When Donald Trump was elected, some people described themselves as “traumatized”. Someone asked me for comment on the record, hoping I would say something like “as a real psychiatrist, trauma is a real disorder with strict criteria, and all you people are dumb”.

I did not, in fact, make this comment.

Suppose that outrage addiction is, in fact, trauma addiction. That means the media ecosystem is a giant machine trying to traumatize as many people as possible in order to create repeat customers, ie trauma addicts. Combine that with the explicit, confessed desire on both sides to “trigger” the other as much as possible, and you have a lot of very clever people all trying to maximize one another’s trauma levels. On the external level, that looks like weaving as strong a narrative of threat and persecution as possible and trying to hit people in their psychological weak points.

I tend to agree with this picture. I would be very interested to hear what you have to say, if you do not agree with it.

If true, it’s obviously a huge problem. It’s as though an entire country of hundreds of millions of people is now employed as its own anti-therapist. Woe is the patient. Or should I say, the impatient. Impatient to get home. Impatient to take out their phone, to switch on the TV or TikTok or YouTube or Facebook. Impatient to suffer, and all the while, impatient for the suffering to end.

Millions of them.

No single bird needs to comprehend what the flock is doing. It merely needs to respond to its neighbours and the other parts of its immediate environment (e.g. don’t collide with the ground). The larger patterns of the flock emerge naturally from all of this local, individual behaviour.

Maybe the picture grows clearer if we scale our attention down, from the level of many individuals to just a single one. I’d be very surprised if any one of your cells knows that it’s participating in your body, or in particular, if any single neuron knows1 that it’s helping to make up your mind. Inversely, the patterns of your mind can surely influence the situation of an individual cell, and that’s becoming even more true as we get better at coordinating our minds and messing with our biology.

When a person has PTSD, or some other mental disorder, what does that imply about what their cells are doing? Do we go looking for some of their neurons that also “have PTSD”? Not really, even though some of those neurons are probably under extra stress sometimes, due to past trauma. Though we might measure that and say it’s “associated with PTSD” or something. On the other hand, if I happened to measure 1,000 neurons and observed that they were all more excitable than usual, should I expect to see that the person they belong to is more excitable in a psychiatrically relevant way? Maybe. In the extreme, that’s just epilepsy. But it seems weird to say that a neuron has epilepsy, even though there is a clearer correspondence in that case.

When a person has PTSD, does it make sense to say that the society has PTSD? I don’t know, and I’m not claiming that’s what Scott was doing exactly. But it’s interesting to consider whether the superorganism itself can “have” an illness which somehow corresponds (or not) to disorders that its constituent organisms might have.

Does it make sense to model a society itself, and not just its individuals, as having a disorder, or what we could call a superdisorder? Are these superdisorders ever analogues of individual psychiatric disorders, such as the ones described in the DSM? Do the disorders of individuals and the superdisorders of groups need to line up?

According to Wikipedia, an egregore:

is a concept in Western esotericism of a non-physical entity or thoughtform that arises from the collective thoughts and emotions of a distinct group of individuals

Now, I’m not going to say that non-physical entities are involved in anything, or that “thoughts are things” in the occult sense. That seems like a map-territory confusion, which I’d amend like so: when I notice or say that I’m having a thought, I’m effectively pointing to the effects of a subset of the physics of my body, though I may not know it. Likewise, an egregore is not some mystical overarching ghost, born separately from its constituents, but a convenient handle on the emergent effects of a group. In particular, egregores may appear to have their own agency that feeds back on individuals, but this is not mystical. The individuals carry it out themselves, by being sensitive to incentives and punishments generally, and sometimes by subscribing to a model or personification of the force that exerts those pressures, i.e. an egregore.

That seems reasonable enough, and it’s basically how rationalists have treated egregores, even when they give them godlike names. The most prominent examples:

• Moloch, evoked by Allen Ginsberg and reified by Scott Alexander in Meditations on Moloch, is the rationalist handle on the behaviour of a group that’s racing to the bottom somehow, where it’s in nobody’s personal interest to be the first to be the better person, and so everything remains shitty.

• Elua, introduced by Scott Alexander at the end of Meditations, is the counter-egregore to Moloch. Elua is “the god of flowers and free love and all soft and fragile things. Of art and science and philosophy and love”. The wholesome uprightness of humanity and civilization.

• Ra, identified by Sarah Constantin, is a force of corruption that causes people to optimize for the appearance of institutional goodness. A vacant prestige, accompanied by a latent hostility towards any request for receipts. Ra is one of the twisted offspring of Goodhart’s law.

Is it appropriate to see an egregore as structurally similar to a particular mental state, or psychiatric disorder?

Scott’s PTSD framing, taken to its extreme, ends with utter partisanship. Two egregores locked in a terminal struggle, clawing at each other forever. Of course it’s actually the individuals that are causing this to happen, though they don’t need to be aware of the big picture. They’re just responding to their local environment, except that “local” now includes an expertly curated smorgasbord of trauma triggers.

Interestingly, there does seem to be scale similarity in the case of “partisan PTSD”: the contractions of the individual at least somewhat mirror the contractions of their society. The PTSD egregore kind of looks like the PTSD individual, fragmented and clenchy.

In the next post in this series, I’ll explore how the correspondence between levels isn’t clean and can break down, or change over history.

Subscribe now

You’ve just imagined what it’s like to be a bird.1 No sense of propriety, decorum, proportion. Don’t like something? Date going poorly, perhaps? Take to the heavens! Find somewhere new! Erratic little shits.

What an apt slur. For oh, how they shit.

The white stuff is uric acid. That’s how they get rid of extra nitrogen: as an opaque solid.2 Apparently, there are some who appreciate these donations. To me, they seem gratuitous at best. The Earth already has lots of nitrogen for the taking. The atmosphere is mostly nitrogen!

Despite a bird’s relative lack of chill, it does enjoy atmosphere. But though its poop’s paleness is nitrogenous, it isn’t the proceeds of the tropospheric air, not proximally. It’s the product of protein, processing inward through the cursed gut of the thing, entering via the clickety little beak on its freaky little face.

You (yes, you) also have extra nitrogen to get rid of. Haven’t you noticed? You discard it as urea though, not as uric acid. Urea likes to be wet, so it isn’t white when you leak it out.

Some skin creams contain urea. And because urea likes to be wet, it pulls water into your skin. Your hands get softer.

I guess it might go without saying, but you shouldn’t urinate on yourself as an alternative to conventional skincare.3

The martlet is a bird without feet. It isn’t real, though. A mythical ornithical. According to Wikipedia, it “never roosts from the moment of its drop-birth until its death fall; martlets are proposed to be constantly on the wing. This condition is an allegory for continuous effort”.

Relatable.

A martlet is a sort of beacon of virtue. I especially respect that it never insinuates its unorthodox body into terrestrial affairs.

Subscribe now

Can you tell that I’m trying to find my authorial voice? Leeleeleeleelee!

Unfortunately, I’ll never tweet as well as a bird. My ancestors grew a larynx, which is basically a flute made of a couple of muscle-tensioned sheets of meat. Revolutionary, at the time. A bird, on the other hand, has a syrinx. This means it has two flutes instead of one, with meat that jiggles even faster than ours can.

I’m not envious; the Earth already has lots of meat for the taking. Now, if only I weren’t terminally confused about whether I should be a vegan...

Birds have better lungs than us. Is there no end to the humiliation?4

Our lungs are made of many little bags that fill and unfill with air. The air in the middle of the bag isn’t absorbed very well on account of being in buttfuck nowhere, alveolarly speaking.

Birds, in an act of utterly-smugly-insufferable evolution, developed the following solution: they pull the air into bags (bigger than ours…) that elegantly absorb nothing. Then they push it out of the bags and through a series of tubes. Blood vessels run across the tubes. Something-something physics, something-something higher ceiling on the amount of oxygen absorbed.

This setup is really helpful to birds because of how much time they spend in flight, away from everything and everyone, at unconscionably hypoxic altitudes.5

For a similar reason, birds also have stronger hearts than us.

Ugh. It’s enough to turn you into a conflict theorist.

Time to smash some sparrows, perhaps?

When the gusts are small and frequent there’s no point in aggressive correction. No single gust is dangerous enough to matter. If they nudge your car a bit, they’ll tend to average themselves out without the need for your intervention, or even your awareness. But on some stretches the gusts trend larger, and a single one might be large enough to push you out of your lane. So you have to act, when it hits. You don’t know what direction the next gust will come from, or if it’ll be the big one, but by gripping and reacting harder, your response will be more robust to any of them, when they do come.

Even if the gusts are small at the moment, you might be unsure if they’re about to get big again soon. So why not grip harder and flinch more, anyway, just in case?

You’re sitting in an airplane that’s just entered a zone of turbulence. Your rational mind knows that turbulence is basically never a real danger, and that the plane’s wings can flex a lot more than that without being damaged.

Maybe you enjoy the ride and let it happen. Maybe your knuckles are white as you grip the armrests too tightly, as if that gives you any control over the effects of the next jolt.

Two Inkhaven residents (A and B) walk into the winner’s lounge at the end of the day, where a third (C) and fourth (unlettered) are already talking in hushed voices. C glances at the two newcomers and half a second later bursts out laughing, then quickly glances away. Both A and B wonder for a moment whether they’re being mocked or something. A quickly lets it go: It’s probably nothing. Residents tend to laugh a lot in the winner’s lounge.

It turns out A is right, but B cannot stop wondering whether C disrespects them. They never ask C for clarification. If C is in fact an adversary, that might be a bad idea. So B just keeps wondering, for all the stress and friction it causes over the rest of the month, whether C dislikes them.

What do these stories have in common? They all show some result of how our minds can come to move differently after they perceive adversaries during learning.

I’m pointing at a single, formal principle here. It is the single most important principle you can learn, that operationally determines mental health. And it’s due to a universal fact about learning,1 so it’s also relevant to the training of AI systems.

In 2023 and 2024, I wrote a series of posts that more or less orbited this principle. Each post was a dive into a different field or concept, such as motor control, self-fulfilling prophecies, canalization, annealing, Buddhism, and meditation. Those posts took a lot of effort, and in hindsight they don’t crisply convey the crucial insight. So, you don’t need to read them unless you want to be filled in on all sorts of surrounding material.

In this new series, I’ll tackle the problem more directly and formally, and with more of a focus on AI safety.2 I don’t know how many parts there’ll be yet. (Probably at least 3, including this one, and none of them nearly as long as some of those old posts.)

As far as I can tell, the pieces of the puzzle aren’t original. You’ll find them or their mathematical equivalents lying under such concepts as canalization, trapped priors (in psychopathology); alignment tax and nearest unblocked strategies (in AI alignment); and H∞ and risk-sensitive controllers (in (motor) control); intelligent paranoia; and Goodhart’s law.

At best, this series will find one or two new framings or syntheses that may be useful. At worst, writing it will teach me to think more clearly about these topics.

I’ll end by reframing the behaviour of Inkhaven residents A and B in terms nearer to our principle.

A has no trouble letting the situation go. If we visualize their mental landscape3, with their chain of thoughts/behaviours described over time by a path followed through the landscape, we’ll find it relatively flat and gentle. When they get knocked towards the valley of “I’m being humiliated”, it’s a shallow one, so they can easily flow around it, or back out of it.

On the other hand, B‘s valley of humiliation is carved steep and deep.4 They flow into it easily, even in response to ambiguous evidence. They only exit it with difficulty.

In the next part of the series, we’ll take our first crack at formalizing what is happening in the landscape, including why adversaries during training tend to make things deeper and steeper.

Subscribe now

Sometimes it’s almost tempting to stare directly into the sun, as if to ask for even more. But I won’t do that. I know what happens.

It can be cold at night. During the talent show I wore three layers. I even drank a small cup of beet juice (for the nitrates) but my frigid limbs wouldn’t have it, and continued to jitter and worm their way inward.

I don’t mean to complain. It was sub-zero and snowing only 3 days before my departure from Montréal.

One evening I struggled to operate one of the gas fire pits. I’m not sure I ever got the gas to even flow. I checked the lines; I tried every position. Nothing. Apparently there’d been a small earthquake on the 2nd which might have tickled it into dysfunction. (I was asleep at the time, and missed out on my share of the fun.)

I didn’t bring it up with the officials. I moved on. It’s not like the flames would’ve been more captivating on that altar, than on any of the others.

I sleep a bit better when it’s cold. Have you noticed the same? It’s a Known Thing, as far as I recall. But memory’s slippery. And I’m not going to ask Claude for confirmation. It lives in the mines and I don’t want to go down there today.

Yesterday I wrote an almost three-thousand-word technical post in about seven hours. That was a surprise. I’m certain I’ve written just as many words in just as short a time before, but this felt cleaner and more consequential somehow. I think my thoughts have been clearing.

I wrote that post in the Glass Room. It’s a bright room; it lets god in through its panes. After having hiked the Stonewall-Panoramic Trail this afternoon, I’m not sure I was as reverent as I could have been, yesterday, my first time writing there.

I went back there today after returning from the hike, but it was too warm. I felt so close to heaven, I almost started to nap. I had to leave, then I drank a small cup of tea out of ‘necessity’. It was only after the tea started working that I realized the problem had been the heat.

Hm. Are there mines in heaven?

I keep referring to myself as autistic in conversations. I should stop doing that.

It’s a hand-wave, something I’ve said when I’ve felt I’ve lacked the time to talk more deeply about myself. But go ahead and ask me. I’ll show you what I am.

“Autism” is a sickness word, isn’t it? Now, before you or anyone else leaps up to smooth that over with platitudes, I want you to see how mostly okay I am. (And what exactly would smoothness serve, if I weren’t already okay? A transient okayness, couched in a larger suffering? At best it would give me some slack.)

I don’t need to see myself as autistic, and I don’t need to see myself as not autistic. It just isn’t that helpful. I guess I’m something more specific.

I’ve started some focaccia dough. It’s rising in the kitchen. I’ve stretched it twice already. I might move it to the glass room in a moment. There’s a poetry reading in two hours and I want it to be ready by then.

Yeast is slow to make its way to the Glass Room, at least on its own, so I help it. It has no trouble with reverence, once it’s there.

I haven’t written about my intentions for Inkhaven yet.

I do want to produce good and even impressive works. I want to say that this might not be one of them (it hasn’t taken much effort) but I’ve been led to believe that my prejudices about which of my words others will most enjoy, are probably wrong.

(Well, are you tickled yet?)

More important than being impressive, though, I want to be healthy. I want my mind to be kind of smooth (uh oh) like a... like a dolphin in its medium? I want to speak from the edge of unspeakable wonder. I want my words to be something like the trees I passed in the hills today, growing and growing and smelling like fresh basil for some reason. I want to breach the surface of thought, just to fall back in again. I want to be filled with light and warmth.

The clouds have arrived as though to prove something, but the Glass Room will probably stay warm enough for long enough to make a difference in the rise.

I’m grateful I’ve had many such experiences already, one week into Inkhaven. But something seems not quite as good as it could be: how those moments get turned into writing. When I sit down later, usually I can recapture my ideas with some effort. But it can feel like wasted effort, when I think of the fluency and ease with which I covered all the points within the flow of the original conversation.

Some experienced writers may see it as a total non-issue. I get it: the more conversations I have, the more socially integrated I become, and the more I follow the advice to “write first, edit later”, the easier it is to write as though I am still in conversation, recapturing something like my original energy and fluency. I can already feel a change for the better, after a week of Inkhaven. But there’s still friction, and I’m not sure that the verbal-first orientation I currently have will ever entirely make way for a “writing is entirely satisfyingly akin to speaking” situation.

A broader skill may have always depended on certain other skills, but that does not imply that those other skills (as currently conceived) should be seen as universal prerequisites to the broader skill. Consider the skill of getting from point A to point B. At one time we used horses for this, but it is no longer necessary to brush and speak softly to most forms of transportation. Nor is it necessary any longer to crank a car to start it, nor to keep maps in the glove compartment, nor even to micromanage the process of navigation by sitting at the steering wheel and manually driving the car.1

The same is true for writing. Of course I can take notes during a conversation. I can even proactively ask to record the entire thing, or sections of it when I suspect I am about to riff productively. But both of these actions disrupt the vibe, and may make participants self-conscious in a way that interferes with expression. And the parts of the conversation that are most worth capturing are often not evident except with a small amount of hindsight, which implies an additional effort to go back and sift through recordings in the case that we attempt to solve the problem by recording everything.

Now, consider the following vision: I’m having a conversation with a friend. We speak at length and easily. Eventually we hit on a subject that’s dear to me. I riff about it for a minute (or five) and then think “wow, that was a good framing/phrasing, I wish I could keep what I just said as a draft for a post, but also I want to keep exploring this and not abandon the conversation for more than 10 seconds to stare at my notebook/laptop”. No problem. I pick up my device, and tap one or two UI elements. My friend is then presented with a dialog and does something similar. Immediately, the transcript of just that part of the conversation appears on my device.

Importantly, in this vision, the disruption to the conversation is minimal and comes at a moment of closure and not one of suspense or anticipation. And when I sit down to write my post, I have to do no additional work to get back into my words.2

How do we achieve something like this? By designing a system that is always recording, but that verifiably discards everything that the participants do not explicitly consent to keep. I’ll call this a secure conversational replay buffer, or SCRB3.

In the rest of this post, I’ll briefly outline the technical, legal, and vibe challenges of such a system. If my initial framing above sets off any privacy red flags for you then you’re in good company, and should keep reading. But note first that the version of this system that we could legally prototype today (e.g. as just a phone app) would not result in as seamless an experience, as a version based on special hardware with baked-in cryptographic verification of what information was discarded. Unless stated otherwise, I will assume that we are talking about an actual cryptographically robust SCRB; how this might actually be achieved will be outlined near the end of the post.

Vibe Concerns

The idea for this proposal first came to me this morning in conversation with some of the other Inkhaven residents. They responded a little uneasily. This is understandable, given that the proposal involves an intentional change to consent dynamics, which should make us at least a little wary. I also had not thought through the cryptographic elements yet, which are central to the privacy concerns expressed by the residents.

There are two psychological concerns I want to point out here. The first is un-self-consciousness, which is one thing this proposal intends to address. You should not need to put effort into pre-emptively worrying about asking for permission at the right time, potentially jarring the vibe, prior to the moment that you feel good about what you have said. If technology can make this possible in a secure way, it seems like a clear win for vibes.

The second: even if users claim to be comfortable with the cryptographic protections, they might still be uncomfortable with the always-on recording, at least at first. This might seem internally inconsistent, assuming verification of discards is genuinely baked in. However, formal guarantees are one thing, and user acclimation is another. It’s understandable that when recording is somehow happening all the time, and the user is not an expert on the cryptographic methods, that they might be anxious or hesitant until the technology has been socially as well as technically proven.

This feels adjacent to a notion of plausible deniability. You tacitly assume that nobody is making clandestine (perhaps illegal, depending on jurisdiction) recordings of your conversations, even though you probably aren’t taking strict measures to ensure this is not the case. If designed properly, SCRBs would be no less secure in this sense than the conversations you are already having.

Of course, it remains to be seen how well SCRBs would be socially proven, and whether observer effects and self-consciousness would persist regardless of technical guarantees.

Legality

Let’s take California as an example, because I’m currently at Inkhaven in Berkeley.

California is an all-party consent state; see California Penal Code § 632. That means it is hostile territory for any technology that records any “confidental communication” without prior consent from all of the parties of the communication, which is a crime. Here’s the definition of “confidential communication” from § 632 (emphasis mine):

“confidential communication” means any communication carried on in circumstances as may reasonably indicate that any party to the communication desires it to be confined to the parties thereto, but excludes a communication made in a public gathering or in any legislative, judicial, executive, or administrative proceeding open to the public, or in any other circumstance in which the parties to the communication may reasonably expect that the communication may be overheard or recorded

If you refer to the rest of the section, you’ll find that it’s the act of recording that’s central to the crime. It doesn’t matter if we immediately delete the recording, or do voice separation to only keep one of the parties, or if the recording happens over the phone or the internet. Consent is required to record at all.

Now, I am not a legal expert, but here’s my reading of how this will apply to a genuinely cryptographically robust SCRB: we are recording, even if it is an encrypted recording with verifiable discards. In California, as soon as we use a system without obtaining prior consent, we’re committing a crime. But this is a novel situation, because by any reasonable measure, zero harm is being done, given the SCRB’s discard guarantee. So we would appeal the criminal charge to a higher court, and attempt to establish a precedent about the use of such a harmless class of recording system.

One additional concern is whether a device might accidentally capture the confidental communication of a passerby. My understanding is shaky here but as far as I can tell, this would not be any more of a big deal than it already is for consent-to-record, because I am neither party to, nor intentionally intercepting the passerby’s communication.

It’s worth pointing out at this point that not all jurisdictions are like California. For example, in Quebec, Canada, recording is legal only if you are a participant or have consent from one party; see the Quebec Criminal Code § 184(2)(a). However, it is appropriate to design our system to do zero harm according to the strictest jurisdiction in which its users (or parties to their conversations) will reside. This legal constraint aligns with the high-trust design we intend for SCRBs anyway.

In the context of legality and in particular civil liability, it’s worth mentioning Vitalik Buterin’s concept of control as liability:

every bit of control you have is a liability: you might be regulated because of it. If you exhibit control over your users’ cryptocurrency, you are a money transmitter. If you have “sole discretion over fares, and can charge drivers a cancellation fee if they choose not to take a ride, prohibit drivers from picking up passengers not using the app and suspend or deactivate drivers’ accounts”, you are an employer. If you control your users’ data, you’re required to make sure you can argue just cause, have a compliance officer, and give your users access to download or delete the data.

If you are an application builder, and you are both lazy and fear legal trouble, there is one easy way to make sure that you violate none of the above new rules: don’t build applications that centralize control.

A robust SCRB would already record only the minimal amount of data that the users had consented to, so that’s good for exposure to unforeseen liability. And we might take this insight further and not provide a cloud service at all, to accompany a hypothetical SCRB helper app or whatever.

Subscribe now

Mechanism design

The State of the Art

Of course, there’s good old audio recordings.4 I already mostly mentioned the limitations of these in the introduction. You could improve the granularity of recording, e.g. by getting prior consent once, and then only holding down a record button during moments you want to record, which is minimally disruptive. Still, you have to remember to get consent before you press the button. And the moment at which you decide to record is prior to actually stating your ideas, not after, so filtering of material is not built-in as it is with SCRBs.

Lifelogging is the practice of making personal records of one’s life, which can involve such technology as wearable microphones or cameras. Bloggers have been writing about it for well over a decade now.5 While the degree of coverage (e.g. text versus audio versus video) varies between users, generally the interest is not in a highly selective, salience-based, secure recording system, but instead a total record of one or more aspects of life. Of course, this does not escape the legal issues with good old audio recordings.

Contemporary wearables like the Limitless Pendant (Meta) or Bee AI (Amazon) are similar, in that they use an always-on design that relies on post-hoc filtering, and do not use a selective cryptographic system to obviate prior consent.

The closest technology to SCRBs in terms of UX is a replay buffer, such as used in live streaming or music production. Basically, data is stored to a circular buffer, which allows the user the post-hoc ability to capture data that had been recorded in the past N seconds/minutes, with anything older being continuously discarded. Existing systems are intended for individual use, however, and aren’t concerned with multi-party consent.

How to do better

This section will be necessarily very brief in this preliminary proposal, given time limitations here at Inkhaven and also my limited familiarity with the cryptographic systems literature.

We tentatively adopt Vitalik Buterin’s notion of privacy pools. In Buterin’s case, this means using zero-knowledge proofs (ZKPs) to demonstrate that your funds don’t originate from unlawful sources, without revealing your entire transaction history to the entity that desires such verification from you. In principle, a robust SCRB would be designed to produce similar cryptographic attestations that it had in fact overwritten (part of) its buffer at a given time T.

There are a number of relatively more incidental decisions that would need to be made about hardware and software design. Presumably the user interface for retention/consent decisions would either be baked in to the hardware itself (e.g. buttons and a separate screen on a small recording device) or else we’d use some kind of helper app running on a paired device, such as a smart phone. I won’t discuss this more now, as it seems quite open-ended and I imagine the specifics would be decided by better designers than me.

I have the seeds of some other ideas in my mind, but these are entirely speculative at this point. For example, we we could use sandboxed language models, whose operation may also be subject to cryptographic guarantees of some kind, to decide on “smart boundaries” instead of relying on fixed discard horizons. In principle this could be used to provide the user with richer choices about what to retain, than “last 1 min” or “last 5 mins”, which would rely on the user having memory of how long they had been talking for. However, this means transferring some information about the content of the conversation to a visual display or helper app prior to the user’s consent, and this seems like it introduces the possibility of harm.

Conclusion

This is just a preliminary outline of a proposal, written in a few hours at Inkhaven, by someone who is not a legal or cryptographic professional.

It’s possible that there would be deep issues with getting the technical side to work, and I do not understand the cryptographic mechanism well enough yet to say with more confidence whether it will work. In any case it would be an engineering challenge that some startup or other organization would need to handle, while accepting the likelihood of a criminal legal challenge to establish the precedent of these systems as zero-harm. That such a challenge is likely in at least one jurisdiction suggests that great care should be taken to ensure the systems actually are zero-harm and function as advertised prior to this challenge, to avoid potentially quite protracted legal complications.

Subvocal recording devices, or something like them, could totally obviate the need for prior consent / SCRBs in common circumstances. Some such devices are in active development, and are interesting in that they might genuinely capture only my voice and nobody else’s. This might satisfy California’s statute by design, assuming we have a clear answer to the question of whether the device records any signal from parties other than the wearer. And they also do not necessarily time-lock the retention decision to the moment of emotional closure, with filtering coming at no extra cost, nor do they necessarily encrypt anything or provide any sort of other guarantees. So even if we adopt these devices, we may still want to adopt some SCRB-like framework for better security and hygiene.

I remain optimistic that I might live to see SCRB-like systems make our writing lives more about ideas and words, and less about distracting chores.

Update: Shortly after posting this, I noticed a potentially serious technical problem: how do you identify the parties that must provide consent at the time of retention? Therefore, I suspect that all parties must provide some kind of prior registration/consent. Still, I think there would be advantages to an encryption + consensual partial retention + verified discard scheme, with an added “joining contract” that defines who is party to the confidence.

But in the end, it did happen to provide for intelligence which, ever-imaginative, tempts itself to impose structures from above. The most famous political example is the French Revolution. The revolutionaries saw themselves as inventors. Some of them had infected each other with the notion that they, as individuals, should and even must grasp the big picture of their society. But when a naïve and earnest progressive finds, inevitably, that the world does not bend to their conceptions, little choice besides force may seem to remain. After all, these conceptions are the truer thing! Have you seen the alternative we’ve been living with?

You probably know how the revolution turned out.

Some individuals don’t flow so well with others. While their neighbours are shaking hands, they miss their cue, breaking the contractual rhythm. Well, it happens to everyone, but it happens to these people enough that they are rather more ejected from the party than usual. Divergent.

Maybe this arrhythmia happens at first due to the peculiarities of personal history, to traumas or other complications. Maybe it arises because of deep personal differences in architecture. In any case, cleaved from the main stream, such a one has a tendency to supply more of their own vibes, at more of their own pace.

Cast adrift, filled with less of the flock-stuff, their reward is more space to think and act oddly. At least, so long as the flock does not set its fear on them.

Obviously, not all freethinkers are autistic. Intelligence is complicated, and the flock holds many patterns. “Being a researcher with original ideas” is one, and there isn’t a bouncer that verifies your neurodivergence credentials prior to entry. All you need are certain skills, which contrary to how the cope of the divergent might whine, are mostly not unique to the divergent.

What is a rationalist? In one sense, someone with a comparative advantage in crossing into the wilds of unorthodox shapes and unthinkable diagnoses. At times, that’s what a society most needs. So it might come as a surprise to some innocent rationalists that there are societies that adapt and thrive for decades, but are not run by rationalists.

It’s interesting that the rationalist community exists in the first place. Such a thing had not quite existed before, not on such a scale. It’s as though, after the right scaffolding sprung up, the intelligent divergents were finally able to locate each other through the flocky fog. Building enough common ground, they created a flock, a school all their own. One that was not quite so conditional on tenure or wealth.

But it’s a sort of simulacrum of a normal human society. A convergence of divergents, not quite dripping with the robustness of normativity. Rationalists like to talk about common ground, but their words may be as wild as they are. The meta-concepts that bind them are as the strongest steel... but bound lightly. As much as they have prodigious discursive flexibility, they aren’t as resolute in evolution’s ancient solution to the coordination problem.

It’s for this reason that populism can be so successful when it’s also so mentally bankrupt. We didn’t need to care about relinquishment and levity! We could have optimized for mindless robustening, organizing not around truth-seeking but around volumes of largely arbitrary statements serving as mere veils over a raw consensus. Then contracting as a colossal muscle, we would crush our adversaries!

And fuck them, right? How dare they be so thoughtful.

Psychedelic research has been dominated by the 5-HT2A receptor hypothesis since the connection was tentatively made by Glennon (1984). Afterwards there emerged a stronger logic: if study participants are first given a drug (ketanserin) that blocks the 5-HT2A receptor, they won’t experience hallucinogenic effects if they then take psilocybin and LSD. So 5-HT2A agonism must be “how psychedelics work”, right?

Okay, first, what’s “5-HT2A”? The 5-HT is just short for serotonin (a.k.a. 5-hydroxytryptamine), and 2A just identifies one type of receptor that serotonin binds to. There are at least 14 types of serotonin receptor, and they’ve been associated with all sorts of different physiological and psychiatric effects.

Each receptor subtype may be expressed more in one part of the body than another, to serve different local roles. They are not limited to the brain. There’s a meme that 95% of the body’s serotonin is in the enteric (gut) nervous system, but it’s also important that most of the serotonin receptors in the gut are of certain types (mainly 2B, 3, and 4) with gut-specific roles.

Serotonin itself has affinity for all of its receptors, and activates them with high efficacy when it’s around to do so. But other substances that bind to serotonin receptors may have affinity for some types but not others. When binding to a given type of receptor, a drug may have full efficacy at activating it, but might also have 1) only partial efficacy, 2) no efficacy, but still block the efficacy of serotonin or other drugs that would otherwise bind,1 or 3) inverse efficacy, causing an effectively opposite effect to what an agonist like serotonin would produce, downstream. Here are a couple of examples:

• The anti-migraine drug sumatriptan binds to the 1B and 1D types more than any of the others (it’s selective for them) and so it specifically causes blood vessels in the brain’s meninges to contract. It typically does not cause hallucinations.

• The antidepressant trazodone is both a partial agonist of the 1A type, but also an antagonist of 2A, 2B, and 1D.

A 2025 study by Jain et al. did a deep dive to quantify the efficacy of 41 different psychedelics at whichever of 318 different receptors they were found to have affinity for. They found that 5-HT2A was by no means the sole target. In particular, some psychedelics have efficacy at a broad range of receptors. One class in particular where efficacy was broadly found, but which has traditionally been neglected, is the α₂-adrenergic receptors. Adrenergic receptor activation is known to alter the response to activation of serotonin receptors in animal models, so we might expect that two psychedelics that share the same 5-HT2A efficacy, but differ in efficacy at one or more α₂-adrenergic receptors, might differ in their subjective effects.

A couple of science journalism posts were written in response to Jain et al. 2025, in the usual tone of look at this new discovery! but without adequately qualifying the claim against the history of the field.

“Psychedelics aren’t just serotonin/5-HT2A agonists” is not a 2025 discovery, just one that had largely been neglected before. The original 5-HT2A mechanism paper (Glennon, 1984) didn’t claim that 5-HT2A or even serotonin receptors in general were solely responsible for psychedelic effects, just that there were high correlations with drug discrimination measures in rats, and rough human dose estimates2. They concluded that the mechanism “involves [5-HT2A]-related events”. The activity of LSD at dopamine receptors has been known since at least 1975. And most significantly, Ray 2010 explicitly quantified how strongly different psychedelics bind to a bunch of different not-serotonin receptors.3

The fixation on the 5-HT2A receptor came later. It’s worth mentioning that David E. Nichols, one of the senior authors on the 5-HT2A hypothesis papers and a leading figure in psychedelics research, is also one of the senior authors on Jain et al. 2025. So this isn’t exactly a rift in academia, or something.

Of course there is still variation between different psychedelics in terms of the specific receptors they bind to, and the overall breadth of their receptor binding (or promiscuity). For example, lysergamides like LSD tend to be more promiscuous than tryptamines like psilocin. And this fits with subjective reports that different drugs in the same class can have substantially different subjective qualities, even when they appear to have similar 5-HT2A efficacy.

The ketanserin studies still have weight, and 5HT2A efficacy seems to be necessary for the most subjectively blatant effects of psychedelics, in any case.

But suppose that by giving ketanserin first, you can eliminate self-reports of hallucinations and “yes I’m on a drug right now”. Where is the line between a blatant hallucination, and other effects which might occur and even be psychiatrically relevant, but not rise to conscious/linguistic salience during the experience?

When measures collapse, does this mean there are no effects? Or are studies statistically hobbled by the limited perceptual/introspective clarity of study participants? Clearly, we have more work to do.

Anyway, I hope you’ll leave here with just a little itch in the back of your mind any time you hear (or get ready to say) something like “psychedelics work by pressing the serotonin button in your brain”.

It isn’t just any egg. It’s sitting in the caldera of a volcano. Maybe it represents the precarity of life?

It took me several hours to coax StableDiffusion 1.5 (SD1.5) to lay this particular egg in January 2023. It totally failed at simple attempts like “egg sitting in the caldera of a volcano”, so instead I started by using the AND keyword to produce the intersection of two prompts:1

1. an egg in a cup

2. an egg in a caldera

But a simple AND of these prompts wasn’t enough to get the result I wanted. Instead, I’d invariably be given some magmatic variant of your boring old egg-in-a-cup.

I suppose this was because the joint egg+cup was much stronger than in SD1.5’s training set than egg+caldera.

To sneak past that, I weighted the egg-in-a-cup prompt more highly for the first 2-4 denoising steps, then quickly turned up the weight on egg-in-a-caldera. The idea was to get the model to put the egg into a noisy proto-cup, thereby locking in the correct overall cup-like composition, but then while the cup’s form was still ambiguous, turn up egg-in-a-caldera.

Here are are two results from during that tuning process. The one on the left is fully egg-in-cup. The one on the right is somewhere in between, the weights being switched too late, and the cup too well established already to be diffused into a volcano.

Here are a few other eggs-in-caldera that I generated with SD1.5 from different seeds.

See how weird they are?

Eggs and egg-shaped portals have become my preferred objects to prompt for, when evaluating the aesthetics of image models. In the process I think I’ve become something of an eggspert on what makes a good egg, as well as what to eggspect from each model.

In the neggst section I’ll show results from some other models. None of them allow advanced parametrization like AND, or weight adjustments over denoising steps, or most of the other fancy things we can do when working with local models like StableDiffusion. But why not?

For one, the frontier labs haven’t been much for giving that level of parametric control over their cloud models in any case. And as models have gotten better at thinking and composition, the eggspectation is that you should be able to simply describe what you want prosaically, and the model will figure everything out for you.

While compositionality and prompt understanding have both improved, I agree with others that the results have lately become less iconic or weird, and more like quality, all-purpose trough fodder. But the purpose of this post not to talk about model collapse or regression to the mean or whatever you think is the cause of ensloppification. This post is eggs.

If this were a rigorous eggsperiment, I would have done some things differently, such as giving a standard set of prompts to each model in turn, for a fairer set of comparisons. But this was not rigorous. This was eggs.

Now some Midjourney v6 from February 2024.

Ostentatious:

Eggsample prompt (bottom left): cosmic egg, pristine yolk, divine whites, smooth shell, sunburst, bicolor, woodcut bliss --no frame

Ethereal:

Eggsample prompt (right): psychedelic portal, smooth and pristine egg in its natural habitat --no cracked, border, rough

Stark? Lines?:

Eggsample prompt (top right): doodle linotype, psychedelic portal, smooth and pristine egg in its natural habitat --no cracked, border, rough

Embirded:

Eggsample prompt (top left): masterful precise illustration, unnaturally smooth egg, mystical experience involving a bird’s unhatched progeny --no smudge --weird 100

DI’:

Prompt: psychedelic color intaglio, salvador dali, smooth and pristine egg in its natural habitat --no cracked, border, rough shell, reflective

Midjourney v6 generated decently interesting images with little prompting effort, so my strategy was a bit different than with SD1.5: generate a bunch of images and choose my favourites. This felt less like “making art” than caldera-egg did.

I was pleased with the variability in its responses, but likewise, it had some issues with the consistency of details. For eggsample:

Prompt: masterful precise illustration, unnaturally smooth psychedelic egg, mystical experience involving a bird’s unhatched progeny --no sane

Kind of a mess, though reasonably symmetric compared to what SD1.5 would have done with something this intricate.

Now DALL-E 3. March 2025.

Eggsample prompt (top left): minimalist egg containing a psychedelic portal, resting snugly in the caldera of a single volcano as if it were an egg cup, in a psychedelic ukiyo-e landscape

Still low prompt effort, and the results were getting smoother and still more coherent in the details. I remember being frustrated that I was limited to prosaic prompts, but couldn’t coax it to rest the egg inside the caldera.

GPT-Image-1 on the first of August, 2025.

Prompt: in the Orion Nebula, a brilliant egg shaped portal, through which is visible a barren stratovolcano under a clear blue sky, with a white egg sitting half-inside its caldera, as though in a cup

This is the first time I was able to get a model to one-shot this composition, though I think the result is still less iconic than the SD1.5 caldera egg.

And of course it still makes mistakes, such as its attempt at the netivot in this kabbalistic tree of egg:

Prompt: in the Orion Nebula, a brilliant egg shaped portal, through which is visible a kabbalistic tree of life against a clear blue sky, where each of the 10 sefirot (11 nodes) is a pristine egg

But the best GPT-Image-1 egg is probably this one:

Prompt: A photo of a luminescent egg-shaped portal, through the sheen of which is imminently visible a vibrant fractal dimension; the portal is floating low above a dark, calm, and florid ukiyo-e landscape

Finally, here’s Nano Banana 2 in April 2026:

Prompt: a print of an egg resting in the caldera of a stratovolcano, as though it were an egg cup

Absolute slop. Big oeuf.

I still prefer the original caldera egg. I had to be thoughtful to create it, and it’s iconically simple. So I decided to make it my avatar recently.

I haven’t tried the newest local models (such as Flux) to see how they compare to SD1.5. If I get a chance, I’ll mess with one for a couple of hours before I post some BONUS EGGS that did not make it into this post. Coming to a day near soon!

Fin.

Egg.

(GPT-Image-1) Prompt: generate an image of an egg person in a serious egg play and they are become kind of eggy because they have forgotten one of their eggs (the image does not contain multiple actual eggs)

Subscribe now

Also, the citation links in this post are temporarily broken, but you can still find the respective numbered references at the bottom.

Eating fruit is good, right?

Let’s explore how some substances1 commonly found in plants might make someone think twice about eating another cup of berries, and how they are probably wrong to worry about it.

Then, I’ll propose a novel mechanism which if true, might strike fear into your breakfast.

In 1973, the pediatric allergist Ben Feingold proposed to the American Medical Association that dietary salicylates2 worsen ADHD symptoms in children [1]. Therefore, he promoted a salicylate elimination diet. Many plant foods contain salicylate [2,3,4]: around 1-5 mg per 100 g for fresh fruits, up to ~35 mg per 100 g for dried fruits, and a whopping ~1 g per 100 g for spices like cumin, turmeric, rosemary, and ginger. Feingold’s fruit-fearing diet would avoid them all.

And your average psychiatrist might have recommended it even today, if it weren’t for a 1983 meta-analysis [5] of 23 studies of the Feingold diet that found effect sizes of 0.015 and 0.052 on scales measuring attention and disruptive behaviour. Statistically speaking this is pathetic, and afterwards, most researchers abandoned the salicylate-ADHD connection.

There’s one valid and lasting take on salicylate intolerance. This one’s inflammatory, rather than neuropsychiatric: some people with asthma or other inflammatory conditions see their symptoms worsen when they take aspirin or other non-selective NSAIDs, like ibuprofen or naproxen. In the case of asthma, this is called aspirin-exacerbated respiratory disease (AERD) [6].

The mechanism? NSAIDs inhibit the enzyme cyclooxygenase (COX).3 This reduces the conversion of arachidonic acid to prostaglandins, which have both inflammatory and anti-inflammatory effects depending on context; the inflammatory cases are the ones helped by NSAIDs. But then more arachidonic acid is available for conversion by another enzyme, arachidonate 5-lipoxygenase (5-LOX). 5-LOX is the rate-limiting step in the production of leukotrienes, which also serve as inflammatory signals; there is an extra synergy in that PGE2, one of the prostaglandins synthesized by COX-1 and blocked by NSAIDs, serves an anti-inflammatory role by putting the brakes on 5-LOX. Usually all of these concerns are clinically insignificant next to the anti-inflammatory effect of reducing prostaglandins. But for those with AERD, it’s a problem:4

1. Eosinophils and mast cells, which are specific types of immune cells, participate in a positive feedback loop where they are recruited and activated by leukotrienes, and also produce more of them.

2. The cysteinyl leukotrienes LTD4 and LTC4, downstream of 5-LOX, act on CysLT1 receptors causing airway constriction, tissue edema, and mucus production.

3. Through the positive feedback loop of 1, mast cells tend to activate explosively, roping in their neighbours and suddenly dumping a bunch of histamine. This can worsen lung inflammation and also cause additional allergy-like symptoms such as hives and itching. However, antihistamines alone are not sufficient to treat AERD, suggesting this is a secondary mechanism.

Why this becomes problematic in patients with AERD is not well understood, but looks like it follows from an infection-induced epigenetic shift in the baseline balance of production of leukotrienes versus prostaglandins, and the reactivity of immune cells.

Should dietary salicylates also be a problem for AERD? They don’t inhibit COX-1 enough to reduce the production of PGE-2 significantly, taking the brakes off 5-LOX, nor do they inhibit COX enough overall to much increase free arachidonic acid.5 So mechanistically, we might expect the answer to be no. The largest RCT [26] that suggests otherwise is too weak to convince.

So: the asthmatics probably need not fear the berries. Well, not on account of their asthma, anyway.6

I’m not going to review the entire cursed realm of internet users claiming to be sensitive to dietary salicylates, polyphenols, and whatever other Trojan berry adversaries that might be captivating their paranoia. But here’s a mugshot. In it we see people:

• Attributing all kinds of symptoms to dietary salicylates including dark undereye circles and adrenal fatigue.

• Fixating on a handful of studies from the 1990s (mostly in autistic children) suggesting that phenol sulfotransferase deficiency is responsible for the accumulation of dietary phenols in the body.

• Failing to rigorously distinguish between polyphenols, salicylates, and phenols in general, let alone different polyphenols, instead lumping everything into “high-phenol” foods. Likewise the recommended treatments for salicylate sensitivity, phenol sensitivity, and methylation disorders more or less overlap.

• Self-diagnosing with conditions in the absence of established diagnostic tests; while in principle elimination/challenge dieting can reveal things, we should expect it to be vulnerable to placebo and confirmation bias.

• Following modern variants or extensions of the Feingold diet. The most structured is FAILSAFE (Free of Additives, Low in Salicylates, Amines, and Flavour Enhancers), which has a large Facebook following but no modern evidence to back up the salicylate claim.

Of course, in the proud tradition of giving the deranged more material to work with, I’m going to do a bit of my own speculation now, and suggest what I believe is a novel mechanism by which the consumption of salicylate- or polyphenol-containing foods might make someone anxious, stressed, agitated, aggressive, or otherwise affect their mood, and in fact how salicylic acid and polyphenols might be synergistic in this respect.

What I am about to propose is probably fake. Like, I’ve eaten a cup of blueberries before and become anxious afterwards. Okay, maybe there’s a connection there. But it does not need to be the one that I’m proposing. Probably the blueberries were just an easy scapegoat. And me, just someone who used to have moderate baseline anxiety yearning to be accounted for.

Now consider:

1. Salicylates might increase brain dopamine synthesis. A single unreplicated study [7] in 2018 found that when mice were given 2 mg/kg/day of aspirin for 30 days, it increased the expression of tyrosine hydroxylase (TH) by 2-3x in the substantia nigra, one of the two major midbrain nuclei where dopamine is synthesized. The upstream mechanism involves the phosphorylation of CREB, one of the classic ways gene expression is regulated in animals.

2. Some polyphenols inhibit the sole dopamine-degrading enzyme found in prefrontal cortex (PFC). Multiple studies [8,9] have shown that some common fruit polyphenols inhibit the enzyme catechol-O-methyltransferase (COMT), which metabolizes dopamine and thus controls dopamine levels. In most areas COMT isn’t the only enzyme that does this. But in PFC, it is! [12,13] This means that, if any of those polyphenols can reach PFC in sufficient concentration, it should increase the concentration of dopamine there by reducing its degradation by COMT.

3. Increased dopamine in prefrontal cortex pushes us away from inattentiveness, and towards rumination. The mainline theory of the relationship of PFC dopamine levels to psychology is the inverted-U model [19]: agonism of dopamine D1 receptors on pyramidal neurons enhances NMDA (excitatory) channel currents and suppresses other inputs to the neuron. This strengthens whichever representations are currently active, while reducing updating based on new inputs. This framing says that when PFC dopamine levels are too low, you get ADHD-like poor working memory because representations are wispy and transient, and this is the major hypothesis about how psychostimulants like amphetamine improve working memory in ADHD. On the other hand, too much dopamine and you tend to become ruminative and unable to stop thinking about whatever you’re currently thinking about. Downstream of that: anxiety, agitation, and whatever else.

4. Dopamine content/degradation in PFC varies between individuals. There are significant differences in COMT activity between individuals, due in part to the Val158Met polymorphism. Individuals with the Met/Met variant (about 25% of Europeans [10]) naturally have 3-4x lower enzymatic activity [10,11] and we might expect them to be more ruminative even prior to the addition of inhibitors.

So: if your fruit contains a COMT-inhibiting polyphenol which can make it into PFC, it could raise the dopamine levels there and shift you towards ruminative on the inattentive-ruminative axis. And if your fruit contains salicylates which upregulate TH similarly to 10 mg of aspirin (which is a very low dose) then this may synergize by increasing the dopamine supplied to PFC. And the magnitude of the effect should depend on your baseline expression of COMT, which depends on your genes.

There are a few potential problems here:

• The 2018 TH study [7] is the product of a single lab and hasn’t been replicated yet.

• Does it take 30 days for the upregulation to happen? No, it probably starts within hours. The 2018 study includes some in vitro experiments that weakly support this. And the 30 day in vivo exposure period is unsurprising; note that the mice need to be killed to do the final analysis, and in the case of a chronic nutritional intervention, steady state effects are important.

• Dopamine projections to PFC actually come from the ventral tegmental area (VTA), and not the substantia nigra. This is probably irrelevant as these two regions are right next to each other and identical in how CREB phosphorylation affects TH expression.

• Do rodent studies translate to humans? Or even, do results from one transgenic mouse model translate to all mice? These are nagging questions in animal research. At least, the 2018 TH study tests two different mouse types: C57/BL6 mice, which are “normal”, and aged A53T mice, which are a model of Parkinson’s disease. Additionally, the CREB phosphorylation mechanism by which TH is upregulated is probably conserved across mammals, and we know that both aspirin and dietary salicylates cross into the brain, in humans [25]. But this is not definitive.

• The study shows that aspirin upregulates TH. Does this apply to salicylic acid, and other dietary salicylates? We know that salicylic acid and aspirin (acetylsalicylic acid) largely share mechanisms, with a significant exception being aspirin’s direct, irreversible COX inhibition which requires the extra acetyl group. More importantly, aspirin is almost entirely metabolized to salicylic acid within an hour or two [25], with salicylic acid persisting somewhat longer. The study does need to be replicated with salicylic acid, but I bet you’ll find a similar effect.

• The dose makes the poison. Assuming dietary salicylates do upregulate TH, how much do I need to consume to do so? The 2018 TH study shows that the TH upregulation happens in a narrow dosage range. So 1 mg or 100 mg of aspirin might not have the same effect, which we might only observe for the dietary salicylate equivalent of around 10 mg of aspirin. The relevant studies are weak (n=20-40 per group) and produced by the same Scottish team [16,17]; one study measured urinary salicyluric acid, and the other, plasma salicylic acid. Their results suggest that vegetarians are similar to people taking 75 mg/day of aspirin, whereas non-vegetarians are perhaps a third of that. Everyone was Scottish and the vegetarians came from a monastery, and we might not expect quite the same results for an average American. But overall these studies roughly accord with expectations, and the equivalent of 10 mg/day probably accounts for a non-negligible subset of most populations.

On the off-chance that this “salicylate and/or polyphenols cause too much PFC dopamine and thus rumination” hypothesis is actually relevant, one of its strengths is that it would validate that some people do experience adverse effects from fruit, while also potentially explaining (say, because they have low COMT activity to begin with) why they are so fixated on whatever explanations they can scrounge up to clothe their miseries. Of course, other mechanisms may still contribute in parallel.7

This post is a speculative review of potential mechanisms. There’s plenty of uncertainty remaining about which mechanisms are practically relevant, about the validity of unreplicated results, and about parameters like dose and timing. And now I’ll finish with a couple of general issues we haven’t covered yet.

First, there are significant differences in salicylate content for each specific food source. We should assume that a blueberry will contain different levels of particular salicylates and polyphenols depending on growing conditions, cultivars, regions, and estimation methods.

Second, leaving aside concerns about mechanism and replication, probably the greatest source of variability remaining is the absorption and distribution of dietary polyphenols. Many polyphenols have low bioavailability [23]8, especially because they are actively yeeted back out of the intestinal wall by the bouncer (P-glycoprotein) right after absorption.9 Likewise, many polyphenols act as prebiotics for gut microbes, and it might be the metabolites that are actually relevant here, and which would need to be assayed in vivo if we wanted to be thorough about it.

Well, there it is. Are you afraid yet? You shouldn’t be. Fruit is probably mostly fine for you.

References

[1] Feingold BF. Why Your Child Is Hyperactive. Random House, 1975.

[2] Swain AR, Dutton SP, Truswell AS. “Salicylates in Foods.” J Am Diet Assoc 85(8):950-960, 1985. 10.1016/S0002-8223(21)03743-3

[3] Malakar S, Gibson PR, Barrett JS, Muir JG. “Naturally Occurring Dietary Salicylates: A Closer Look at Common Australian Foods.” J Food Comp Anal 57:31-39, 2017. 10.1016/j.jfca.2016.12.008

[4] Paterson JR, Srivastava R, Baxter GJ, Graham AB, Lawrence JR. “Salicylic Acid Content of Spices and Its Implications.” J Agric Food Chem 54(8):2891-2896, 2006. 10.1021/jf058158w

[5] Kavale KA, Forness SR. “Hyperactivity and Diet Treatment: A Meta-Analysis of the Feingold Hypothesis.” J Learn Disabil 16(6):324-330, 1983. 10.1177/002221948301600604

[6] Baenkler HW. “Salicylate Intolerance: Pathophysiology, Clinical Spectrum, Diagnosis and Treatment.” Dtsch Arztebl Int 105(8):137-142, 2008. 10.3238/arztebl.2008.0137

[7] Rangasamy SB, Dasarathi S, Pahan P, Jana M, Pahan K. “Low-Dose Aspirin Upregulates Tyrosine Hydroxylase and Increases Dopamine Production in Dopaminergic Neurons: Implications for Parkinson’s Disease.” J Neuroimmune Pharmacol 14(2):173-187, 2019. 10.1007/s11481-018-9808-3

[8] Chen D, Wang CY, Lambert JD, Ai N, Welsh WJ, Yang CS. “Inhibition of Human Liver Catechol-O-Methyltransferase by Tea Catechins and Their Metabolites: Structure-Activity Relationship and Molecular-Modeling Studies.” Biochem Pharmacol 69(10):1523-1531, 2005. 10.1016/j.bcp.2005.01.024

[9] Lu H, Meng X, Yang CS. “Enzymology of Methylation of Tea Catechins and Inhibition of Catechol-O-Methyltransferase by (-)-Epigallocatechin Gallate.” Drug Metab Dispos 31(5):572-579, 2003. 10.1124/dmd.31.5.572

[10] Egan MF, Goldberg TE, Kolachana BS, Callicott JH, Mazzanti CM, Straub RE, Goldman D, Weinberger DR. “Effect of COMT Val108/158Met Genotype on Frontal Lobe Function and Risk for Schizophrenia.” PNAS 98(12):6917-6922, 2001. 10.1073/pnas.111134598

[11] Karayiorgou M, Altemus M, Galke BL, Goldman D, Murphy DL, Ott J, Gogos JA. “Genotype Determining Low Catechol-O-Methyltransferase Activity as a Risk Factor for Obsessive-Compulsive Disorder.” PNAS 94(9):4572-4575, 1997. 10.1073/pnas.94.9.4572

[12] Gogos JA, Morgan M, Luine V, Santha M, Ogawa S, Pfaff D, Karayiorgou M. “Catechol-O-Methyltransferase-Deficient Mice Exhibit Sexually Dimorphic Changes in Catecholamine Levels and Behavior.” PNAS 95(17):9991-9996, 1998. 10.1073/pnas.95.17.9991

[13] Käenmäki M, Tammimäki A, Myöhänen T, Pakarinen K, Amberg C, Karayiorgou M, Männistö PT, Gogos JA. “Quantitative Role of COMT in Dopamine Clearance in the Prefrontal Cortex of Freely Moving Mice.” J Neurochem 114(6):1745-1755, 2010. 10.1111/j.1471-4159.2010.06889.x

[14] Kalhan R, Smith LJ, Nlend MC, Nair A, Hixon JL, Sporn PHS. “A Mechanism of Benefit of Soy Genistein in Asthma: Inhibition of Eosinophil p38-Dependent Leukotriene Synthesis.” Clin Exp Allergy 38(1):103-112, 2008. 10.1111/j.1365-2222.2007.02862.x

[15] Smith LJ, Kalhan R, Njoroge T, Gibeon D, Convertino M, Welti R, Holtzman MJ, Sporn PHS. “Effect of a Soy Isoflavone Supplement on Lung Function and Clinical Outcomes in Patients With Poorly Controlled Asthma: A Randomized Clinical Trial.” JAMA 313(20):2033-2043, 2015. 10.1001/jama.2015.5024

[16] Blacklock CJ, Lawrence JR, Wiles D, Malcolm EA, Gibson IH, Kelly CJ, Paterson JR. “Salicylic Acid in the Serum of Subjects Not Taking Aspirin. Comparison of Salicylic Acid Concentrations in the Serum of Vegetarians, Non-Vegetarians, and Patients Taking Low Dose Aspirin.” J Clin Pathol 54(7):553-555, 2001. 10.1136/jcp.54.7.553

[17] Lawrence JR, Peter R, Baxter GJ, Robson J, Graham AB, Paterson JR. “Urinary Excretion of Salicyluric and Salicylic Acids by Non-Vegetarians, Vegetarians, and Patients Taking Low Dose Aspirin.” J Clin Pathol 56(9):651-653, 2003. 10.1136/jcp.56.9.651

[18] Hosseini SA, Shateri Z, Abolnezhadian F, Maraghi E, Haddadzadeh Shoushtari M, Zilaee M. “Does Pomegranate Extract Supplementation Improve the Clinical Symptoms of Patients With Allergic Asthma? A Double-Blind, Randomized, Placebo-Controlled Trial.” Front Pharmacol 14:1109966, 2023. 10.3389/fphar.2023.1109966

[19] Arnsten AFT. “Catecholamine Influences on Dorsolateral Prefrontal Cortical Networks.” Biol Psychiatry 69(12):e89-e99, 2011. 10.1016/j.biopsych.2011.01.027

[20] Laidlaw TM, Boyce JA. “Aspirin-Exacerbated Respiratory Disease — New Prime Suspects.” N Engl J Med 370(1):21-32, 2014. 10.1056/NEJMra1403070

[21] Kopp E, Ghosh S. “Inhibition of NF-kappa B by Sodium Salicylate and Aspirin.” Science 265(5174):956-959, 1994. 10.1126/science.8052854

[22] Pearce FL, Befus AD, Bienenstock J. “Effect of Quercetin and Other Flavonoids on Antigen-Induced Histamine Secretion from Rat Intestinal Mast Cells.” J Allergy Clin Immunol 73(6):819-823, 1984. 10.1016/0091-6749(84)90453-6

[23] Manach C, Williamson G, Morand C, Scalbert A, Rémésy C. “Bioavailability and Bioefficacy of Polyphenols in Humans. I. Review of 97 Bioavailability Studies.” Am J Clin Nutr 81(1 Suppl):230S-242S, 2005. 10.1093/ajcn/81.1.230S

[24] Bhardwaj RK, Glaeser H, Becquemont L, Klotz U, Gupta SK, Fromm MF. “Piperine, a Major Constituent of Black Pepper, Inhibits Human P-glycoprotein and CYP3A4.” J Pharmacol Exp Ther 302(2):645-650, 2002. 10.1124/jpet.102.034728

[25] Needs CJ, Brooks PM. “Clinical Pharmacokinetics of the Salicylates.” Clin Pharmacokinet 10(2):164-177, 1985. 10.2165/00003088-198510020-00004

[26] Sommer DD, Rotenberg BW, Engel Patel R, et al. "A Novel Treatment Adjunct for Aspirin Exacerbated Respiratory Disease: The Low-Salicylate Diet: A Multicenter Randomized Control Crossover Trial." Int Forum Allergy Rhinol 6(4):385–391, 2016. DOI: 10.1002/alr.21678.

I held her hand that day.

I don’t know what made me do it. We happened to be alone together for a few minutes, before I left with the rest. They’d all already gone out of the room for some reason or other.

So I took her hand. And for a couple of minutes, there was calm.

She had lost consciousness a week before. I hadn’t reached out to her since. What was the point? Was she even there to feel me? I remember reflecting at the time how out of place it seemed.

The next morning we learned early what had happened. Not too early. Maybe we needed our sleep.

Obvious feelings aside, I wasn’t shocked. That’s not to say I was a particularly sane or well-calibrated teenager. Just dissociated enough from my feelings that across the ample days whose daylight hours we’d lurked away in those sterile halls, the conclusion had met little resistance as it settled to the floor of my mind. Inevitable.

My grandfather was offended when I did not cry just when he thought I should. I wonder if, grasping in vain to teach me that tears are not only possible but necessary, there was a moment when he considered that pressing harder on the insect would merely increase its resolve. You know, like a cat automatically leans into you when you nudge it.

My siblings were inconsolable.

The last thing thing she did for me was to sing me Happy Birthday. She slurred her words. Obviously a shadow, but not a distorted one; still the shape of herself, just blurry and sleepy and slow.

I think it was mostly the phenobarbital and the morphine and whatever other lesser evils were sparing her from unending torment and seizures.¹ Tumours had long since moved into her brain, but apparently they were not pressing on the right places to erase her before she went.

I can’t remember if she looked at me as she sung to me. I think she did. I want to assume she could still see me clearly.

I’ve never quite been able to let it go: why did I reach out like that? Did I predict what was going to happen that night? That I should say goodbye?

I don’t see how. Oh, I’ve ruminated on it enough. Am I forgetting something about what I perceived? Was she breathing differently? Was the look on her face subtly wrong? Or peaceful? Did the nurses sense something, and somehow telegraph it to an autistic child who could hardly look them in the eye?

If I’d been a cat, would I have curled up beside her, like the ones that pick out all the patients on the threshold?

I don’t know... I don’t know. There’s no record except the images in my mind, ever more distant.

It’s been 18 years. I miss you, mom. Maybe not as much as I should. I’m a strange person, but I guess you’d be proud of me anyway.

It hurts me to say that I don’t believe I’ll ever see you again. Instead, I’ve shared all these words with the part of you I keep with me. And it helped me. You helped me to say this.

You’re still teaching me. You see?

I miss you.

For how exactly could I hope to conjure a robust-enough depiction of robustness?

It felt as though, if I did not begin with just the perfect words I should never earn my license to start speaking. Yet, for so long I had also felt the crushing urge to speak, as if I might suddenly justify why I exist, once and for all. My words would pour and pour, until…

Irony’s a symptom. Trapped, I feared I had no right to speak on being trapped. Once the words were out, my prison would lay plain for you to see. And then I knew you’d say, “who are you to lecture on escape?”

I knew your words would burn me, to my shame, as I had wreathed my cell with judgment’s flames.

I denied myself this sacred beauty: I don’t know what you will say. Maybe you’ll judge; maybe you’ll cheer. Maybe you’ll listen, and maybe forget. Will you laugh? (Will I learn?) Maybe I will go unheard.

Still I’m drawn by my desire to speak and to be heard. And still at times I’m overcome by lowness, struck by some perceived inferiority, by my memory of all the vaster-minded, more-respected people who inspire the things I share.

In some ways still, my trap remains.

But now I’ve gone outside. And my symptom dissolved on approach, a mirage I had fearfully spied from a window, but never up close.

This post, this right here’s what first I longed to say. It didn’t feel right, starting with it. How would he, the guy I was, respect himself if in beginning, he began complaining and appealing? So here I am to make the futile reach to validate myself to you at last!

Well these few paragraphs will be enough to let it go, I think, now I’m not grasping near as hard for painful, pointless things I know will simply vanish when I do not grasp at them.

Subscribe now

Now, as if 44 footnotes weren’t enough, let’s expand a bit on some things I haven’t addressed yet, and which I sometimes still find kind of confusing.

1

Suppose a brain can enter a high-energy state, reorganize its activity, and rebalance its influences. Psychologically this should mean the freedom to resettle on better patterns of behaviour.

But really, being free to move doesn’t mean I’ll necessarily move to a better place. The context in which I transit a high-energy state is critical:

• If the system is being driven into bad patterns during the annealing process, then those patterns will also have a chance to contract into the influences of the system.
  
  We could think of PTSD as an annealing event gone wrong: excited by intense and unrelenting inputs, my brain enters a high-energy state and renegotiates its inner contracts under the duress of terror.
  
  This shouldn’t make us fear annealing in itself, any more than we should fear our capacity for learning more generally. While it’s possible to teach your mind to suffer, with no upside, that doesn’t mean you shouldn’t ever learn again.

• Meditation is nice because it’s semantically neutral with respect to sensory data. But the influences that already exist within the brain are not necessarily neutral with respect to each other! Inducing strong annealing by prolonged meditative practice might end badly if the practitioner does not carefully integrate the sometimes weird and dreamlike consequences that follow when boundaries are softened between flows that had previously been held apart.

• Drugs, such as psychedelics, may intervene directly in the brain to induce a high-energy state, more or less independently of the sensory data being consumed.

  
  It does seem pretty useful to have a pharmacological cheat code for annealing. But in the moments after someone takes a drug, it remains to be seen how the learning process will play out, given the sensory data they’re actually consuming. In fact they have to take special care, because they didn’t need to take special care to induce the state in the first place. This is already a meme: set and setting.

2

Michael Edward Johnson suggests that annealing might be the way we should view learning in the brain. The subtitle of his annealing post is in fact “Toward a Neural Theory of Everything”.

Okay… I don’t take this to mean that mechanisms like Hebbian reinforcement (HR) aren’t also “learning in the brain”, but that annealing is part of a larger account of how such lower-level mechanisms are coordinated, up to the behavioural level. I agree with Johnson (and others) that a central, unresolved problem of neuroscience is to bridge the neural and behavioural levels. It makes limited sense to ask how a synapse is changing, without asking what the organism it belongs to is doing.

3

I’ve played a bit fast and loose with thermodynamics. This series wasn’t meant to be formal, but I think I should give a little more, here.

Classic equations for free energy have a form like:

Here, U is the internal energy. The more energy is bound up in the structure of the system, the higher U is. A really hardened piece of steel with lots of little stresses has a relatively high value of U. There’s a lot of energy trapped in such a system; certain processes can potentially relieve the stresses and release that energy. So we might also call U the potential energy.

The interpretation is similar for brains. Higher U can mean more complex models, which are necessary and good because the world is complex. But it also means more unnecessary complexity, structural mess, or technical debt. That’s bad, when it interferes with the movement and harmony of the system.

Thermodynamics is about taking a bird’s-eye view: instead of keeping track of all the little details, we keep track of summary values like U, the total internal energy. The temperature T is another such value, summarizing how much kinetic energy all the system’s parts have — how much they’re moving around.

S is the entropy. It measures our uncertainty about the detailed state of the system, given that we’re working with summaries. If there are lots of ways that we could subtly rearrange the system without changing what the summary says, then the summary can’t tell us which of those subtle differences actually describes the system at the moment. So S is high.

Our equation describes how U, S, and T combine to give free energy F. But free energy isn’t really meaningful on its own unless we’re considering some process (like annealing) that changes S and/or U. So we write instead:

Here, ΔU and ΔS are the changes in the values of U and S, due to the process.1

Processes that are “allowed” to happen, that are spontaneous or favourable, have negative ΔF. This implies negative ΔU and/or positive ΔS: processes that release bound energy are favourable, as are processes that increase the entropy. Sometimes the two can be at odds with each other (e.g. negative ΔU and negative ΔS); changing the temperature T is one way to alter this balance.

The relaxation of stresses due to heating should have negative, favourable ΔU. But is the ΔS of destressing positive or negative? Watching explanations on YouTube and skimming materials science papers has left me confused. On the one hand, it seems intuitively correct that destressing decreases entropy: a more regular lattice is more correlated with itself. But maybe this is misleading, an overly static snapshot. Atoms are always moving, and if the atoms in a regular lattice are more free to shift around (i.e. their potential basins are wider) than tangled atoms are at the same temperature, then perhaps the regular lattice is higher entropy?

In any case, imagine that the ΔF of a process like destressing is technically favourable at a lower temperature, but in practice we still need to heat the system to actually see something happen. Why would that be?

Well, a process can be favourable, but that doesn’t mean it happens quickly enough to matter. The free energy equation is a summary of everything that happens between the endpoints of a process, and doesn’t account for the detailed dynamics of the subprocesses by which things actually play out in between. If we look at the ΔF between two room-temperature pieces of metal in different states, it might seem that one should just spontaneously change to be more like the other. But that won’t happen until we provide enough energy to activate the intermediate processes. The rate at which those processes happen typically accelerates rapidly as temperature increases.

Now, the free energy principle is concerned with a problem of information: how to infer which hypotheses H are the best explanations of some data D. In those terms, this is how it defines the free energy:

This looks complicated, but it’s analogous to the equations we’ve already seen.2 The two terms (“𝔼…” and “H…”) still correspond to energy and entropy3, and they still describe something like the internal complexity of our model, versus how uncertain we are about the state of the world.4

Needless to say I’m not exactly sure how all of this applies to annealing in metallurgy, let alone to the brain. (I’m not the only one.) But importantly, the free energy principle is a principle. It’s a statistical summary of what the system’s information will be doing overall, and not a claim about the details of the system’s implementation.

4

I’m a little confused about the balance between coherence and incoherence, and what exactly constitutes a high-energy state for the brain.

This paper suggests that the brain at baseline is already slightly above criticality, and that LSD significantly increases the disorder of the system:

However, at least in our model and as already described in previous work using similar methods [5, 6], we found that the resting brain in the placebo condition was already above the critical point—that is, the resting, the wakeful brain is in a supercritical state as observed through the Ising framework lens. This is consistent with the idea that, while mutual information peaks at the critical temperature, information flow in such systems peaks in the disordered (paramagnetic) phase [98], which is in apparent contradiction with other studies suggesting a subcritical nature of healthy brain dynamics [39].

And apparently there isn’t a single, invariant way to define criticality anyway:

Diverse and inconsistent uses of the terms “critical” and “criticality” have led to confusion. In this review, criticality has mostly referred to avalanche dynamics that behave at the limit between stability and instability. But other variants of criticality exist (Wilting and Priesemann, 2019). These include criticality between ordered and chaotic phases called the “edge of chaos” (Boedecker et al., 2012), criticality between synchrony and asynchrony (Botcharova et al., 2014), and multiple paradigms for the time-evolution of a critical phase transition as in extended criticality, intermittent criticality, and self-organized criticality (Saleur et al., 1996; Huang et al., 1998; Sammis and Smith, 1999; Bowman and Sammis, 2004). These forms of dynamical criticality are also distinct from statistical criticality (Mora and Bialek, 2011; Tkačik et al., 2013). How these all inter-connect is a topic of ongoing research.

Say we want to study multi-scale learning processes like annealing. How should we define and measure brain criticality, or high-energy states? Will the answer vary across the brain, the behaviour, the experiment? How will we notice when we are lumping too much stuff together in our summaries, relative to the causality we should want to describe?

5

I wouldn’t blame you for imagining totally distinct phases of relaxation and contraction in the brain, given my simplified description of annealing. But is it plausible that the mechanisms that reinforce versus rebalance the influences between neurons are engaged exclusively, one at a time?

Pure HR causes instability in networks because of runaway positive feedback between synaptic weights and firing activities. Normally though, it doesn’t cause instability in the brain. Something must be balancing things out; some other mechanism must be active around the same time. Homeostatic plasticity was one candidate for such a balancing mechanism, but it turned out to be much too slow to keep up with HR. Some other mechanisms, like heterosynaptic plasticity, are faster and might serve a similar role. But it remains to be seen exactly how the balance is maintained, and how much it depends on the detailed structure of individual cells (which varies quite a bit across the brain).

Anyway, we should probably not assume that 1) rebalancing does not happen outside of high-energy states, and 2) that HR cannot be active when the system is moving through such states. In fact, interventions that are intended to induce rebalancing might instead be “hijacked” by excessive HR and lead to further entrenchment of harmful behaviour.

we speculate that there at least two ways in which stress may contribute to canalization in brain circuits and synaptic connections, one would entail chronic stress-induced cortical atrophy (Dias-Ferreira et al., 2009; Duman and Duman, 2015) and another would entail a Hebbian ‘hijacking’ of stress-induced increases in TEMP (Brivio et al., 2020), see also (Parr et al., 2020; Friston et al., 2012b, 2021) – and (Dias-Ferreira et al., 2009; Moda-Sava et al., 2019).
- The canal paper

6

Given some of the things I’ve heard from meditators, I’d almost think it’s better to go through life without ever having contracted at all. But also I can’t imagine how it’s possible to totally flatten all of the attractors out of a living thing, without killing it.

I use “contract” very generally to point to the (emergence of) coherence in any dynamical system, though I put special emphasis on relations in living systems. This aligns with the very general view taken by the free energy principle, and also with all the common meanings of the word as far as I can tell.

On the other hand, sometimes practitioners are pointing at something specifically psychological when they say “contracted”.

Well, we might treat the mind as made of parts (thoughts, concepts, vibes, sensations, fragments, etc.) which interact and change according to certain dynamics. And we can think of flows through psychological state space as the movements of attention, or shifts in the active contents of our minds over time. Here, contracted means stabilized in attention; practitioners are looking to relieve themselves of attentional states that are too robust.

While I think we can assume that contractions of the mind are intimately parallel (or perhaps even identical) to respective contractions of the body, still not all contractions of the body may be felt. An expert meditator may achieve complete cessation of subjective experience, in which their attention remains totally uncontracted, unconscious. But their other body processes survive, preserving the skills and memories that’ll still be there for them once they uh, cease ceasing and wake back up.

Anyway, “contract” has nice semantic qualities whether we’re concerned with the dynamics of physics or of phenomenology. We just have to keep the context clear.

Subscribe now

7

The authors of the canal paper point out that their state-space representations are “over-simplifications”, which “depict dynamical phenomena, [yet] remain static images”.

There are a few related takes, here:

• When we model a process as a dynamical system, we need to decide which state variables to define or include, and how they relate to measurements of the process we are able to take. Maybe we’ll neglect some relevant aspect of the process, and in that sense our model will be inaccurate and perhaps “over-simplified”.

• Let’s assume we are capable of arbitrarily detailed measurements and calculations of just the right variables. The processes we’re interested in tend to have huge state spaces, easily over a million dimensions. Their dynamics are also potentially highly non-linear. Non-linear flows can be counter-intuitive, and this can be arbitrarily problematic in high-dimensional spaces. But many of the standard ways of analyzing and depicting states are based on assumptions of linearity; if we apply them inappropriately we will arrive at distorted interpretations of what the system is actually doing.

• Assuming the system’s parts are acting coherently, moving together, we can use dimensionality reduction methods to try to capture and analyze the system’s flows with as few variables as possible. But these methods cannot preserve all of the features of the flows as they exist in the high-dimensional space. We have to make assumptions about which features are worth preserving, and this might end with us going in circles.

• When we want to visualize a dynamical landscape, we need to use tools like dimensionality reduction to put our data in a low-dimensional, consumable form. Again, this requires great care. What features are we trying to represent to the viewer?

• Our depictions don’t need to be static. We can construct dynamical models of the brain; those models might entail evolving landscapes; accordingly, we might make dynamical predictions about the brain.

• Even our visualizations aren’t necessarily static, as we can produce animations of brain measurements and model landscapes.

And what about the equivalence we can draw between dynamical landscapes and probability distributions? Consider my confident prior belief that the sun will set around 8 pm today. We might represent this belief as a rather pointy distribution over the (easily interpreted) face of a clock. But suppose an agent of chaos doses me with an amnestic drug, and I become mildly confused about which day of the year it is: my less-confident belief would be represented as a less-precise distribution.

If we cleverly measure my behaviour, we can extract such distributions. This is observer-centric: it depends on the choice of relevant variables and how to measure them. But it’s not at all clear how a brain “chooses variables”, in the sense that there are uncertain relations between different kinds of information that are somehow encoded by its many parts. This is especially troublesome when in order to measure something or other, we decide to set up experiments that “unnaturally” constrain or collapse what might otherwise have been an open-ended and emergent process of interaction and attention in the world. So the results of our measurements may rightly be seen as snapshots, and their interpretation quite possibly corrupted by circularity.

But let’s be charitable. Canalization suggests that steeper (i.e. pointier) distributions should correspond to steeper gradients of dynamics: a peak in a probability distribution should map to some attractor or valley in the energy landscape, whose deepness and steepness is what’s really encoding the precision or confidence or peakiness of some belief. Mechanisms that decrease the entropy of a neural network (e.g. making the neurons fire more similarly) should decrease the entropy of whichever beliefs are encoded (i.e. making the distribution pointier).

That seems fine, and there certainly are formal ways to treat probability distributions in dynamical terms. But though this equivalence may be correct in general, it cannot make our particular representational and experimental choices for us. There’s no free lunch.

8

We’ve focused on over-canalization, or dynamics that are too robust. But under-canalization could also be a problem, when dynamics aren’t robust enough.

All mental disorders are caused by over-canalization? Wouldn’t you expect some to be caused by under-canalization? Where are they? The paper admits that psychedelics (which probably decrease canalization through 5-HT2A agonism) can contribute to some mental illness, but seem at a loss to explain this.
- Scott Alexander

The canal paper specifically associates the general factor of psychopathology ‘p’ with over-canalization.

[…] cognitive and behavioral phenotypes that are regarded as psychopathological, are canalized features of mind, brain, or behavior that have come to dominate an individual’s psychological state space.

But does ‘p’ correlate with disorders like autism and ADHD, in which there’s an incoherence in high-level predictions that seems a lot like under-canalization? How could it be a general factor of psychopathology, if it only accounts for one of the two tails of disorder?

I do think over-canalization will tend to show up more consistently than under-canalization in biological systems, as clenching seems adaptive in skill-limited situations, and under-canalization is by its nature incoherent. Still, is it even appropriate to characterize a behaviour or disorder as being either a disorder of over-canalization, or of under-canalization? If autism is a disorder of under-canalization, how can we explain the symptoms of rigidity (e.g. literalism and routine-following) that locally look a lot like over-canalization?

The Deep CANAL paper suggests we start to reconcile this by making a distinction between learning and inference, like we usually would in machine learning. This distinction roughly lines up with the one we made between influences and activity in the brain.5 Canalization during inference means the activity landscape is steeper, and the system is more robust against changes in its activity patterns. Canalization during learning means the loss landscape is steeper, and the system is more robust against reorganizing its internal influences (with respect to some goal).

In the brain, there certainly are differences between activity (e.g. more transient firings of neurons) and influences/connectivity (e.g. more lasting structures of synapses). But the brain is not a machine learning model. Dynamically, there is no clear distinction between an inference phase where the brain is active but its influences are held steady, and an offline learning phase where its influences are updated according to a formal loss function.6

Similarly, though we can construct convenient working categories of mechanisms and states, there really is no precise physical separation between what we might call my brain’s training parameters, and any other aspect of its state. There are just different types of state that update on different timescales and which we might represent in our models. Crucially, the brain’s own loss function is not a distinguished entity that’s held all apart from the rest of its functions, though some parts of the brain may deal more than others in the value of outcomes. A definitive duality between an activity landscape and a loss landscape is a representational choice. But the brain was not designed by an engineer imposing elegant categories.

The brain’s layering of processes which we might categorize as more inference-like or more learning-like is complex. So while the Deep CANAL paper is a good start in decomposing over-canalization versus under-canalization, I think we can go further.

How precise can we get about the architecture of the system? As much as different parts of the brain (e.g. cortical regions) may have local dynamics which are conditionally independent of each other, there will be local opportunities for contraction, too much or too little. How do these local contractions cohere (or not) into the contractions apparent at the level of behaviour, and its potential disorders?

What’s more, how will we know once we’ve accounted for all the parts of the system whose contracts might be relevant?

9

Which physical parts of the brain matter to our models? For over a century, the paramount view among scientists has been that the brain should be treated as a network of neurons: the relevant object of study is a single neuron, or a neural circuit. And all the evidence does suggest that neurons must be a larger part of the picture.

Is there a bit of institutional clenching happening here? What are some other brain structures which tend to be treated as mere sidekicks, yet which might be just as relevant as neurons to the functions of the body?

Let’s check out one example.

Blood vessel walls are filled with vascular smooth muscle cells (VSMCs). Adjacent VSMCs are connected by gap junctions, and can influence each other electrically. By contracting together, they can narrow vessels and constrict the flow of blood. Resonances can emerge across the vascular system.

An organ’s incoming vessels — its arteries, and the arterioles that branch off them — are more muscular and more in control of the flow than its outgoing veins. When different parts of the branching tree of arterioles contract differently, they can control the relative blood supply to different parts of an organ, and to different organs across the body. Classically this is seen as a mechanism for metabolic and homeostatic regulation: changing the relative blood flow changes which parts of my body have priority for nutrient supply, heat redistribution, and waste disposal.

Of course, all of that also applies to the brain. Some parts of my brain are in higher demand than others, and this keeps changing depending on the particulars of whatever I’m doing at the moment.

Michael Edward Johnson thinks something deeper is really going on here:

The vascular system actually predates neurons and has co-evolved with the nervous system for hundreds of millions of years. It also has mechanical actuators (VSMCs) that have physical access to all parts of the body and can flex in arbitrary patterns and rhythms. It would be extremely surprising if evolution didn’t use this system for something more than plumbing.

Johnson’s vasocomputation hypothesis is that blood vessels aren’t just witless infrastructure, but active participants in the local dynamics of computation:

• Both firing neurons and contracting VSMCs induce changes to local electromagnetic (EM) fields, and are probably sensitive to them. That means vascular contractions might be triggered by the changes to local fields induced by the firing of nearby neurons, but also that the firing of neurons may be affected by the changes to the local fields induced by nearby vascular contractions.7 That forms a closed loop, and potentially a feedback control system.

• As contraction can vary across the branching trees of arterioles, vessels in one brain region might be contracted differently from those in another region. So the computational properties of the neural-vascular feedback system could vary by region, with respect to the particulars being computed in different places.

• Apparently vascular contractions have a compressive effect on nearby neural resonances, causing them to collapse into more defined (i.e. less uncertain) states. That’s an anti-entropic effect; it should imply the canalization or robustening of a regional attractor. In that sense, vascular contraction could serve as an alternative or complementary mechanism to Hebbian reinforcement, though with some important differences in terms of implementation.
  
  One implication is that a region of the brain may be “locked in” to a particular prediction. Locally, while vessels remain contracted, neural dynamics will also remain contracted, constraining them to a subset of the dynamics, computations, or predictions they might possibly implement.

• The “locking in” of predictions might become chronic, because the latch-bridge mechanism allows smooth muscle to stick itself at a particular level of contraction without continuing to spend mechanical energy. In this way, brain regions could become stuck implementing something particular for a long time.

• The latch-bridge may help to explain how parts of the brain become trapped together in low-energy states, or how it is that poor exploration is a feature of certain disorders (like depression), and so on.

• Johnson proposes that the local robustening of network dynamics due to the induction of vascular contractions is the mechanism of tanha, the rapid attention-narrowing, vibe-doctoring process that’s the eventual source of suffering in Buddhist psychology.
  
  In particular, Johnson suggests that vascular contraction should occur in reaction to dissonance in the local EM field. To put it cutely: when neural harmonics go locally “out of tune”, this triggers a vascular contraction, which has a compressive effect that “squeezes down” the bad vibes until they collapse into something simple (even too simple, and at odds with everything else).

I find these hypotheses fascinating. The “clenching in the brain” analogy might be even more literal than I’d already suggested.

There are some things that I’m still mulling over:

• Do the scales work out? Capillaries, the smallest blood vessels, are not lined by VSMCs. They may have the means to contract, but whether this is relevant to regional brain function is unclear. How near do neurons need to be to arterioles of different sizes (particularly the smallest ones) for local field coupling to be non-negligible?
  
  In any case, coupling will fall off with distance. If neural-vascular EM coupling does matter, we might expect some structuring of the influences between neurons lying closer to arterioles, and those lying farther away.

• I’m not sure why vascular contraction should necessarily cause neural dynamical contraction, i.e. why contracting blood vessels should have a physical compressive effect on the local fields’ resonances.
  
  From an evolutionary point of view, it makes some sense: constricted blood vessels necessarily imply reduced blood supply, which prevents nearby neurons from remaining in an excited state at high metabolic cost. So a certain direction exists to the causality (contracted vessels → (metabolic constraints) → lower entropy neural dynamics) and it would be natural for evolution to elaborate on the same causal graph (e.g. contracted vessels → (field coupling) → lower entropy neural dynamics).
  
  Does say, physics provide a stronger reason why the local field coupling should necessarily proceed in this direction?

• There’s a puzzling tradeoff between vascular clenching (e.g. the latch-bridge mechanism) and resonance (i.e. vasomotion). On the one hand, field coupling may imply synchrony between vascular and neural dynamics. On the other, while a vessel is latched, vascular resonance may be completely damped and thus uncoupled from neural resonance.
  
  It’s easier for my entire vascular system to resonate when nothing is latched, and presumably that’s what can happen when I’m really mentally healthy?

• Johnson draws his own line between learning and inference in the brain:

  Phrased in terms of the Deep CANALs framework which imports ideas from machine learning: the neural weights that give rise to [self-organizing harmonic modes (SOHMs)] constitute the learning landscape, and SOHMs+vascular tension constitute the inference landscape.

  It isn’t clear to me, for one, why the latch-bridge mechanism is more appropriately seen as a mechanism of inference than of learning, except that this may be a convenient categorization for the sake of analysis.

Thankfully, Johnson’s hypotheses are falsifiable and we have the means to test them, if we will. But also of course, our general notion of contraction (or canalization) doesn’t hinge on the truth of vasocomputation alone.

10

We’ve already taken a bird’s-eye view of harmonic modes, which may be a useful way to frame the emergence of multi-scale features of brain activity. But how exactly do Atasoy et al. compute their connectome-specific harmonic waves (CSHWs), given some brain measurements? Here’s a summary of the methods:

1. Use T1-weighted MRI to identify the cortical grey matter surface, then split it into 20,484 pieces (in the 2016 paper at least).

2. Use DTI-based tractography to identify white matter fibres (i.e. tracts of axons). For each fibre, estimate which of the 20,484 pieces of grey matter it connects together.

3. Define a simple (unweighted, undirected) graph whose nodes/vertices are the 20,484 pieces of the grey matter surface, and whose edges include both 1) a short-range connection from each node to each of its nearest neighbouring nodes on the gray matter surface, and 2) a long-range connection between each pair of nodes joined by a white matter fibre.

4. Calculate the graph Laplacian, L. This is a straightforward calculation, but the output is a rather large 20,484 × 20,484 matrix.

5. Solve the following eigendecomposition:8

   The resulting h_j are the connectome harmonics.

We can list the harmonics in order, from lowest frequency to highest: h₁, h₂, h₃, and so on. Each harmonic is a vector (list) of 20,484 numbers, which we can interpret as the relative activation across the cortical surface.

These harmonics are based entirely on the graph structure inferred from anatomical MRI measurements, without any consideration of neural activity. So to support these results in their original 2016 paper, Atasoy et al. include an additional neural field model of activity, and they show how CSHWs might emerge from neural field dynamics.

Neural field model abstract away the particular connections between neurons, and instead treat the brain as a continuum with a particular geometry, over which neural activity can spread. Weirdly, the harmonic modes might actually depend more on the geometry than on the connectivity. But what could this mean, at the level of implementation? Don’t the resonant signals propagate through the “continuum” via synapses?

Maybe these neural field models are simply statistically valid and the specifics of neural connections become irrelevant on scales much larger than a single neuron. But shouldn’t such an assumption break down a bit in bottlenecked regions of the network (e.g. hub nodes)? I still find this rather baffling.

That’s it for now!

Earlier in the series, we asked: How do we see from outside? How do we unclench?

Now we can also ask: How can we escape a steep valley? How can we flatten its walls and re-sculpt the landscape? How can we undo over-canalization?

How can I smooth out the wrinkles from the fabric of my mind?

Every valley shall be exalted
And every mountain and hill brought low;
The crooked places shall be made straight
And the rough places smooth

- Isaiah 40:4-5 (NKJV)

We’ve seen that Hebbian reinforcement strengthens the contracts between neurons, and that this has the effect of contracting the paths of their firing activity in state space. It’s oh so convenient that the different everyday meanings of “contract” align with the different views we’ve taken on this phenomenon.1 This is not a coincidence.

(By the way, I’m going to start contracting “Hebbian Reinforcement” down to HR, though sadly I’m obligated not to joke about it.)

Another way to say that HR is contracting, is that it’s entropy-reducing.

How uncertain or variable are the neural firing patterns? That’s their entropy. A group of neurons act with lower entropy when they fire more predictably or redundantly. And the influences between the neurons are lower entropy when each influence is either very strong or very weak, rather than one of the many moderate values in between. HR causes influences to be either strong or weak, and neural activity to be more redundant and predictable, so as a mechanism it is entropy-reducing.2

I know you must be as excited as I am to have even more words to choose from to say almost the exact same thing. And you should be excited, because these new words will help us to make the most thrilling of physical analogies!

Consider a liquid that freezes into a solid crystal when we lower the temperature enough. The molecules of the material settle and bond into a lattice with an extremely regular structure. If we increase the temperature again, the molecules start vibrating into wilder configurations, until suddenly they break free of the lattice completely and return altogether to the liquid state.

The atoms of a metal also prefer (thanks) to contract into a kind of crystal. Say I have a piece of cold, solid iron that’s in this crystal state. Its atoms3 are arranged in a regular lattice, full of planes that can slide past each other, or be cut apart.4 The metal is relatively soft and ductile.

If I work the cold iron by hammering it repeatedly, I can knock its atoms out of their regular places. This hardens it, because the crystal lattice has become a tangled mess, full of little tensions set at odds with one another. And as long as the metal remains cold, the atoms are too chilled to untangle themselves. Nothing can easily slide past anything else. Hardened iron is a more suitable material for the blade of a sword or the beams of a building, though it’s also harder to rework — and more brittle.

Now heat up the hardened iron for a little while. The atoms gain enough energy to vibrate out of their stuck positions. Then slowly5 cool the metal so its atoms have time to settle into their favorite, untangled crystal lattice. This process of gradual heating-then-cooling to relax stresses and reform structure is called annealing.6

Importantly, we don’t need to fully melt the metal to anneal it. When the metal is cold and solid, it’s very ordered — predictably stuck in whichever pattern it already had, each unmoving part held fast by all the unmoving parts around it. When the metal is a hot liquid, it’s extremely disordered: it’s not even pretending to be a lattice, and its atoms are influencing each other much more incoherently. Near the boundary between these two extremes, there’s a sweet spot of entropy — where the coherence between the atoms has relaxed, but not to the point of total molten chaos.

This point is called criticality and it’s where the potential for the system to reorganize itself is at its greatest — where the heat-driven forces of entropy are at their most constructive. Different parts of the system come into energetic contact, without totally coming apart.

The child learns to believe a host of things. i.e. it learns to act according to these beliefs. Bit by bit there forms a system of what is believed, and in that system some things stand unshakeably fast and some are more or less liable to shift. What stands fast does so, not because it is intrinsically obvious or convincing; it is rather held fast by what lies around it.

- Ludwig Wittgenstein, On Certainty (144)

Back to our original questions: Say I’m trapped, flowing through some deep valleys, caught in some tangled mess of coping mechanisms and outdated adaptations. My brain’s parts are stuck, playing out the same patterns against each other. Nothing can move smoothly past anything else. Maybe it feels like I can’t learn anything anymore.7 Working the system makes it even messier. And it’s brittle — if it’s put under a very large and sudden stress, it tends to break rather than flex.

How can I clean up the mess, and pay off some of my technical debt? How can I unstick the patterns in my brain?

By analogy to annealing, we want to heat the system until we reach a high-energy state.8 In this state, if the analogy holds, the brain’s parts will have the chance to reorganize their patterns of activity — hopefully into something more untangled, parsimonious, or adaptive. As this continues and as the system cools down again, the new patterns should “crystallize” into the influences between the parts of the network, so that it can reproduce those patterns more easily even when it’s chilling.

So far, this analogy is pretty sloppy. How do we “increase the temperature” of the brain? When we reach a high-energy state, what allows the brain’s influences to relax, if they start off contracted? And what are the actual things I might do, to make this happen?

We don’t want to anneal the brain into a near-perfect crystal, the way some simple substances might settle if we heated and cooled them gradually enough. A perfect crystal would be an entirely redundant thing — it would “forget” all the history of how it had been previously worked. It would lose all its “memory”. Total amnesia is definitely not our goal.

But that isn’t a huge concern, since our analogy isn’t an exact fit for the brain. To see this more clearly, let’s return to the frame of predictive coding. A simplified view of predictive coding says something like this: to survive — or to achieve any goal, really — your brain needs to predict what your senses will say.9

Suppose the brain tries to cancel out upcoming sensory signals — say, from your eyes or your skin — by sending down prediction signals to meet them. When the two fail to cancel out, the uncancelled remnant is a prediction error which proceeds “upward” and excites the neurons directly responsible for sending down the bad prediction.

Extending this, we might treat the whole brain as a kind of prediction hierarchy. Each brain area G is excited by unresolved prediction errors from a “lower” area F, which are combined with predictions from a “higher” area H, and the result is another prediction error which G passes up to H, and so on, and so on. So prediction errors enter as bare sensory data, percolate up some hierarchy, and at each step there’s another renegotiation, another opportunity to suppress my senses and chill out.10

In this way, predictive coding gives us a way to think about “heating” the brain: we should excite it with sensory details it won’t suppress as predictable or boring.

Easy enough? Not really. Brains love to “predict away” entire worlds of sensory detail by wielding an all-purpose hammer like “just chilling — good vibes only! (That’s not a real tiger, right?)”.

Let’s make our annealing analogy a little crisper now, given that we’re treating the brain as a bunch of neurons and their firing rates:

• Compared to relatively simple crystals of metal atoms, the brain is a ridiculously complex web of cells. For one, an individual cell is far more complex than an individual atom, in fact containing many trillions of them.

• Neurons are of course made of matter — of many atoms joined together. But while the atoms in a lump of iron join into an ordered structure because of simple metallic bonds between immediate neighbours,11 the various “bonds” or influences that neurons have with each other are complex, layered, and sometimes distant. A neuron can be excitatory or inhibitory; it can grow an axon toward other neurons; an axon can join onto another neuron at many possible positions on its surface, forming a synapse; a synapse can be strengthened or weakened in a variety of ways.12

• Iron atoms in a high-energy state are jittering around in their literal positions. Neurons are in a higher-energy state when they are excited — when their firing rates are higher and more variable. That means increased movement less so in their literal position, than in the state space of their firing activity.13

• Whether iron is in a high-energy state depends on how much we literally heat it. Whether some neurons are pushed into a high-energy state depends on the strength of the signals reaching their part of the web, unsuppressed.

• In iron, the movement of atoms is directly related to the adjustment or rearrangement of the bonds between them.14 In a neural network, the activity of neurons relates to the influences (aka connectivity) between them more indirectly, in complex and various ways.
  
  A network can quickly be driven into an unusual pattern of activity, but some influences cannot change quickly—and some change more quickly than others. It’s faster to boost the signal at a given synapse, than to grow a new synapse, or a new axon. For example: my neural activity might be more chaotic for a few hours while I’m in a high-energy state, and yet the bulk of my long-term memories tend to be secure.15
  
  Importantly, there’s a difference between 1) activity patterns that are supported by the existing influences between neurons, and which are easy for the network to produce even at lower-energy states, and 2) activity that the system may be driven into by large signals coming from outside.

• A bit of iron probably isn’t going to heat itself. On the other hand, the brain is a living system that apparently maintains itself kind of close to criticality even when it’s chilled out, for those sweet computational benefits. So an everyday low-energy state might only be slightly below criticality, and entering a high-energy state might mean flipping to a state a little bit closer to (or a little above) criticality. This is fine, because small shifts can cause complex changes in the behaviour of a complex system.

OK. So neurons are excited into high-energy states when their predictions are inappropriate and fail to suppress the sensory signals which excite them. The increased entropy of these states means the network reorganizes into patterns of activity it previously would have had difficulty reaching. And some aspects of those patterns may be good to keep, especially if they would have better predicted the signal that drove the system into the high-energy state.

Keeping a pattern means forming a habit — re-contracting the influences between neurons in the network, so it can reproduce the pattern even when it’s chilling.16 But how can we make sure the existing influences can actually be altered, and don’t just stubbornly clench their way along through the surf of new activity?

We can’t say for sure it won’t happen. But neural influences do relax sometimes, or we would never be able to learn our way out of mental traps. Thankfully it’s plausible that short-sighted neurons can locally detect the fast, erratic high-energy firing of their neighbours and themselves, and in response they may have a mechanism that relaxes their existing influences with their neighbours, to make room for the reinforcement of new patterns.17

Heterosynaptic plasticity is an example of such a mechanism. While HR specifically strengthens a synaptic connection from one neuron to the next, heterosynaptic plasticity non-specifically strengthens all the inputs to a neuron. This re-balances the neuron’s inputs — reducing the relative influence that any other single neuron’s activity has on it.

The authors of the canal paper refer to this kind of high-energy-induced rebalancing as “Temperature or Entropy Mediated Plasticity” (TEMP):

Our model states that an effective intervention for psychopathology should, through an acute action analogous to an increase in system temperature or entropy — as per the Ising model (Suzuki et al., 2007; Ruffini et al., 2022) (see also (Singleton et al., 2022a)), trigger a downstream sub or post-acute effect that is analogous to a rebalancing or recalibration of synaptic weights, i.e., a counteraction to canalization.

a transient increase in TEMP within a system should help rebalance [flatten] its global state-space (Singleton et al., 2022b; Daws et al., 2022)

It might be appropriate to make a distinction between driving a single region of a brain into a high-energy state, versus driving the whole brain into one.

A small part of the brain might be annealed on its own, to achieve a more useful model of something specific. For example, maybe a specific part of my motor cortex becomes excited by a specific prediction error when I fail to land a specific throw, resulting in a bit of local annealing that might improve my game.

Still, my entire brain needs to coordinate with my body to maintain the behavioural context in which it’s even meaningful to anneal on specific errors. Why am I throwing something right now, and why should my motor cortex be updating on whether that throw failed or not? What game am I even playing? If a bunch of little brain regions are updating themselves separately all the time, what happens if they go out of sync? All the actors might get better at playing their own parts, while together they gradually lose the plot. So it’s probably useful for the whole (or most of the) brain to sometimes enter a high-energy state, where all the regions can reform together.

Doesn’t it seem psychologically risky for my brain to enter a high-energy state? Sure, it could be a short-term risk for robustness: as long as I’m in such a state, my brain probably isn’t able to coordinate as stably around any single pattern, including whichever pattern happens to be most effective one for my goals at the moment.

However, my goals are also contracted structures. Being biased against entering high-energy states means insisting that all my preferences and skills are already sufficiently coordinated into whatever structures they ought to have. Well… if my goals and skills are already optimal, then at best the reformation of influences in my brain will leave me performing no better than before. But it’s also a pretty big risk to assume that I am already close to optimal, or that my everyday brain is hot and critical enough to reveal any entrenched or concealed structural issues, before long.

At the level of behaviour, there’s a tradeoff between the risk of relaxing and the risk of contracting18, and I suspect we are biased toward contracting:

• Notice that once you feel threatened, everything is more likely to be perceived as a threat, including a person trying to convince you that you shouldn’t feel threatened.

• For almost the entire history of our evolution, mortal threats could appear often and without warning. When they did, there’d be no more time to learn, only to vigilantly execute whichever skills were already nearly loaded up. Why should I relax my attention? Think of my children! Think of the tigers. Hey, I’m pretty good at stabbing and running and hiding!

• Intelligence is a universal solvent, capable of inventing contexts in which tigers are truly irrelevant as threats. When was the last time you had to stab something? But intelligence is a relative latecomer: hiding and running and stabbing were the intellectual peak of threat-handling for hundreds of millions of years. That was the relentless context in which evolution laid the foundations of our brain’s architecture. And evolution cannot backtrack on its creations, only elaborate on them, however exquisite those elaborations may be.19

So we’re prone to a low-level clenchiness, a flinchy short-sightedness which can corrupt our intelligence. On the other hand, given the chance, we can moderate clenchiness by making wise decisions like “I should probably anneal a bit more than the heirloom inductive biases I got at Tiger World have led me to expect”. Our goals and skills are more complex than ever, and more in need of careful and repeated revision. How do you know you’re doing as well as you could? Evolution is merciless. Self-deception is easy. If there are tigers in this city, they are in a zoo.

Corruption or no, I shouldn’t be surprised when most of the time, my adult brain wields its powers of suppression to keep itself in a relatively low-energy state, where it can coordinate more robustly by leaning into the patterns it already knows, while suppressing “irrelevant” details.20 Over time though, circumstances may build and build to the realization that things could improve, and that progress isn’t being made by incrementally adjusting the individual parts while otherwise maintaining the status quo. Then it’s probably time to risk a temporary hit to robustness and experience a whole-brain annealing event: let the parts disintegrate and reform!

Michael Edward Johnson has been writing about annealing and brains for a while, and he’s made some deeper connections than I’ll make with this post.

One thing I want to highlight is his discussion of Selen Atasoy’s studies of brain harmonics. It’s new to me, but I’ll make a first pass connecting Atasoy’s work to what we’ve already discussed.21

First, here’s a new analogy: when we fix a guitar string at both ends and tension it, it tends to oscillate at certain natural frequencies, also called harmonics. This is because when each point on the string moves up and down, it influences each of its neighbouring points on the string to do the same. As long as they remain physically connected, neighbouring points must move together and oscillate at similar frequencies. The overall result is that when the string is excited, the entire thing “locks on” to a particular resonant state or mode where the entire string oscillates as a whole. The frequencies at which this happens are determined by the influences physically holding the string together, and their geometry. In the case of a guitar string, that includes the material properties of steel or nylon, the thickness and length of the string between the tensioning points, and the tension the string is under (i.e. how much we turn the tuning peg).

Again, a brain is more complex than this — but let’s speculate. Imagine a brain as a kind of complex resonator. Which modes will it resonate at? That depends on the influences between its parts, which we can also frame as the connectivity of the paths along which neural activity travels and loops and reverberates.

Harmonics may arise within a brain region due to the local influences, but also across the whole brain due to the larger-scale influences across all its regions, sometimes called the connectome. Atasoy’s theory is called Connectome-Specific Harmonic Waves (CSHWs); it’s a way to predict the large-scale resonant harmonics of the brain based on MRI measurements of the connectome’s structure:

[CSHW is] a method for applying harmonic analysis to the brain: basically, it uses various forms of brain imaging to infer what the brain’s natural resonant frequencies (eigenmodes) are, and how much energy each of these frequencies have. The core workflow is three steps: first combine MRI and DTI to approximate a brain’s connectome, then with an empirically-derived wave propagation equation calculate what the natural harmonics are of this connectome, then estimate which power distribution between these harmonics would most accurately reconstruct the observed fMRI activity.
- Michael Edward Johnson

Johnson proposed a continuum of harmonics, across scales of the brain. Smaller-scale harmonics — which he categorized as region-specific harmonic waves (RSHWs) — should have higher frequencies, be confined to local regions, and correspond to (sub-)behavioural particulars. Larger-scale harmonics (Atasoy’s CSHWs) should have lower frequencies, pass through many regions, and reflect behavioural coordination or summary states.

I find “CSHWs” and “RSHWs” a little disorienting to look at, so for the rest of this post I’m going to say LHs for little/local harmonics and BHs for big/brain harmonics.

LHs are high-frequency harmonics that are confined to local brain regions which deal in particulars. It makes sense that they are confined; we should expect different particulars not to interfere with each other by crossing their local boundaries, otherwise our brains might have trouble keeping track of particulars. (It does strike me as redundant to say that LHs are locally confined when we’ve already said that they are high-frequency, since the geometry of confinement is precisely what determines the frequency of resonance. I’m probably missing something.)

How can all these isolated LH particulars be coordinated into coherent behaviour? That should happen at the scale of the BHs, which being lower frequency have a more coherent identity across the whole brain. They travel widely, and should be able to interact with the many local LHs. And that goes both ways. For example:

• LHs to BHs: If the right LHs start resonating differently after being struck by some new and surprising details, their voices might be able to resynchronize the BHs and produce an overall shift in emotions or behaviour.

• BHs to LHs: An especially loud BH may be able to forcefully synchronize all the LHs and quickly suppress “irrelevant” details.

Johnson associates BHs with emotional states. This seems natural, if we treat emotions as a kind of high-level judgment the brain makes about everything it’s doing at the moment. We should expect such judgments to integrate many details from across the brain, and also to influence the processes that deal in those details. BHs are one plausible way in which this could happen.22

This framing is kind of nice, because “the brain maintains a status quo by suppressing apparent irrelevancies” can be bridged to something a little more intuitive or psychological: a game of “good vibes only” is always playing out in the synchronizations and suppressions of harmonics:

we doctor our [harmonic modes] *all the time*—when a nice sensation enters our awareness, we reflexively try to ‘grab’ it and stabilize the resonance; when something unpleasant comes in, we try to push away and deaden the resonance
- Michael Edward Johnson

Johnson makes an interesting connection to trauma and depression. In the event that some LHs become seriously miscalibrated with respect to the particulars they intend to model, it may make sense to turn down the BHs so that miscalibration (misinformation?) cannot spread across the brain like a harmonic infection and cause a “cascading system failure”. The result? The now-quieter BHs aren’t enough to coordinate the LHs, which fall more and more strongly into their own distinct routines, and in a vicious circle this makes it more difficult for a stronger BH to arise again and coordinate them. (Sound familiar?) And so sometimes the outcome is a muted cacophony of unreconciled horrors, bound by history, full of little tensions set at odds with one other.

Brain-wide annealing can be framed as relaxing and energizing the harmonics, so that the whole system can reharmonize. It’s like introducing Whoopi Goldberg to a bunch of nuns. And given that the brain is usually trying to maintain a status quo, we might expect annealing to happen locally to the LHs before it happens globally to the BHs.

Insofar as partitioning is possible in a broadly-coupled harmonic system, these [free-energy increasing] perturbations [that would induce annealing] will tend to be ‘local’ as the brain has strong incentives to preserve structure that doesn’t need updating.
- Michael Edward Johnson

Finally, let’s review a few methods we might expect to induce annealing. (We might also call them treatments to relieve clenching.) Keep in mind that these aren’t recommendations — the circumstances of your own life will determine the best course, of course.

I’ve largely based this section on insights I’ve obtained, again, from Michael Edward Johnson. Before we begin, I want to refresh by sharing his convenient summary of the neural annealing process.

1. First, energy (neural excitation, e.g. Free Energy from prediction errors) builds up in the brain, either gradually or suddenly, collecting disproportionately in the brain’s natural eigenmodes [i.e. resonant harmonics];

2. This build-up of energy (rate of neural firing) crosses a metastability threshold and the brain enters a high-energy state, causing entropic disintegration (weakening previously ‘sticky’ attractors);

3. The brain’s neurons self-organize into new multi-scale equilibria (attractors), aka implicit assumptions about reality’s structure and value weightings, which given present information should generate lower levels of prediction error than previous models (this is implicitly both a resynchronization of internal predictive models with the environment, and a minimization of dissonance in connectome-specific harmonic waves); 

4. The brain ‘cools’ (neural activity levels slowly return to normal), and parts of the new self-organized patterns remain and become part of the brain’s normal activity landscape;

5. The cycle repeats, as the brain’s models become outdated and prediction errors start to build up again.

Here, “self-organize into new multi-scale equilibria” refers to the fancy things that happen near criticality, the sweet spot between cold coherence and hot incoherence.

In particular, leading from step 3 into step 4 is where we expect mechanisms such as heterosynaptic plasticity to be activated so that new patterns can be reinforced in the influences within the network, to remain accessible outside of high-energy states.

Here’s a cute version:

1. “Heat” the brain, until

2. it flips into a high energy state, in which

3. activity reorganizes into new patterns, and after a while

4. things cool down again, and winning patterns contract into influences.

5. Repeat as appropriate/inevitable.

How do we initiate this cascade from step 1?

Meditation

Earlier in the series we discussed some meditative practices. Now our hypothesis is simple: when applied effectively, meditation pumps excitatory energy into the brain’s networks, potentially inducing high-energy states and reorganization events.23

Excitatory energy? This might seem to contradict the popular view of meditation as tranquil or something. But consider what it means, here: to excite our brain networks, we either need to stop our top-down predictive models from suppressing incoming sensory signals, or we need to boost the intensity of those signals. Or both. This rather lines up with the distinction between insight and concentration practices:

• Insight practices, such as mindfulness, sort of look like removing the habitual top-down suppression (or “judgment”) of “irrelevant” perceptions, allowing them to rise up freely. This should increase the variety (i.e. entropy) of the states we end up reaching.

• Concentration practices look like focusing all our attention on a single point of perception, greatly boosting its signal.

Outside of meditative practice, it’s totally possible for your brain to arrive in a high-energy state because of large, surprising signals related to your children, or your job, or an illness. High-energy states induced in such ways are like loaded questions: they do induce freedom of reorganization… but the system is forced to reform around something specific. Existing influences relax, only to forcibly contract on whatever meaningful and inevitable pattern is being driven into the network at that time. Sometime it’s a pattern we want, or will be happy we found, which is great. But sometimes we’d prefer to clean up the mess we already have, before taking on some more.

What’s special about meditation is that it isn’t necessarily about anything. Insight practice often avoids privileging any particular thought. And while concentration practice might seem to privilege some object of attention or other (e.g. the breath), consider that the end result of focusing attention to a truly single point is the collapse of all informative contrasts and associations. Because of this, we should expect these meditative practices to be very useful for cleaning up messes without forcing them to be replaced with something else.

Johnson has a term for practices that pump in “content-free” energy: semantically-neutral annealing.

“Semantically neutral energy” refers to neural activity which is not strongly associated with any specific cognitive or emotional process. […] usually energy build-up is limited: once a perturbation of the system neatly falls into a pattern recognized by the brain’s predictive hierarchy, the neural activity propagating this pattern is dissipated.

[…] effortful attention on excitatory bottom-up sense-data and attenuation of inhibitory top-down predictive models will naturally lead to a build-up of this ‘non-semantic’ energy in the brain

Psychedelics

Why are serotonergic psychedelics increasingly studied for the treatment of mental illness? Because they might be effective. Why should they be effective? The hypothesis is, again, that they can drive the brain into a high-energy state.

Meditation adds energy to the system by changing how we actively reflect, attend, and (dis)inhibit our mental processes, which are either identical or interdependent with our brain states (thus influence them). On the other hand, if psychedelics add energy it’s by directly intervening in the mechanisms of the brain: by turning a kind of chemical tuning knob, they cause some types of neurons to be more easily excited, so that you’ll go high-energy no matter your prior mental state. And the rebalancing mechanisms presumably still get activated — a high-energy state is a high-energy state, to a short-sighted neuron.

From the canal paper:

We are intrigued to speculate that heterosynaptic plasticity (Chistiakova et al., 2014)—seemingly the type of plasticity induced by the synaptogenic effect of psychedelics (Ly et al., 2018; Shao et al., 2021)—occurs downstream of an initial increase in the entropy of on-going (e.g., cortical) neuronal ensembles

The authors do acknowledge that this speculative mechanism might not be the only one, that the connection between increased entropy/energy and plasticity mechanisms is not known for sure, and anyway that things are probably pretty complicated.

When viewed through a lens of simulated annealing, other neuronal phenomena such as increased: asynchronous glutamate release (Aghajanian and Marek, 1999), expression of neural activity markers (Gresch et al., 2002), complexity of spontaneous oscillatory activity (Schartner et al., 2017), entropy of connectivity motifs (Tagliazucchi et al., 2014), high-frequency harmonics (Atasoy et al., 2017), near (Toker et al., 2022), or supercritical dynamics (Ruffini et al., 2022) sensitivity to perturbation (Jobst et al., 2021), and global integration (Tagliazucchi et al., 2016) can all be linked to either an early, upstream temperature or entropy increase (Schartner et al., 2017) or a later, downstream synaptic reweighting effect e.g., as implied by preclinical data (Shao et al., 2021; Daws et al., 2022; Moda-Sava et al., 2019; Raval et al., 2021).

This process may depend on how the acute entropic action relates to increased post-synaptic gain in deep-layer pyramidal neurons translating into an enhanced sensitivity to bottom-up signaling or prediction error (Carhart-Harris and Friston, 2019)

Two of the canalization authors (Carhart-Harris and Friston) published the REBUS model24 of psychedelics in 2019, which — along with adopting the annealing analogy — explained the action of psychedelics in terms of a reduction in precision-weighting on priors, which means the same thing as flattening the energy landscape.25 With respect to psychedelics, the main contribution of the canal paper was to give the REBUS model a more dynamical (and perhaps more intuitive) frame — in terms of shallower channels in an energy landscape, rather than decreased pointiness of prior distributions.

Exercise

There seems to be plenty of evidence that exercise is good for mental illness. It could lead to rebalancing in at least a couple of ways:

• Rhythmic or textural changes in perception can be exciting, but are often relatively semantically neutral.26 For example, running on uneven ground creates rhythmic variations which are not very relevant to higher-order predictions, but should not be entirely suppressed.
  
  As we move through a changing and unfamiliar environment, our sensory fields are filled with shifting textures and motions. As I begin my run, maybe my brain is wearing an inhibitory mask that filters out many of the details; but after an hour or two of attrition, the neutral energy might start to creep in.

  Let’s walk, or bike, or hike a while away

  until the world, long pouring through our eyes

  unfills our minds — and as recedes the day,

  our joy flows in much fuller; much more sound

  our words ignite; our dreams grow more profound —

  and with the sun but brighter will we rise.

• Exercise could induce specialized mechanisms outside the nervous system that ultimately lead to neural rebalancing. For example, if during evolution it was typical for exploratory behaviour to be selected-for following increased use of the muscles, then following exercise we might expect to observe our muscles sending signals to the brain to activate rebalancing or neurotrophic mechanisms which are adaptive for exploration.
  
  More generally, it may be possible to activate neural rebalancing mechanisms without first inducing a high-energy state directly in the brain.

There you have it: hot atoms, cool lattices, rebalancing acts, upward-downward negotiations, suppression signals, brain resonances, and clenching treatments. Quite enough for one post, no?

Next time, we’ll dig just a little deeper into some potential problems with all of this.

A stone is soft as wax,—tribunes more hard than 
    stones;
A stone is silent, and offendeth not,
And tribunes with their tongues doom men to death.

Titus Andronicus, 3.1

Subscribe now

“Canalization” is an awkward bit of jargon, right?

I'll probably keep using it. It’s good to lean on terms that already exist. At least a few people are already canalized into using “canalized”, and they’ll understand me better if I am, too.

And what about everyone else, the majority who don't speak fluent landscapese? Mental health professionals, can you imagine telling a patient "your energy landscape is over-canalized"?

I have a similar but less averse reaction to Scott Alexander's phrase trapped priors. It’s not hard to figure roughly what it’s pointing at, even if you haven’t studied Bayesian probability. Maybe “trapped” is doing more of the lifting, there. But trapped how? Sometimes I find myself wanting to evoke more of the embodied roots of the trap.

I'm no psychiatrist. Maybe y’all have this figured out. Do you need precise terms when speaking with patients? I guess it’s not so important for you to share literal jargon.

Still, haven’t we found some nice jargon! I worship idols as much as the next guy, but do we really need this? What are the best words that people already possess, that we can use as-is for our purpose?

As you’ve probably guessed, I usually prefer contract1 to canalize, and clenching2 to over-canalization. They call forth images not just of the strengthening influences among the system’s parts, or the pinched contours of its landscape, but also the tightness of mind and muscle we animals have suffered since long before the first mumblings about mathematics.

“Canalization” is certainly the more precise term, in a sense. When I speak it, I know that the people who’ll parse my meaning are those who are already close to seeing what I’d like them to see. The canal paper has experts speaking to experts. But in hindsight, at the time I first read the paper, I don’t think I was aching for a new word to talk about this particular phenomenon. Of course I enjoy having a more robust vocabulary. But when I use clenching as a metaphor for trapped behaviour, people catch on quickly because it’s hardly even a metaphor.

When I say “canalization” in public, on the other hand, I can’t help but worry that I’m doing someone a disservice. (I’m probably just clenching.)

There’s a cost to being precise. Precision is canalization. You and I are contracting over how to use language — over when to be precise and exclusive, versus flexible and wide. With respect to this tradeoff, we have our stances. All of this is part of mine.3

Here’s a challenge: find a usage of “contract” or “clench” that’s maximally at odds with the way I’ve used them in this series — a usage that doesn’t describe parts of a system coming together, and moving more predictably with respect to each other.

Oh, can you name some nicer valley I should move to?

Subscribe now

Where living things appear, so does robustening. And as is typical of inevitable human phenomena, people have described robustening many times before, more or less implicitly. In this series we’ve already seen it in the context of muscle clenching, self-fulfilling prophecies, and predictive coding. We’ve talked about the tradeoff between stubbornness and science—what it means to give structure out, versus take it in. We’ve also discussed some practices that people use to try to escape from harmfully robust mental states.

Now let’s elevate those earlier discussions with some ideas about dynamics, optimization, and brains.

We’ll pass arbitrarily close to the free energy principle. Some might say it’s what I’ve been talking about, all along. But literally everything is about the free energy principle, according to the free energy principle. More and more I’m stealing its language and ideas, but we won’t be focusing on it, this time.

This post is for those of us who are still beginning to understand.1

Everything changes — before too long, we should assume.

Sometimes a process changes slowly enough that we can treat it as though it is composed of some stable things or objects or parts. How long will the assumptions remain valid, that allow us to distinguish those parts? Perhaps things won’t change so much and so soon as to undermine our effort, and we’ll gain a little understanding before we’re forced to try again.2

A dynamical system is a precise, formal way of describing change over time. A classic example of dynamics is the movement of a pendulum, which swings back and forth under the influence of gravity. A simple pendulum is an easy thing to model as a dynamical system.

A complex system is a way of describing many parts that interact with each other. Out of all their local interactions, coherent behaviour can emerge. Individual birds in a flock make decisions that may be quite short-sighted, or local — something like follow the average direction of your neighbours. When many individuals get together and follow such a local rule, their movements cohere into flocking. They don’t scatter in random directions; something larger emerges from all the individuals. And no single bird has to ponder the flock as a whole, for the flock to behave as something like a superorganism3.

We can treat a flock as a dynamical system, with birds as its parts — its stable parts, which we assume aren’t doing much transforming into not-birds so long as we’re asking our questions.

We can describe the flock’s movement over time as a path through a state space. A single point in a system’s state space tells us the state of all that system's parts at a point in time. While the relevant state of an individual bird might be its current location, heading, and speed, a point in the flock’s state space gives us that information for every bird in the flock. Then a sequence of points forms a path, describing how the entire flock moves over time.

We can also treat a brain as a dynamical system. A brain isn’t made of birds: its parts are conventionally taken to be cells — specifically neurons. The state of each individual neuron might be its firing rate — how many times it emits an electrical discharge, every second. Then a single point in a brain’s state space represents all its neurons’ firing rates at once, and a path through the state space gives the history of firing rates for all the neurons over a given period of time.4

Maybe this all seems a bit too simple. A bird isn’t just a location, speed, and heading — actually it has its own brain full of neurons which effect its movements. And as it flies around, its muscles get tired and start to respond differently to the commands from its brain. There are many other bird-variables which might start mattering, depending on the questions we want to ask.

Similarly, a real neuron isn’t just some formless drum machine that pulses at a given rate, which we can summarize with a single number that changes over time. It’s much more complex. It has shape, metabolism, and a lot of other structure that isn’t captured by “firing”. Besides, neurons aren’t the only kind of cell in my brain.

We can wonder, how do researchers decide which variables, terms, or properties5 are important enough to include in their simplified models of a brain, or a neuron, or a bird? This question is related to whether the chosen parts of a system can be treated as stable enough to be modeled as such. It’s a very important question, but let’s leave it aside for now. Pretend it's just neurons and firing rates.6

Like birds, neurons don't act in just any way, to effect the larger whole of behaviour. To form a flock, birds don’t scatter in random directions. Likewise, my neurons don’t fire independently of each other. They don't follow just any path in state space, but something a little more ordered. Though each neuron acts locally, like a short-sighted bird, order arises because the neurons and their neighbours follow something almost like a rule — something like fire if enough of the neurons connected to me are firing.

The actual “rule” is more complex, though. A typical neuron has many other neurons connected to it, influencing it. It doesn’t listen to all of them equally. It doesn’t start its life with a grown-up balance of all those incoming influences. And it’s only one of many millions of neurons that needs to solve its own local variation of this problem, for useful order to arise.

Yet the neurons do learn to move together — to change all their local balances, to collectively form coherent patterns that effect how an animal acts in the world.

Learning means 1) deciding how well I’m achieving a goal, and 2) changing myself to better achieve it.

Looking at my brain as a dynamical system made of neurons, changing myself looks like adjusting the contracts — the strength of the influences — between neurons.

Of course I am not directly aware of the little adjustments that occur inside my brain as I learn. But that’s unnecessary, given that “I” operates at the level of goals, while the full machinery of my brain works out the little details. Neuron A starts listening to neuron B more than neuron C, and my behaviour changes very slightly. If enough of the right neurons adjust themselves, maybe I reach my goal a little better than I did last time.

And what about deciding how well I’m achieving my goal? Back in preschool, excited to reach for my favourite toy, I wished I would arrive in a definite state: I’ve grasped the toy in my hand. When we know exactly what my task is, we can use a loss function to describe how well I’m achieving it — precisely, explicitly, with math.

A loss function is just a way of keeping score. It takes some combination of information about my body, and how it just moved, and any targets it hit, and so on —and gives back a number. The score is kind of upside-down, versus what you might expect: the lower the loss, the less I’m losing. My hand’s getting closer to the toy, and it’s getting there faster, and my brain spent less energy to do it…

Imagine we’re measuring the activity of some of my brain’s neurons, and writing down their history as a path in a state space. We can’t measure them all; if we’re lucky, maybe we measure 0.01% of them. The many dimensions of the space describe the firing rates of the neurons we’re measuring. At the same time, we score my behaviour with a loss function. We can insert that score as an extra dimension in the state space. In this state-plus-loss space, the added dimension becomes the direction of “up and down” — downhill means lower loss means better.

To reach my goal better, of course my neurons can’t just fire independently of each other, doing their own things as separate individuals. And not only should my neurons fire in some interdependent patterns, and follow some kind of coherent path — they should specifically follow paths in the state-plus-loss space that fall downhill to better performance.

The firing patterns that are described by downhill paths are the useful patterns, in terms of the score we’re keeping. They’re the patterns that are not only coherent, but more coherent with respect to my goal.

Your loss

That’s all fine, if a loss function is given. But when is it explicitly given? Can I maybe design it? What’s the best scoring method? Well, do I know what game I’m playing? What is my goal? That’s not just something unchanging that’s imposed on me.

My brain wasn’t born with a hard-coded list of all the tasks I’d ever have to perform and precise ways to score them.7 We start our lives as babies squirming and flailing, trying to figure out how to even move our limbs coherently. Months later, we might decide to reach for one toy rather than another. And as we continue to grow, our reaching gets increasingly complicated and abstract. Eventually, we might learn some mathematics.

In each moment, even when I’m not thinking about loss functions, I act like there’s something I’m trying to achieve — a score I’m trying to boost. My brain interprets my senses, channels my attention, creates my task. And it implicitly keeps score, even when I’m not explicitly quantifying anything.

The brain’s own score isn’t a single number that we’ve calculated separately and tacked onto some meager measurements of a fraction of the brain. A state-plus-loss space like the one mentioned earlier is just a poor imitation of some even larger state space describing everything about the brain8 — something scientists so far cannot measure.

Anyway, the kind of information we’d use for scoring an explicit loss function isn’t something that individual neurons generally have access to. They’re like birds who know things because they watch their immediate neighbours. Meanwhile, I know the things I know because I am the flock. I don’t see them, and they can’t see me.

And yet, as parts of the complete and complex physical process of the brain, my neurons fly together down the slopes of an energy landscape, where “energy” is the implicit loss I’m always trying to minimize.

We exist as we have survived, and survive as we have predicted.9 A goal is like a mismatch between prediction and reality, and downhill is the direction of impetus or behavioural gravity which best attempts to resolve the mismatch. Should we be surprised that downhill movement — goal-directed movement — inevitably emerges in living systems?

We survive as we have predicted, and predict as we have learned. All the parts of my brain are always trying to fall downhill together, but in doing so they also continue to refine and redefine what “downhill” even means. Baby’s first goal is flailing and squirming, which transforms into reaching and grasping, and later into planning and writing. I move down an energy landscape—but the landscape itself changes!

Anything that pushes me uphill even a little is something I wasn’t predicting. Call it a surprise, or a disturbance, or evidence. If I had already learned about it, I would’ve predicted it. “Downhill” would already factor it in. My plans would cancel it out, in effect. I wouldn’t have been diverted off the slope. A thing can’t surprise me if I already truly, deeply accept it!

… but what if I don’t?

Then when the surprise hits, it’ll nudge me uphill.

And then?

I learn. I hope.

Our tendencies, habits, or preferences are like valleys that are worn into the landscape of our behaviour. Channels into which we tend to fall into and flow, like a marble rolling into the nearest groove. These attractive areas of state space are conveniently called attractors.

Flows in deeper, steeper valleys are more robust. When our path through the landscape is surrounded by high valley walls, then stronger forces need to interfere to push us over and out of the valley — and disrupt our habits. If we only get pushed halfway before the disturbance lets up, it’s easy to fall right back onto the path already followed.

So what’s stubbornness? It’s like a valley that keeps getting deeper and steeper; a region of the landscape that “refuses” to allow flows within it to be diverted, or for itself to be worn into some other shape, regardless of mounting evidence or disturbances.

Canalization is a concept which originated in genetics, to explain why phenotypes — for example that a fox looks like a fox, or that a bird nests its young in a particular way — may remain stable even when there are local disturbances, such as mutations in single genes, or changes in the weather.

Canalization is not about the mere presence of valleys or attractors, or else it's totally unoriginal. It's about the self-reinforcing nature of the process of channel-cutting or valley-eroding. It’s the tendency of the landscape to deepen and steepen. It’s attractorfication.

What if canalization is excessive? What if it keeps us captive, falling through our lives along the bottom of inescapable chasms of habit? What if a piece of critical evidence isn’t enough to nudge us over the walls of harmful normalcy? What if some large valley dominates the landscape, its many smaller branches or tributaries always pulling us in, no matter where we start? What if I never shut up?

Recent evidence hints at a “general factor of psychopathology”, conveniently called ‘p’.

If such a factor is real, and a reliable test shows I have more of it, then we’re more likely to observe that I have mental illness of any kind. Of course, whether I actually exhibit some mental illness or other still depends on the influence of specific hereditary or environmental factors — for example, being wealthy enough to experiment with therapies and experiences as it suits me.

Scott Alexander mentions that a general factor like ‘p’ is not the same as specific dependencies between disorders, where one condition tends to lead into another. He gives the example of developing an anxiety disorder because you're terrified of the demons you keep hallucinating. A general factor would increase the probability of both the anxiety disorder and the hallucinations, without requiring that one precedes or causes the other.

The 2023 paper “Canalization and plasticity in psychopathology” proposes that canalization is ‘p’.10

Our general theory states that: cognitive and behavioral phenotypes that are regarded as psychopathological, are canalized features of mind, brain, or behavior that have come to dominate an individual’s psychological11 state space. We propose that the canalized features develop as responses to adversity, distress, and dysphoria, and endure despite, rather than because of, evidence. We also propose that their depth of expression or entrenchment determines, to a large extent, the severity of the psychopathology, including its degree of treatment-resistance and susceptibility to relapse.

[…] we can imagine a valley as representing a cognitive or behavioral phenotype, feature, or ‘style’, and its depth and steepness is intended to encode its strength of expression, robustness, influence, and resilience to influence and change

(You should check out Scott's review of the paper for plenty more discussion.)

I find it hard to disagree with this. Anything we might define as mental illness and which was acquired through learning, may find its remedy through even more learning which shifts us away from the pathological pattern — so long as the pattern isn’t robust to the shift. It makes sense that regardless of the particular origin or character of a mental illness, it may persist when its dynamics are too robust.

Of course, the particular origin or character of a mental illness does still matter. I might say I learned a pattern that’s now so robust it’s hurting me—but which pattern is it, and where did I learn it? The paper spends some time exploring how canalization might matter to specific classes of mental disorders, for example those that are internalizing (e.g. depressive) versus externalizing (e.g. addictive).

A pattern judged to be too robust is called over-canalized. One of the canal paper’s authors is Karl Friston, the instigator of the recent formalism of active inference. So of course, the paper also frames over-canalization as predicting too hard:

Cast in Bayesian terms, the pathology we have chosen to highlight in this paper pertains to when the precision (or confidence) of prior beliefs (a prediction or model) becomes inappropriately high, leading to a failure of adaptability and the perpetuation of cognitive or behavioural entrenchment.

Over-canalization may be like pulling too long on a finger trap — while being caught in memories of situations where pulling had been a good way to free your fingers:

Moreover, apparently maladaptive phenotypes may initially have been adaptive for a given context - but critically, have become too entrenched, influential, and insensitive to new or changing conditions, i. e., they become, through canalization, insufficiently adaptive.

Canalization isn't inherently bad. We rely on “well-trodden paths” to make our way in the world. The answer to clenching cannot be to never contract any muscles again. We can view all tendencies or habits, even good ones, as channels in which we prefer to flow. For example, when performing familiar tasks we can often be very confident about the sequence of movements we need to make.

Given that you’ve started moving your leg to walk, you have a high prior (or an “extremely precise belief”) that you should bend your knee a certain way.
- Scott Alexander

Whether a behaviour is “too robust” depends on context.12 Here’s the tradeoff: the valleys in the landscape should be deep and steep and stable enough to keep up our good habits, but not so deep and steep and stable that new evidence or experiences can never push us over the walls of the valley, and begin to sculpt even better behaviour.

The right amount of compression pressure is not zero.
- Michael Edward Johnson

Hebb and flow

So far we’ve seen that:

• Viewing a brain as a dynamical system, the paths the system follows in state space describe the overall patterns of “movement” of the brain’s parts—say, the firing rates of neurons over time; we could also call these paths the system’s flows.

• The parts of living systems generally flow downhill on an energy landscape, improving a physical score which, if we could quantify it, would describe how well the system is achieving its implicit goals.

• The shape of the energy landscape itself changes as part of this process—in other words, learning can occur and goals can change.

• Valleys or steep areas in the landscape may tend to be self-reinforcing, and we call this tendency canalization. It keeps behaviour locally robust, and it may be either adaptive or dangerous, depending on what the behaviour is robust to.

Now: when we speak of valleys, we’re looking at an overall description of the state space of the brain or mind. That’s a top-down view of the whole forest — of the patterns that emerge because of the local actions of all the system’s parts. But what’s changing about those individual actions, that collectively results in canalization or robustening?

My brain wasn’t born with a hard-coded list of all the tasks I’d ever have to perform and precise methods to score them.

Memories and behaviours are things that gradually emerge from the coordination of my brain’s parts, as I interact and gather in the world.

Baby’s first goal is flailing and squirming, which transforms into reaching and grasping, and later into planning and writing.

Babies’ movements grow more coherent and complex, as their brains have ways to update the structure of their cells in light of the perceived success or failure of their movements to achieve certain results.

“I” operates at the level of goals, and the full machinery of my brain works out the little details.

Scientists have discovered a number of mechanisms that neurons use to adjust how they influence one another. One of these is Hebbian reinforcement, which the authors of the canal paper hold especially responsible for canalization.

Why?

The rough idea of Hebbian reinforcement is this: when one neuron (N1) fires and even slightly influences a next neuron (N2) to fire shortly afterward, the connection between them tends to be strengthened, so that N1 influences N2 more strongly in the future. N2 associates harder with N1 — it comes to fire more predictably with respect to N1.

When neurons fire more predictably with respect to each other, that implies a contraction of the landscape they move over. You don’t need a 100-dimensional landscape — 100 different changing numbers — to describe the path in state space that 100 neurons follow, when they all fire similarly. So Hebbian reinforcement leads to redundancy. At the extreme of redundancy, when all neurons are firing exactly the same way, we could describe them all at once by remembering just one number, over time.

Every single one of those damned birds is heading due east at the speed of light.

There are two related and crucial effects of Hebbian reinforcement. When a system of neurons is stimulated by novel information from outside the system — say, visual patterns from the eyes — that information will drive the patterns of N1-then-N2 firing in the system. While that’s happening, Hebbian reinforcement would contract the system onto firing patterns that reflect the structure of the information. In other words, the strengthening influences between neurons will reflect the strength of the associations in the sensory data. So the first crucial Hebbian effect is the acquisition of new patterns.

But we’re not concerned just with whether a pattern is present or absent, but with how present it is. A little Hebb can incept a pattern; a lotta Hebb can make sure it’s the only pattern around.13 So the second crucial Hebbian effect is that neurons obtain solidarity — they vibe14 more strongly.

For better or worse, it’s harder to disrupt a strongly shared vibe. This is our plainly-stated hypothesis about how the steepening of the valleys occurs: the steeper boundaries in state space reflect an increase in the coherence of the dynamics of the system, which is due to a mechanism that locally reinforces influences.

In parallel to redundancy, Hebbian reinforcement should also lead to exclusivity. If neurons A and B share a little of vibe X, we can reinforce vibe X between them. And if neurons C and D share a little of vibe Y, we can reinforce vibe Y. At the beginning, neuron A might have been able to influence neuron C with a bit of vibe X, and neuron C might have been able to influence neuron A with a bit of vibe Y. But as the A-B and C-D groups vibe more strongly on X and Y respectively, the members of each group become less likely to influence the members of the other. In fact, to maximally reinforce their own vibes, they should really try to suppress each other. So we might expect Hebbian reinforcement to partition (rather than mingle) a networked system, splitting it into self-reinforcing vibe cliques15 with exclusive dynamics.

These insights about redundancy and exclusivity are simple enough to apply to a small network — but what about a brain?

For example, some regions of my brain send down signals to my muscles through my spinal cord, to control my movements. What happens if different parts of those regions vibe among themselves in conflicting ways? I can’t take multiple, mutually exclusive actions — I can’t reach one of my hands to the right at the same time as I move it to the left. Brains have to solve this problem by selecting (at most) one of the exclusive alternatives16, or by finding a compromise when it makes sense to do so.17

Because of behavioural constraints like those, but also because of the particular architecture of the brain, we have to be careful when using our intuitions about Hebbian reinforcement to try to describe the relationship between neurons and behaviour. My brain’s entire state space isn’t a monolith that can split any which way into vibe cliques. The different dimensions of the space represent the different parts of my brain, and those parts aren’t all directly connected.18 The Hebbian mechanism acts locally between parts that are directly connected, and how its reinforcing effects spread through the network depends on the complexities (and oddities) of the actual architecture.19

The authors of the canal paper see Hebbian reinforcement as just one potential part of the mechanisms of canalization in the brain.

We speculate that the canalization could occur via (at least) three mechanisms: 1) an atrophy-related reduction in abundant synaptic connections, selectively sparing well-reinforced synapses and associated circuits, 2) top-down inhibition or neglect of specific circuitry (e.g., linked to traumatic memories), or 3) strong potentiation of selective synapses and circuits (Bliss, 1990). All scenarios should reduce freedom within the global system by biasing certain circuits or sub-states.

The potentiation or atrophy of the influences between neurons occurs through local mechanisms, but its relationship to behaviour emerges in larger-scale interactions across the brain. In a given circuit, Hebbian mechanisms may strengthen some vibes to dominate over others. But at a critical moment, maybe a circuit’s now-dominant vibe makes a cringey or dysphoric contribution to the larger negotiations playing out across the brain, and the entire circuit ends up inhibited by the processes that select non-conflicting behavioural alternatives. Is Hebbian reinforcement also at work in the selection machinery, to make this inhibition more robust? If it is, wouldn’t that be in opposition to what Hebbian reinforcement originally achieved in the now-inhibited circuit? If we’re interested in canalization at the level of behaviour, clearly we won’t always be able to associate it with a single Hebbian reinforcement event.

Exclusivity can play out in funny ways. One wrinkle in the system offsets another, and before long it’s wrinkles all the way down.

And should we be satisfied with that?

For everyone strives to keep his individuality as apart as possible, wishes to secure the greatest possible fullness of life for himself; but meantime all his efforts result not in attaining fullness of life but self-destruction, for instead of self-realisation he ends by arriving at complete solitude. All mankind in our age have split up into units, they all keep apart, each in his own groove; each one holds aloof, hides himself and hides what he has, from the rest, and he ends by being repelled by others and repelling them.

- Dostoevsky, The Brothers Karamazov

Subscribe now

Here are some of mine.

I’m hardly still that much the cynic you will meet below, but I’ve recalled his pain.

Labor

What kind of music's even left to write?
What fills my soul but something someone's felt
before? If I forget, can this be new
again? Must living mean dying? Oh why,
why don't we want to admit what is true?
In just a second, inevitably
a quiet space will empty in my head
where for a moment longer, surging sound
—ecstatic, holy, newborn, beautiful—
reigns indisputable, and death's a joke.

Social odia

In this space so vast and liminal,
slighted, I get slightly criminal—
seizing on today's defection,
ruining some cringe erection.

Elon's sub—or was it, Greta's?
terse—and much too versed in meta,
stressed—obsessed with alpha/beta,
raging—hemorrhaging my data.

On this branch each bird picks up a tic
and talks and talks and keeps god sick
with blue bird flu or some sus shit
behind a mask, before the front of it.

In your face and supercritical,
pointing, psycho-analytical—

frightened, righteous, egotistical,
twisted, traumatized, statistical—

paralyzed and apolitical,
staring down the parasitical

in this space so bare and clinical
marketing our best infections,
mirrors stiffen into cynical
clenching clenching clenching clenching
clenching

Contract

A minute since the heavens wept
and wet the moss which sponging mid my toes
solidified my spirit out the fog
of swimming shapes and incoherent ghosts.
Now reining in my eyes, down from the clouds
to watch the dancing lights upon the weave,
these ripples over silt and stones: a world
endures in just this corner's corner's edge!
Yet Earth moves ceaselessly beyond my sight—
its boundless tears of joy and pain misplaced
by me, in tending with this waving light.

Ineffability

I cannot write without a thought
for words, explicit forms, that strip
the meanings from my mind, like trees
torn from primordial tracts, or clothes
from off the prisoner's weary back.
Whatever bliss or fear I'd find
and living lightly, set it free,
instead I scheme to court with corpses
feeling nothing, noisily.

Content warning

To whom it may concern: just go ahead
and plunge and dig your claws into my chest.
Just break my ribs apart my sternum. Peel
the lining from my bones. You hear it peel?
Go on—put on your act and cringe in shock,
as though you weren't my very butcher. Feel
whatever you feel while you grasp my heart,
like you have always wanted. Squeeze it dead.

Theory

They claim to seek our optimum
but what is optimal to them
was drafted as they grew into success
and ratifies itself inside that game.

You'll hear them say from time to time,
alone and analytically,
"don't ditch the baby with the bathwater"—
then, en masse, in print, denounce entire fields.

Well how else should they stop the funds
from flowing where they shouldn't flow?
How will a human cope with clenching?
Clench?
Now everyone, please form your gangs.

Modern

There can be no more consolation got
by those now gone beyond our skill to save;
those all too many never will give thanks
for futures spared for them not soon enough.

All terrors past will always have occurred:
the blinding pain that bore our crueler ways;
the wounded, confused, stifled whimpers of grief;
the restless longing fading into black;
ultimately, the silence.

Indignantly and proud I seek the dawn
a second sun shall rise to share the sky,
and we'll forget that once, one couldn't ask,
"so where were you, the day the Devil died?".

Hereafter

Invigorated by the urge to speak
and keep on speaking, all too much to share
—an hour, a day, a year could not suffice—
my mouth rides on, inertially possessed.
Words pursue words no listeners will catch.
My audience disbands, but still there's hope
convulsing miserably in the dust,
chanting the unbelieving mass of death,
hunting for one more fragile syllable
in vain to banish silence for all time.

Subscribe now

During a recent conversation about how best to situate my use of the word "clenching" to respect existing knowledge and terms, Cube Flipper mentioned an uncanny parallel with the Buddhist concept of taṇhā, and suggested I read Romeo Stevens’ post (mis)Translating the Buddha.1 It was my first encounter with his writing, and the first time I really engaged with any writing about meditative practice.

Stevens explains how the conventional Western translations of Buddhist psychological terms are largely ill-chosen, even harmfully so.2 He gives revised definitions for some terms from the Pali canon, a major text of Early Buddhism. He emphasizes that these terms refer to identifiable mental events, not to vague metaphysical energies, like we might mistake "impermanence".

Tanha is the cause of Dukkha. Understanding Anicca and Anatta inclines the mind toward Nibbana by removing the Upadana that maintains the creation of Sankharas.

The above sentence full of Pali terms is about detectable mental events within the stream of physical sensation, feeling tones, mental talk, and mental image that makes up moment to moment experience. If something sounds weird it's just that you've never reified it before because it goes by very fast (10-40hz range). Meditation is about training the mind to be able to notice these more subtle events and then instructions for noticing certain things about the causal relationship between these subtle events and how good your moment to moment experience really is. The purpose of meditation is not to become a really good meditator, to experience certain cool temporary states (though some are helpful), etc. But direct insight into the basic building blocks of your experience.

I recommend reading through the entire post. Let’s walk through some parts of it and relate them to what we’ve already seen in this series: clenching, or robustening, or predicting too hard.

Before we begin, we’d better start by considering whether different people will even experience the same basic classes of mental events. Are some shared universally, and already waiting to be noted in the workings of each of our minds? Or are they created like self-fulfilling prophecies, as patterns that meditators incept in themselves under the influence of culture and memes? How consistent is human phenomenology?

I haven't had much time yet to digest the terms we’ll discuss. How might they refer to things I've already noticed about the movements of my mind? Do some of them refer to things I've never noticed? Maybe. I guess I'll have to meditate more. Will I notice which new things I create that weren’t there before?

I’m not sure. For now, I remain confused. I’m writing this to air out my current understanding and confusion.

Which brings us to one more preliminary moment of skepticism. Remember that we tend to take on technical debt? We’re evolved animals, impulsive about pushing through, enduring immediate encounters by any means, more than slowing down to perfect our minds before we act. We pile on OK solutions that work well enough for now. The more we’re stressed or uncertain, the harder we lean into the mess of good-enoughs we already have, patching and patching them instead of feeling free to stop to untangle and consolidate them.

Well, we’re also that way with words and ideas. So we should be wary of using semi-formal language to talk about the kinds of mental events that precede language. It is easy enough for a mind to objectify something it fancies about itself… to reify, to substitute something real with merely your idea of it. When we give our ideas names and build them into stories, they can achieve their own stable lives in our attention, even if the things they were originally intended to reference aren’t so stable themselves, or even real.

Sometimes we lean in a little too much, stubbornly steamrolling the subtlety of reality beneath our too-solid story of it.

A lot of the confusing stuff in buddhism is trying to talk about these direct insights. e.g. 'grasping' isn't talking about a cognitive understanding but rather a direct perception of an automatic mental move that happens in the pre-conscious perceptual stack (at least pre-conscious prior to a bunch of work). Buddhists do this because sometimes, if a person is ready, you can speed things up by just pointing the thing out rather than waiting for them to figure it out all on their own. […]

I do think that this has resulted in ambient memetic immunity of the same type hypothesized by Scott Alexander and others about why new psychotherapy methods work for a while and then seem to stop working. People get some sort of idea of what all these experiences are supposed to be and as a result ignore actual moment to moment sensation.

- Romeo Stevens, “Dukkha: created vs discovered”

Sometimes you can make prophecies come true by faking it. But if you really want insight into the workings of your mind, you can’t keep dressing them up in stories. Your actual mental events are not your firstborn concepts of those events, nor your second-born, nor hundredth.3 Eventually my perceptions and ideas and words may reach a sufficient approximation of nuance that their interplay can do good work to relate some of my inner movements. But that’s secondary to learning to pay any attention to those movements, as long as I’m concerned with actual introspective competence. If you lean too hard or too often on any particular story about what an experience should be, perhaps you’ve acquired a piece of identity that may stubbornly override your capacity for honest experiences.

So each time you discover a pleasantly novel personal insight or meditative state, if you find yourself excited to name it and hold it and keep it: beware the little debt trap you might be building.

The investigation you do into your own experience has to be a real investigation, and not one in which you are highly confident about what there is to find. i.e. people feel like they're meditating wrong if their experience doesn't seem to match whichever map they are using.

- Romeo Stevens, “Dukkha: created vs discovered”

Of course that doesn’t mean you can or should abandon words and ideas completely. Just, maybe don’t grasp them so quickly or hold them so tightly.

Don’t flee so fast from confusion.

And when I say you're dreaming, I'm dreaming too. So if you were to 'agree' with these words as right, I would name that nothing more than a way of offering condolences for the demise of their strangeness.

- Zhuangzi, Ch. 2 (trans. Ziporyn)

Well then, let’s get started. Here’s Stevens' explanation of taṇhā (emphasis mine):

taṇhā is usually translated as desire or craving but this is wrong and misleading. taṇhā is more literally translated as 'fused to' or 'welded to' [all bold emphases are mine]. It immediately follows the mental moment that you zoom in with the attentional aperture on something. It could be that a flower or an item on the shelf at the supermarket captures your attention, or you turn your head to catch more detail as you pass by an accident on the road. Many hundreds of thousands of such events take place in the course of a single day. With most of them attention then relaxes and makes space for the next thing. But with some small proportion you find the mind doesn't quite 'unclench' from the object or some aspect of the object. This tension aspect is why it is sometimes translated as ‘grasping’ which is closer. Imagine something you aren’t finished with being pulled out of your hand and you tensing your fingers to resist.

As I read this for the first time, immediately as my eyes landed on 'unclench', my mind did a kind of apocalyptic Dolly zoom. I hadn’t seen anyone use my word like that before. That old ugly thought, best translated "I'm not original!" reared up as suddenly as it hadn't in a while. It wouldn't let go for a few minutes. I could hardly see past it.

I realized that while taṇhā might not be identical to what I'd been calling "clenching", the kinds of stuff described by the two terms at least had to be intimately connected.

So taṇhā was part of what I was experiencing in that very moment? I was clenching on the use of “clenching?! Such an absurd recognition, a ridiculous shock, immediately satisfied and terminated the conditions of the painful discord.

A moment later, there was only joy.

Since then, I think I’ve noticed taṇhā on a short subjective time scale when there are sudden unexpected sounds during meditation. Across the room, an appliance makes a small but sharp popping sound, and it also pops in my awareness for a fraction of a second. It’s like the “spacetime” of my perception is suddenly and subliminally warped towards the source of the sound in space, as it seems to me. At the same time, my body snaps into hidden preparations to move in the opposite direction, away from the blip. But I’m confident the source is harmless, and there’s nothing top-down to sustain the contraction, so it quickly relaxes away. (Mostly.)

In hindsight I do think this contract-relax sensation was there all along, discovered more than created. I just didn’t make note of it as clearly in the past, and hadn’t settled on a name.

Importantly, taṇhā names a specific subjective experience, an attentional move that we all do automatically and preconsciously, on the order of ten times per second while awake. But the way I use “clenching” is broader than that: I’m pointing at a general principle of dynamical robustening, of which taṇhā is one very important expression. Other instances may happen in parts of our bodies which we do not perceive as conscious. They might also not be local to a single mind or body at all! Societies and cultures also clench.

Continuing on, we have upādāna. Stevens says that upādāna immediately follows taṇhā4 and that it involves processing potential aspects of a vibe/object and filtering out the ones that feel the least stable or pleasant:

Upadana is literally translated as ‘fuel’ but also is used in gardening metaphors to mean seeds as well as having connotations around ‘pulling towards oneself.’ I already alluded to why this is an important concept with the idea of maladaptive strategies that generate their own fuel. Upadana is a mental event that immediately follows Tanha. It can be thought of as the opposite of Equanimity. We instinctively pull or push away aspects of mental objects/representations that we do or don’t like. We try to ascertain the aspects of objects that are stable, controllable, satisfying so that we can own, or associate with those aspects. We ignore or try to push away aspects of the objects that make us feel hopeless, helpless, worthless, empty.

“Follow the good and stable vibes” makes sense as a view from inside a greedy process of something like active inference: we try to isolate what we think we can reliably observe or predict or enact, for immediate profit. We lean into the search for a model that we can lean into harder, instead of budgeting for elegance and integration.

Frustrated pulling on a finger trap, what does a child count as unacceptable violence? Maybe, anything that keeps them from pulling.

The process of taṇhā-upādāna happens over and over across many moments. It grasps constantly for simplifications. The longer-term result is that all the simplifications become aggregated and organized into structures called saṅkhāras:

Sankhara can be translated as either 'that which has been put together' or 'that which puts together.' […] If Upadana is a seed, then Sankharas are the warped houses we build out of the twisted lumber that grows. Living in these poorly made houses we don’t understand why we are miserable. To speak less metaphorically, a Sankhara can be thought of as a collection of mental events put into a story about how the world is.

Saṅkhāras are your bullshit. They’re fetishistic expressions of the mess of your technical debt; bureaucratic propaganda from parts of your mind (asserting they’re) keeping things running on time; the terms of the constitution of a tyrannical little government.5

Together, my saṅkhāras form a kind of toxic ecosystem in which all my self-centered simplifications, no matter how individually pathetic, are granted niches that mutually support each other, and influence the outcome of the next round of taṇhā-upādāna in a kind of vicious circle.6 If I start waking up, if I try to fix one part of the ecosystem, I risk destabilizing other parts of it. I may be scared away from ever even trying.

But this fearfully corrupted opportunism leaves us with patchwork solutions, stiffened by their internal tensions, and often verging on insensate to outside views.

Clenching is immobilizing.

Onto dukkha:

Dukkha is usually translated as suffering, which sort of works but misses important stuff. A more literal translation is 'a difficult emptiness.' Approaches, even quite effective ones, for dealing with the suffering of life were already in existence at the time of the Buddha. Both schools that preached constant absorption into pleasurable meditative states, and schools that preached a doctrine and practice of 'non-duality.' Both of these approaches survived, became mixed up with Buddhism, and today there are schools claiming to teach Buddhism which actually teach these methods. These methods do in fact decrease suffering, but they are only partial solutions. Both because they are reliant on maintenance of certain states and ways of being, and because while they deal with suffering caused by the immediate senses, you are still left with a more fundamental suffering related to feelings of emptiness or, Dukkha's other translation, 'worthlessness' and related feelings (nihilism etc. in the west). You've encountered this for yourself if you've experienced something cool during contemplative practice but then had a kind of 'so-what?' moment. The sense that this experience, while interesting and probably a temporary respite from your worries, hasn't actually addressed the core problem. People especially have this coming back from retreat. If this were just considered on its own, without the teaching of the antidote, this might be called worthlessness, that it seems like things are never satisfying and thus nothing has any value.

Perhaps I'd also call dukkha disappointment, or “lacking closure”. It's the buried cringing that happens when taṇhā and the rest fail to hold their ruse together. Which is inevitable! All models are wrong.

If you dream of drinking wine, in the morning you will weep.

- Zhuangzi, Ch. 2 (trans. Ziporyn)

Dukkha is the discord or bad vibes that result from a partial vibe collapse. Partial, because if the vibe fell away completely, if the amnesia were total, would an unpleasant contrast remain?

Dukkha leaves you hanging.

Why? Suppose taṇhā keeps happening even after the shattering invalidation of the vibe it had grasped and distilled through upādāna. We cling to the shattered pieces as if they would join back together if we squeezed enough. As if we could escape the finger trap by pulling hard enough. As if there’s no time to step back and learn or feel anything better right now.

A central insight is that dukkha really is inevitable, and that “good vibes only” is a farce. We impulsively push unpleasantness away. We try to filter it out of existence. But this only insinuates and prolongs the conditions for our suffering. Across countless moments, our impulses commit us to a hidden and vastening mess. And when we see a flash of the mess, our quick-patch solution may be to ignore and deny it altogether.

On the other hand, consider the practice of trauma work7: you stir up repressed bad vibes, then diffuse, dissolve, accept their contents back into yourself. This works very well sometimes. It kind of makes sense, if you think of dukkha as something that becomes memorized, and that the conditions for a partially collapsed vibe8 may be stored long-term. We generally try to avoid recalling such vibes, because it’s immediately unpleasant. If you’ve ever been suddenly struck by a cringey memory of your past, you know what this feels like. Trauma work says: stop dwelling in denial. Bring back that bad vibe with a vengeance. But don’t clench on it like usual. Don’t try to force a resolution until the pain compels you to hide it away again. Stop treating it as Other, as outside interference. Let the broken pieces dissolve back into mental potential, fertile with revenant agency.

If suffering were truly just coming in from the outside [link is mine] in thousands of different forms (i.e. the way things seem on cursory inspection) then we wouldn't have much hope of a single intervention helping us. Nor would we be confident in any such intervention since some new form of suffering can always show up. But if suffering is a result of something we're doing, then if we can figure out how to stop doing that, the suffering stops. Which we can confirm for ourselves in moment to moment experience.

In Buddhist psychology, dukkha is one of the “three marks of existence”. That is, it’s part of existing, and to think otherwise is delusion. No story, no saṅkhāra, no bullshit can spare us from any and all bad vibes. No model makes predictions perfect enough that it can avert any and all mismatches between conception and reality. By witnessing and accepting dukkha, we can escape the trap where we make bad vibes worse by impulsively trying to avert them.

The other two marks of existence are anicca and anattā, which are the mental principles we come to witness and accept as we free ourselves from two other delusions: nicca and attā.

Nicca is our tendency to believe that things could or should be maintained to our satisfaction. This is an identifiable mental event in how we reify an object or concept. Ignoring the very ephemeral nature of moment to moment experience in favor of only noticing those aspects which do occur as stable. Spotting it for yourself is very powerful. If this were just considered on its own without the teaching of the antidote it might be related to feelings of hopelessness. That there is no hope of maintaining the conditions that lead to things we like. Thus, the flow of positive and negative experiences are undependable, indefinite in duration, intensity, and frequency. That our hopes of forcing them to be stable with our mind will be in vain.

Nicca is a kind of reassurance that prevents the relaxation of taṇhā, and the unravelling of saṅkhāras. I think I've felt nicca in my self-soothing reaction to any perception that seems to threaten my enjoyment of a favourite or newfound word or vibe.

Atta is a little hard to translate, we can translate it as more like a verb or more like a noun (Pali is weird). If we see it more like a noun it might be translated as ‘essence’ and if we translate it like a verb it might be translated as ‘to take/have control/ownership of.’ Together we have the notion that if something has a real immutable character or ‘essence’ to it that we understand, then we can really control it and that this control won’t be subject to change.

If upādāna pursues simplified essences across the narrowing process of taṇhā, and nicca is a kind of compulsive moment of reassurance that coaxes us to follow through on the pursuit, then attā may be the moment that those essences are appropriated by a central coordinator of action, "the self". It’s the impulse to seize agency by wielding the essences: I’ve found this tool. Now it serves me! But is it really a tool at all? Does it serve? Or am I taking on technical debt, just as soon as the form of the essence seems sufficiently solid to manipulate, because I really like the idea of having a new tool? Stevens uses the example of Maggie’s fake steering wheel in the intro to The Simpsons, which brings her the illusion of control.9 We pretend useless things are tools because playing with them conjures an aura of doing something. Yes, maybe they were useful for a moment, once. Maybe they’ll be useful again sometime, too. And maybe I’m just pulling on a finger trap recreationally.

To put it cutely: taṇhā says “this”, and upādāna makes it “this”. Nicca says “this is fine”. Attā says “this is mine”.

By noticing that nicca and attā ultimately cause more pain than they solve, we may begin to want to escape them.

Anicca: no, you won’t always feel this way. Existence does not abide unchanging. The classic translation of anicca is “impermanence”, as in the freedom from the delusion of permanence. It’s the unclenched state we want to approach in mindfulness practice: not partaking of those hits of comfort that’d delay a vibe from its natural expiry, but giving way to whatever comes next.

And anattā:

Anatta is to point out the error in [attā’s] way of seeing things. The point is to notice the mental event that represents objects or concept as though they could or should be inherently or essentially controllable/ownable. If this were to just be taken on its own without the teaching of the antidote it might be called helplessness, that things are without the possibility of being controlled. We use the mind to falsely pretend we are more in control than we are.

We can look for essences or simplifications, and sometimes they may even be locally useful. But when we itch and we push to wield and to control them, we commit to a narrowed vision of causality, which before long will be revealed as a mere caricature of control.10 Anattā is refusing to partake of the impulse that grasps an essence in the active hand of attention, as if that essence must be a tool because I need to act.

Our brains11 keep something like a body schema, a model of our body in space that helps us to coordinate how we move.

The inner movements of our attention are kind of like the movements of invisible limbs. So, analogous to the body schema, we may also have an attention schema that helps coordinate our attention. An attention schema is meta-attentional: it attends to attention.

Now this is particularly speculative: nicca and attā seem to have a reflective or meta-attentional flavour. Taṇhā happens first and continually, and is always beginning to narrow my attention onto some thing, and the next thing, and the next. Sometimes attention lingers somewhere and taṇhā proceeds through upādāna, with nicca- and attā-type events acting to stabilize the process. Something like, it’s OK that I keep attending to this. It feels correct. And, it’s useful that my attention is narrowed like this. See how much everything’s under control?12 Maybe these nicca and attā events arise from contractions in the meta-attentional mechanism, like attractor states in the brain’s substrate for its attention schema. Evolution could easily favour such features, if they make the grasping movements of the “limbs of attention” more locally robust to disturbances.

Intelligent reflection is useful, but it’s also open-ended and distracting. It’s not enough just to narrow my attention onto some shortlist of movements or actions. There needs to be a way for an action to gain traction, to “move up the stack” and win me over, so I can commit to something actually executable, even while I might be considering multiple courses in parallel.

Do nicca and attā contribute to behavioural robustness also by “protecting” my narrowed attention from the “outside interference” of overthinking, which would distract me from faking it and forcing my way?13

How much of what I'm saying is bullshit, an evocation of an unrefined saṅkhāra? I haven’t meditated very much by meditator standards. A few hundred hours. Am I just reifying clumsily, because I haven’t had enough experience with the feelings themselves? Being picky about words, because I haven’t smoothed out what’s underneath?

Is there something of nicca in the urges I feel to keep using "clench", because I like its vibe? Something of attā in my residual sense of ownership of the word, and my violated feelings of originality?

Am I still clenching on "clenching"?

At least a little, I think. But I understand that my understanding and my objects must be flawed. Anyway “clenching” is the best word I’ve found for this thing I’ve seen. Maybe I’ll learn of a better one. In the meantime, why should I be surprised that others have come to the same conclusion?

We can't go entirely without better models and better words. The Buddha certainly didn't go without them. Totally-falling-apart-forever contradicts the spirit.

The answer to muscle spasms isn't never contract a muscle again.

The answer to taṇhā isn't abhor the slightest bit of attention.

Taṇhā happens easily, automatically, continually. And it’s the springboard for the whole twisted circus. To live freely and gracefully, we might want to know its opposite.

Nibbana is generally thought of as an exalted state of being that is free of all suffering, all desire, etc. etc. It […] leads to people assuming that Buddhism is wireheading. The best translation of Nibbana (for the purposes of practice) IMO is ‘cooling down.’ If we think of the above strategies as a sort of tensing, a sort of effortful exercise, a sort of heating up if you will, then we can contrast it with untensing, non-efforting, cooling down and relaxing. The simplest way to think of this is that Nibanna is the opposite of Tanha. Often translated as the mind ‘inclining towards relinquishment’ (of that which was grasped). The nature of this experience is relief. And here it means not only relief from the particular stimulus that was stressing us out, but the (normally experienced temporarily) relief from compulsive grasping, the relief from wanting things to be other than they are, relief from the belief seemingly pressing down on us that we need to act just for things to be okay. A kind of happiness that comes from a halting of believing that we need to get happiness by arranging things to match up with mental projections.

So nibbāna is “keeping unclenched”. It looks like not immediately and impulsively pulling on the next finger trap you encounter, by assuming you’ve already got it figured out (I know how to pull!) only to ignorantly commit yourself to even more mess and stress and ignorance.14

Importantly, to escape the basic clenching of taṇhā, I need to escape the meta-attentional clenching that stabilizes the consequences of taṇhā through upādāna. Thus nibbāna does not arrive without anicca or anattā.

Now what if I actually want to be stubborn, to “[arrange] things to match up with mental projections”? Won’t nibbāna make me passive? And won’t I stop noticing evil in the world, if I’m able to find “a kind of happiness” regardless?

Well, progress is relentless. Stress looms. Depression is baffling. Even when mental health issues are skill issues, they are also visceral, historical, dynamical issues. Traps exist! I do not have to become permanently dumber for a series of salient events to narrow my vision, until in some sense I am merely churning at pre-ordained tasks… perhaps while the outside view grows hazy. Insensitivity rationalizes itself to insensitivity. Opportunity lies forgotten.

Relinquishing the bleak urgings of stress doesn’t strand us without recourse to any action. It releases us into the potential for fluid action.15 It is not necessary to become a hermit in nibbāna, to usefully incline towards it.

People in the left tail of neuroticism become more functional [with practice], not less. Their fear of becoming even less functional is a stability defending meta-aversion to modifying things they think are keeping them alive. In the high-threat mode, everything is flagged as a potential threat, including exiting the high-threat mode.

- Romeo Stevens, “Core Transformation”

For most of history, death stalked close behind relaxation. The fear of unclenching still lurks in our reflexes, and in the other precedents of our biological bodies.

But letting go is a prerequisite for growth and agency!

Of course, we don’t want to abandon everything. And we cannot leave behind all saṅkhāras, all mental framings or dispositions. But we can untangle and refine them to be less crude, compulsive, and fetishistic.

We'd like to find ourselves more optioned and free. Less constrained, convulsive, and clenchy. The miserable don't start their journey wanting to die. But when the trap seems to grow inescapable…

Romeo wraps up his post with an excellent synopsis of the whole ordeal:

By default, the mind becomes stuck to mental representations that have more to do with our desires than how things really are. This leads to aversive experiences of emptiness, hopelessness, worthlessness, helplessness when we bump into evidence about how confused we are. We come up with plans for avoiding these experiences, but these plans don’t really work, leading us to repeatedly encounter flashes of the undesired experiences. Our response is to try to push on the plans even harder, which doesn’t work. But once we get wise to this process we can incline in the opposite direction, pushing less hard on experience. The relief from doing this wakes us up to the idea that we’ve been fueling the above vicious cycle and live in a house built from these sorts of knots of confusion. Instead of trying to hold the house together with constant maintenance while simultaneously trying to find the exact right decorations, we start tearing down the house. We discover that the very idea that we needed an unchanging, beautiful house that definitely belongs to us was just another of the confused knots. House building, maintenance, and dwelling becomes just another human activity that can be engaged with or not as is convenient. As these activities were previously taking up huge amounts of our attention and resources, we find ourselves much more relaxed and able to enjoy things. Because others still live within the paradigm of seeing everything as related to houses, they are inclined to perceive you as living in a shiny spiritual house, and try to figure out which sets of maintenance activities and decorations will grant them these ‘spiritual’ advantages.

My use of "clenching" arose in the context of literal muscles, of motor control and reaching experiments. Soon, I realized its value as a metaphor. Its meaning was already so close to being shared between us in the right ways! It makes sense that others thought to use it, too.16

Now I'm no longer so afraid of being unoriginal.

I’m just here learning to relate.

Do not be trapped by the need to achieve anything. This way, you achieve everything.

- Dune Messiah

In the next part: brains and dynamical systems.

Vipassana training consists of both working on mindfulness, careful observation of what enters the mind, and concentration, holding attention on some object. These capacities are seen as somehow opposed but also complimentary to each other, and it’s important that they develop in tandem, that one doesn’t lag behind. This roughly corresponds with predictive coding’s idea that perception consists of a negotiation between ‘bottom-up’ raw sense data, and ‘top-down’ models of reality which provide context for the raw data and fill in any missing gaps. Likewise, predictive coding suggests that if one of these processes is much stronger than the other, problems occur– e.g. if bottom-up sense-data dominates we’ll experience noise and confusion as we struggle to sort things out, and if top-down predictions dominate we’ll experience hallucinations, stories not connected to facts. My intuition is that modern life’s focus on planning and abstract thought tends to make us a little ‘top-heavy’ here, so at least as beginners, most people would probably benefit more from mindfulness meditation

To survive, to thrive, we need to predict and take actions in the world. Our subjective experience lives in our heads, in a simulation of the world that contacts the world only indirectly, through the vast details of our senses—say, light falling on parts of our eyes, or the stretching of receptors in our skin. But raw sensation isn’t delivered with an interpretation attached, and as we grow from infancy our brains have to learn for themselves how to pick apart the details and start to perceive objects and causes.

I’m staring at an apple I’m holding. What this apple “looks like” in terms of raw data is a bunch of changes in the firing of certain nerves in my skin due the apple contacting my fingers in certain places, and of certain cells in my retina due to light interacting with the apple and my eye.1 But it doesn’t feel so technical to me. I just see “an apple”. I’m an adult, and I’ve already internalized how to meaningfully correlate all those little bits of sensation. Now the processes of my body (brain) can intuitively anticipate all the apple-related details. Those details might’ve made an incoherent soup once, but now the object apple is part of my world simulation.2

Predictive coding says that my nervous system forms a kind of hierarchy that builds up my model of the world, based on raw information3. Sensory receptors at the “bottom” of the hierarchy feed their raw data “upward”. Higher up, increasingly abstract4 models push their predictions about sensations back “down”. The downward-flowing predictions “prime” the lower levels of the hierarchy to await the sensations that should show up if those predictions are correct.5

If my model of the world is excellent, the predictions it sends down will match up with the incoming sensations, and the two will cancel out nearer the bottom of the hierarchy. But if something about my model is wrong or incomplete, there’ll be a mismatch between prediction and sensation. This prediction error will percolate up the hierarchy until it gets corrected, such as

• by making me realize I misjudged the context, so I switch to some known model that’s more appropriate for the context I’m in6;

• or by incepting a brand new model, as must often happen in children;

• or by reaching the top of the hierarchy and fizzling out, undecidable;

• or by being blocked by a stubborn prediction.

In any case, prediction errors are disturbances or surprises that we can avert by making our predictions better. When we can’t avert them, they work their way inward by a process of negotiation, through escalating levels of predictive models.

If the negotiation is well-balanced then errors will teach us—that is, improve our models—without being excessively distracting. But if a certain prediction is too strong/stubborn/robust, it may keep us insensitive to some of our actual sensations, overriding them with hallucinations or delusions. And when our predictions are not strong enough, we may be distracted by constant and pointless negotiations over the meaning of very slight changes in sensation.

Back to meditation: when we lean in to a particular state in concentration practice, maybe we stabilize a top-down prediction to the exclusion of bottom-up disturbances. And if some concentration practices are meta-unclenchy, maybe we choose a type of top-down story that "cleans up” or “makes room” for bottom-up signals7. I can fill myself with the story I’m here to see, and to learn, and to love, and it can protect me from other stories that might keep me burdened and insensitive.

On the other hand, in insight practice perhaps we learn to more directly quiet our top-down predictions (stories, identities), relax our stubborn blockages, and ease entry of details into awareness.

But no story is meta-unclenchy if I spend too much of my time with it, to the exclusion of others. And too little stubbornness renders me ineffective.

The top-down versus bottom-up distinction can help reframe the tradeoff between structure-forcing and structure-fitting: forcing is the influence of prediction/action signals that proceed outward or downward, whereas fitting is the influence of evidence/sensory signals that proceed inward or upward. As infants we start with very little in the way of top-down models, so the balance of information flow is skewed toward the inward, structure-fitting direction. This fits (ha) with children being balanced more towards exploration, as they have fewer expectations to guide them yet. They have no choice but to act more broadly—and incoherently—to sample from the world and acquire structure, which later they can stubbornly exploit as adults.

Finally, I want to point out that Johnson characterized vipassana as consisting of both concentration and mindfulness components, while Romeo Stevens categorized vipassana and mindfulness under insight practices. Perhaps this shows the fuzziness of these categories.

I don’t know about the history of vipassana, but let’s say we recklessly invent a new practice superthink that’s purely about insight. It’s practiced by actual people—before long, some of them will notice the over-insighting imbalance, and import some concentration practice into their routine. But maybe they’ll keep calling the whole package superthink, because to them the term has become a social signifier more than an indication of practical content.8 So does superthink refer to an insight practice, or a mixture of insight and concentration practices?

This is just to point out that there’s a difference between the content of a practice and whatever name you give it, especially as that name becomes shared by other people—which is the most exciting and useful and dangerous part of naming!

After reading the paper on robust reaching, I began to use "clenching" as a metaphor: not only to speak of lingering tensions of the muscles, but also tensions of the mind—when I lean in and fake it a little too long and hard. And also to refer to situations where, looking in from the outside, a person or a group appears to be trapped1: doubling down, neglecting what might free them, often scarcely even able to look for it.

So, how can we unclench? How can we see from the outside?

Most of us are falling in most of the time. We wake up and we’re immediately pulled onto the tracks, onto a train of feelings for the people and things around us. Our eyelids lift, our ears alert, and we’re carried forward by sensations, one after another, on automatic. Feelings, into motions, into sensations, into feelings—I want, so I move, so I see, so I want, so…

We go back to sleep filled with expectation—an inertia, more or less subliminal—for the next episode2.

It’s easy to forget that comfort is not a literal property of a lover, nor satisfaction a property of chocolate, nor pleasure a property of porn. Comfort, satisfaction, and pleasure are feelings we have. They happen inside our heads, where our subjectivity actually lives, where a simulation of the world blazes along, fed by our senses. We easily confuse the simulation with the outside world it simulates—the map with the territory. Because it’s convenient. Because it would be a disorienting burden on our limited attention, to be always and viscerally reminded that we don’t experience the rest of the world directly. Instead, evolution has seen that it’s robust for us to lean right in to whichever simplifications bring us those feelings. As much as we can get away with it, anyway. So we keep falling, keep pushing buttons, keep forgetting they’re buttons. And before long this fails, inevitably, and leaves us shaken and disappointed.3 The world is always one step beyond button-mashing story-logic. My lover is a real person. Chocolate isn’t a balanced diet. Porn isn’t living sex.

How can you become wise to these property illusions? How can you see from the outside? By practicing not falling in, like any other skill. By taking regular time to intentionally redirect attention back upon itself: to learn what it really means to give attention, and to stop giving attention. To make room, and witness as it fills with something new.

With practice, we can learn to change how attention and feelings move, for the better.4

I can still keep the patterns I want to keep—still feel comfort from people, and satisfaction from food. Of course I don’t want to abandon everything, to jettison all meaning and enjoyment and success. I want to notice a little sooner when willing attachment starts to turn to willful captivity. To see beyond the familiar playground-prison of my room, my home, my neighbourhood, my school, my workplace… To learn what other movements could be possible.5 To be ready to act (or not) at the first spark of evidence or opportunity. To be empowered, and walk in a wider world. And to find out just how deeply and smoothly and vastly my feelings can work.

What is meditation? Maybe, the practice of not just falling in. But the meditative practices that people actually follow are much more varied and messy than that little summary suggests.

Buddhists and meditation nerds seem to refer to pretty much any intentional activity that’s relevant to personal or spiritual development as practice. I want to take some time to share my current thoughts about practice, and how they could relate to the other things we’ve discussed in this series.6

Romeo Stevens suggests three types of practice:

1. Concentration practices are about intensifying certain states of mind, and unifying attention. These states include metta, or feelings of benevolence towards others, as emphasized by “loving-kindness meditation”. They also include absorption in the jhānas, which are states of detached bliss—mental euphoria, bodily pleasure, and ultimately equanimity—that can be reached when we are freed or isolated enough from familiar objects, and the distractions of falling-in. On the surface the jhānas might seem kind of dangerous. We’ll return to that that shortly.

2. Insight practices are about picking apart states of your mind—contemplating aspects of them, and witnessing patterns and conflicts. An example is vipassanā, which is based in the contemplation of the fleeting nature of experience.

3. Integration practices are about relating and connecting your insights to your entire life. Many of these practices don’t look like “meditation”. For example: seeing therapists, having conversations, writing online, changing habits (including meditation practices)…

The three kinds of practice are mutually strengthening:

1. Better concentration creates the conditions for insight. Mindfulness practitioners emphasize concentrating on the present moment to enable a clear, non-judgmental awareness of mental events. In vipassanā, concentrating on the breath is a basis for insight into the transitory. And the concentrated positive vibes of metta and the jhanas are powerful vantages from which to contemplate goodness, love, and happiness, from above the clouds of petty inclinations and grievances.

2. Insight provides material for integration practice. Knowledge of our conflicting patterns—mismatched predictions, “cognitive dissonance”—helps us to make good decisions about when to change, and when to stay stubborn.

3. The more integrated we become, the easier it is to concentrate, which creates even better conditions for insight, and so forth.

Sounds pretty good, right?

We are what we pretend to be, so we must be careful what we pretend to be.

- Kurt Vonnegut

Like pretty much everything, meditation is not completely safe. Some practitioners claim emphatically that it’s dangerous, at least once you’ve reached the scale of hundreds or thousands of hours of practice.

As we grow and learn across our lives, our minds build up structure. Our mental movements come to weave among each other like so many threads. As Mark Lippmann7 has pointed out, this process of accumulation of structure is like the concept of technical debt in software engineering, “the implied cost of future reworking required when choosing an easy but limited solution instead of a better approach that could take more time”:

When a mind is really surprised, or things are happening too fast, or something is just too hard, or a mind enacts ingrained bad habits, in all these cases a mind takes on technical debt in order to keep dealing with the world in real time. The more technical debt a mind has, the harder it is for that mind to solve problems moving forward, so technical debt begets more technical debt.

To put this in terms we’ve already used: when we cannot predict what will happen, when we have no time to try out new models or respond precisely to disturbances, we find simplifications to lean into. We fake it ‘til we make it. But our fakery will inevitably be inappropriate to our changing situation—like pulling on a finger trap isn’t the way to escape it. We ought to replace the old simplification with something wiser, which would work in both the old situation and the new. But before we think to do that, we’ve often gone through other trials and layered other fakery on top of the original fakery.8 This makes a larger mess that is even harder to backtrack.

In the short term, it’s easier to add just one more patch to deal with the newest challenge, than to untangle the whole mess and rebuild things more sensibly. And how can I even start to take an outside view of what “more sensibly” could mean, when all the ways I can experience the world are already embedded in a monstrous patchwork that no longer even registers as a mess to me?

Let’s say a person has 10 different mental constructs (beliefs) that they use to make sense of their situation and employ strategies for getting their needs met. That these beliefs strongly resist updating in light of new evidence makes perfect sense in the context in which any one of the beliefs changing makes the whole structure worse than before. The idea that there is a much better way of being somewhere far away in mind architecture space requires quite a bit of faith. Or, as weird sun twitter put it: “‘That way lies madness’ He said, pointing in all possible directions from the center of the attractor.” So the beliefs resist change by virtue of being load bearing, by having had lots of important structures built on top of them. To change them would feel like invalidating the suffering that one (or others) underwent to attain meaningful outcomes within that framework. 

- Romeo Stevens, “(mis)Translating the Buddha”

When I finally try to untangle my mess, and gain insight to loosen some tension or untie some knot, I also alter the tensions in the surrounding weave. Unexpected things may happen. Sometimes a load-bearing beam9 or a section of scaffolding collapses, and I have to put in the work to restructure, or risk a larger collapse.

Dissolving the constructs that lead to you prioritizing exercise, eating well, and sleeping should be seen as dissolution of the goodness of the means, not the ends. e.g. you were using fear based motivation to keep you exercising, which you subsequently saw through. This doesn't mean exercise was bad, it means your method was bad and you should find an upgraded one.

- Romeo Stevens, “Why do contemplative practitioners make so many metaphysical claims?”

The jhānas are worth dwelling on for a moment. On the surface they seem kind of concerning. You’re saying if I just step back from all the other stuff I do with my life, I can manufacture extremely pure and potent positive vibes for myself? Yes, pretty much. Isn’t that just wireheading? No, not quite. Why not? Won’t I get addicted? Well, in principle these vibes can be abused, but according to Romeo Stevens: “every teacher who I trust reports that this is pretty rare, and much more likely to happen to people who practice without the feedback of a teacher and community”.

I’ve not reached the jhānas yet, but I’m also not particularly worried about them. Why?

Because of the nature of falling-in, and my tendency to treat feelings as essences of whichever familiar objects or situations “give” me the feelings, there’s a tyranny those objects10 can hold over my mind. The jhānas make it plain that this is an illusion. The pure feelings were in me all along! Freed from the need to pursue any particular object, I can learn about the feeling itself, unconditionally. In doing so, I’ll tend to grow further beyond the compulsive grabby quest for pleasure, and into the peace of the realization that enjoyment can be everywhere and nowhere—and that nothing can hold me hostage to it. This isn’t the usual kind of easy trick I could abuse to get a hit of feeling, only later to be left wanting.

The jhānas take patience. I expect them to deepen and purify my appreciation and capacity for joy. From the vantage of the jhānas I expect to be more able to cultivate insight, clarity, and the motivation to share in the freedom I find. But I can’t share this freedom simply by dwelling in positive vibes. I have to stand up and head back into the world—there are still problems to solve, and I don’t want to ignore them for feelings of boundless rapture and tranquility.

What if vulnerable or corruptible people find the jhānas? Oh… but what if we don’t find the jhānas?

Life is everywhere life, life in ourselves, not in what is outside us.

- Dostoevsky (letter to his brother, on the day of his mock execution)

How might we squeeze the different types of practice into the categories we’ve already seen in this series? This might not be absolutely correct, but maybe it will yield some insight.

Concentration practice kind of looks like leaning in to certain mental states, and making them more stable or robust. Practitioners would tend to say this shouldn’t look effortful—fine, okay.

Importantly, this doesn’t usually mean leaning into just any mental states, like Taylor Swift earworms, Sunday brunch FOMO, or fantasy league planning. These would all count as fallings-in. They don’t fit our purpose. We want to practice entering states that it might be useful to enter again later. But perhaps more importantly, we want practices that are meta-unclenchy.

… what? Well, practicing concentration means flexing certain states, and in principle this could turn into pull-too-hard-on-the-finger-trap style clenching. But we can choose a type of narrowing that tends to lead back into a widening. This is kind of like cleaning your room11: the act of cleaning itself is narrowing, and might be stressful at first, but the context it creates is freeing. Cleaning can be a bit clenchy, but it’s also meta-unclenchy.

In contrast to concentration practice, insight practice is more about directly leaning out from any particular mental state that could inhibit the emergence of patterns to be observed, noted, or contemplated. Just unclench, and let things happen. Don’t contract too long or hard on any of them, but watch how they move and disturb and relate to each other. Practicing insight means fishing for the position of our thoughts in their larger context, and for those “Aha!” moments that reveal the patterns we’re trapped in. To continue with an imperfect and too-small metaphor: insight is like looking at your room, and letting your imagination run wild about how you’d rearrange the furniture12.

Concentration helps with insight because there’s much more clarity and pleasure in reimagining and disrupting the order of things, when they aren’t covered in dirt and cobwebs. Cleaning and rearranging are synergistic.

So there is a kind of duality between concentration practices and insight practices. One seeks to purify our attention, the other to unsettle it. And kind of like there’s a trade-off between exploit versus explore, and a trade-off between structure-forcing stubbornness versus structure-fitting science, there is a balance between concentration and insight.

Stevens mentions that too much concentration practice can cause “stuck” feelings. Well, concentration practices exploit certain focused states, which creates room for exploration. But if we don’t eventually use that space for exploration… that’s like always cleaning your room, but never rearranging anything. Living in the same sterile space, forever.

And too much insight practice can “[stir] up trauma and not [deal] with it”, which is what happens when we explore too much without creating a context in which we are happy to exploit the results of our search. This is kind of like pacing around your dirty room every week, thinking of how to rearrange it. Mostly you just eat dust and become discouraged.

Should we expect things to be quite so binary? Healthy behaviour often isn’t. Still, specific meditative practices may be kind of lopsided, at least in the beginning.

Most meditation systems are depth-first: you plunge, deeply, into one area of your mind—by, say, learning how to concentrate on your breath with heroic clarity—and then you take the mental machinery thus developed and try to iron out your life with it.

- Sasha Chapin

A role of integration practices is to monitor the usefulness and moderate the balance of the specific meditative practices we pick up, by keeping them in contact with the rest of our lives. Of course, different sets of practices will work better for different people with different problems at different times. We should want to choose practices that form tight feedback loops and bring noticeable improvements to our problems. If we can’t see this happening, we leave those practices behind, or try to balance them better.

No surprise, but too much integration practice can also be problematic, leading to “endless analysis and working on oneself but never really getting to big shifts”. On its own, non-meditative integration doesn’t provide enough intensity or clarity to overcome our mundane inertia—it happily settles into just another activity we can fall into.

We can reflectively apply our insight about property illusions and not-just-falling-in, to our meditative practices themselves13: Useful change is not a literal property of a meditative practice, but something a practice might accelerate. “Meditation” isn’t just a button I can push to get a desirable result. My practice is built up of little behaviours—sit there, close eyes, breathe deep, observe thoughts—that I can clench on and over-emphasize just like any other behaviours, even when the immediate consequences of my practice might seem especially calming and virtuous. If I neglect to keep making this reflection—what am I really using this for, what direction am I moving, and is it staying relevant?—I might stagnate, or worse, fixate on the delightful power of a newfound accelerator until I accelerate right into a wall, or off a cliff.14 In less extreme terms: a practice can become a fakery like any other. There’s no guarantee it won’t worsen my mess, my technical debt. I should keep reflecting.

[Mark Lippmann’s meditation system15] is breadth-first: he wants you to mindfully do a whole bunch of different little things—small intentional changes in your thinking, exploration of different states, reflection, reverie—and thus take a million little steps towards the goal of understanding the way you function.

- Sasha Chapin

But meditation […] will give you some extra tools and perceptual clarity around negative patterns. It won't magically eliminate the work you have to do to tinker with those patterns and implement better patterns. It does have a tendency to make that work feel less aversive.

- Romeo Stevens, “Orientation on the Contemplative Path”

The usual “social animal” advice applies: keeping up with friends is a superior source of motivation to continue with practice, and to do it well. The most important antidote to meditation risks may be participation in a community of teachers, “noble friends”, and therapists. It’s not difficult to set a trap for yourself that’s all-but-invisible to you alone. A small conversation with an honest and perceptive person can be a powerful seed for insights. Let’s cover each other’s blind spots.

Not all community is equal. There’s danger in applying powerful accelerants of self-alteration when the feedback you get is overconfident (everything will be better if you do X, it worked for me!) or isn’t really feedback at all (you do you!!). It’s important that the friends and communities you relate to in your practice should value pragmatism and skepticism at least as much as metaphysics and spirituality.

In my personal opinion, shooting openness sky high without a balancing increase in healthy skepticism reliably lands you in whacky belief town. Most practitioners are not starting with solid prerequisites about map-territory distinctions, probabilistic over binary reasoning, and strong ability to demarcate is and ought (positive and normative) claims. Most schools are not, in my experience, emphasizing the very skeptical nature of the Buddha's inquiry into his own mental processing.

- Romeo Stevens, “Why do contemplative practitioners make so many metaphysical claims?”

It’s easy to forget that our feelings are intrinsic to us, and not the things and people that are most familiar to us. So we keep falling into our lives, and into a kind of private tyranny.

Confronted with urgent or difficult problems, we find good-enough solutions even if they are patches atop patches. This makes a mess of our minds, until the mess becomes familiar and forgotten.

Meditation is about spending some time not doing these things—mentally “cleaning your room”, and finding a clarity in which it’s a pleasure to witness how things are arranged, and then to imagine rearranging them for the better.

We might consider three categories of practice: concentration, insight, and integration. Too much of any one of them could be problematic.

We want to actually solve problems. Sometimes, meditative practices can accelerate that. But what are we accelerating towards? It’s crucial to continually, reflectively judge our choice and balance of practices, and the direction we are heading. Being part of a community of skeptical practitioners is invaluable for covering blind spots and keeping up motivation.

I haven’t described how to engage in specific practices in this post, which is only a survey of ideas. I’m no expert, and you can find much more detail elsewhere, e.g. from Shinzen Young.

In the next part: the language of Buddhism.

Doubling down is robust. By leaning harder into our favourite actions, we minimize the effect of outside interference whose exact form we can’t predict. But how do we know our favourite action—our best model—is really what it claims to be? How do I know “I can’t predict”, or find a better way?

To escape a finger trap, a child ultimately needs to discover a new pattern, like push inward and use my thumbs to hold the trap while I pull my fingers out. Unlike someone making separate movements that only last a second or so each, affected by unpredictable disturbances that cannot be entirely compensated, the child is immersed in a continual situation. We know they have time to recognize their confusion, experiment with movements, and learn the solution—for good.

What counts as "outside interference" to a child pulling on a trap? Well, anything that interferes with pulling.1 Any shift in their attention to alternative approaches. Any experiment involving sustained pushing, that could reveal the winning strategy.

The narrowing of attention and action may be exclusionary of alternatives.2 We only have so much attention, and distractions are dangerous where tigers are concerned. "Pull harder" bets against the finger trap having an accessible and efficient solution. I don’t have time for that right now.

Taking the time to experiment and learn would save the child the energy and futility of pulling harder for who-knows-how-much-longer. But how do they know that there is even a solution that needs learning? How long will the experiment take? What evidence have they seen of an alternative, until they've tried it and it works?3 Maybe pull-harder will keep working like it always has. Our hindsight tells us it's improbable that this kid will defeat the trap by pulling harder. But at the moment, they don't seem to believe it.

If the kid's IQ is high enough, will they simply see past this obstacle immediately? Perhaps. Then they'd move on to the next challenge, and the next, until the metaphorical traps become subtle enough that they can no longer merely peek over the horizon at the solution. Wherever such an impasse is met—a pressing uncertainty without a ready remedy—even a genius kid can only lean harder into what they’ve already got.

And maybe they're having too much fun being trapped.

Does the prophet see the future or does he see a line of weakness, a fault or cleavage that he may shatter with words or decisions as a diamond-cutter shatters his gem with a blow of a knife?
- Dune

In the theory of active inference, prediction and action are not entirely separable events in a sequence, but dual views on the same thing.4 Different words for changes in a living process.

I predict that food should be in my stomach, but my senses tell me my stomach is empty. This mismatched prediction is tension in the process that is my body. Action is the tension resolving itself—the result of a force applied across the process. I feel the tension of hunger, and my hand is driven to reach for an apple, which I start to eat5. I'm still predicting there'll be food in my stomach, and now I sense there is. No more mismatch.

Predictions are about the world. Actions alter the world. We can act to make the world more predictable. We move, and in moving we create context. We enact structure.

Legislation and government aren’t primordial parts of physics, or even psychology.6 They are prophecies that people invoke, until they are more than prophecies. We predict that something like them should exist. They begin to come into existence. Being subject to them, our lives become more predictable. We depend on that predictability, and invest more in the structures. They grow.

Importantly: we are part of the world that changes. Our predictions are also about us, and about our predictions. We act internally to make our selves predictable.7 Our identities crystallize out of fluid informality. Adopting the context of an identity, we can force ourselves to be understood. Legible.

Sometimes, we have little power to change the world by leaning into our actions. But sometimes “I can’t change” is an illusion—a finger trap, a story we keep pulling on. How can we see past that?

Sometimes we do little more than observe. But we’re not passive observers: in observing, we act. We orient and reorient our bodies to seek out evidence. We point our eyes toward what we want to see, to force things to make sense. We may look away, when the tension seems unsolvable.

He was comfortable with this world, and he knew it. He knew it, and he liked it.
- Succession

We must predict as well as we can, or risk being overwhelmed and destroyed by forces we’ve not predicted.

We can improve our predictions in two directions: by acting to change the world to look more like our predictions, or by acting to change ourselves (our predictions) to look more like the world. On the one hand we can force our way, on the other we can be scientists.8

Too much science won’t do. In order to change the world, you cannot be too willing to change yourself. If you see that the world is one way, but you want it to be some other way, your prediction has to be stubborn. You’ve got to maintain the tension, the mismatch inside you between is and ought, so that it will drive your actions to change the is to be more like the ought. If you’re too much the scientist, you’ll eliminate the mismatch by simply acting internally to overwrite your ought with the is. I’ve seen the way things are. I’ve gotten used to it. (I’ve forgotten my dreams.)

Too much stubbornness won’t do, either. The world isn’t a blank slate for the inception of prophecies. There are actual constraints—stuff out there to be learned, stuff not easily changed. And who isn’t thankful for science? But it took us so, so long to arrive where we are. We’ve had had a similar biological capacity for intelligence for many thousands of years, right? We’ve spent almost all of our time leaning into what we already had.

Weirdly, we can become stubborn about the consequences of not being stubborn enough. I’ve wanted it to be this way all along, of course. And many literal scientists are clearly driven to their investigations by the stubborn urge to solve some problem or other. The two directions for improving predictions don’t correspond to two distinct identities—people who are entirely scientists, or people who are entirely stubborn. We’re all some mix. The two are entangled. They deeply reflect and layer upon each other.

Some kinds of stubbornness are more “built-in”, and hard to change no matter how stubbornly we think they ought to change. We can’t decide to not feel hungry, though we can learn to ignore the tension of hunger, and out-stubborn it. Maybe with enough time, stubbornness, and science we’ll alter our biology to eliminate the very feeling of hunger. Would we remember to eat then, or would we simply starve? Those rare individuals who are born unable to feel pain tend to die young. They lack that stubborn, driving tension that makes us act to save ourselves in the face of injury or illness.

Stubbornness versus openness gives us something like a trade-off between structure-forcing and structure-fitting.9 There is structure “out there” in the world, and structure “in here” in our bodies—which are also part of the world. We might prefer to say process instead of structure, but in any case the trade-off asks: How does the “in here” change the “out there”, and vice versa? How does evidence—information from “out there”—change us, and how do we resist being changed by it?

What’s the proper balance between stubbornness and openness?

We can re-frame a reach-harder/pull-harder/structure-forcing/stubborn strategy as predict harder that my predictions are true, which just means predict harder.10

I intensify my prediction—so, my action.

I escape this trap!—I pull harder.

I get my toy!—I reach harder for it11. When I go-harder, my wish is a little more likely to come true. But the world can also tell me no, not this time.

Maybe if I don't reach harder, predict harder, the other kid gets the toy. But maybe they do anyway. And maybe I’ve been predicting so hard that I keep on predicting I’ll get it, after it’s no longer possible.

Maybe I should've paid attention to something else. Seen something new. Escaped some trap.

Maybe learning comes too late.

Children are awesome explorers. Most of them will solve the literal finger trap before long.

In a very recent study, we gave participants a reinforcement learning task in which they could sequentially choose whether to place blocks on a machine. We told them that some blocks would lead to rewards and others to costs but did not reveal what differentiated the two. The actual rule was two-dimensional (e.g. black striped blocks were costly but white striped blocks or black spotted blocks led to rewards). After one negative trial, adults quickly assumed the most obvious rule, that a single feature differentiated the blocks (e.g. all black blocks were costly), and so avoided all the blocks with that feature. But this meant that they never received evidence that showed that the actual rule was more complex and so failed to learn the correct rule. As in other studies, they fell into a ‘learning trap’. Preschoolers, in contrast, continued to try all the blocks on the machine, and so learned the rule correctly.
- Alison Gopnik, “Childhood as a solution to explore–exploit tensions”

It helps that for the trickier problems which require a lot of experience, kids can have adults around to speed them along. Adults have already explored plenty when they were kids themselves, and imitation and language are powerful ways of sharing knowledge.12

Are there any super-adults around here that could speed us normal adults along? Many of us wouldn't trust them anyway. They're probably just trying to get me to stop pulling so they can steal my finger trap.

As adults, we become exploitative of the things we’ve already learned. Our lives can become tyrannically finessed. Pretend becomes obvious, tacit. We keep exploring, but our exploration—and by our designs, even our children’s—is bounded by our exploits. Our curricula, rules, governments...

Our best ways of doing things are effective because we lean into them, and force them into the world. But the more effective they are, the larger they become, the more exploitable and popular and normal… the less incentive to see beyond them. What was once mere prophecy can become a prison.

What’s innovation that isn’t marketable? If nobody has use enough to pay for it, who cares? Except: the superorganism of all of humanity has its own traps, which constrain the nature of the markets—the arenas we all play in. And how can we see past that? As individuals, we probably can’t, at least not fully and directly. It’s too big.

So how can we speed along our world-spanning child of a civilization?

We all live in intricate nests of mental and social bonds. Which of them are traps? Past some horizon, we cannot tell. Intelligence is whatever pushes the horizon further—gives us the vision and the option to escape. But do we, then?

Don’t we13 too often tend to pull stubbornly, coordinate stiffly, prolong history, mock intelligence, even though we’re probably dwelling just beyond a long-neglected hill, with who knows what freedom waiting in the next valley?

Too cynical? Well, as much as cynicism may be a trap… watch me escape!

Fake it ‘til you make it: Force your predictions on the world.
Be a scientist: Let the world force itself on your predictions.
These aren’t separate options, but deeply entwined.

Sometimes we fake it too hard. How do we know before it’s too late?

In the next part: meditation and Buddhism.