(Also thousands of hours of meditation). I didn't get any of this. But, somehow, it seems meaningful. I did get some of it.
Anyway as a long-term meditator myself. There is definitely something 'here', which is also truly profound (and very mundane). For me the profound part of it is that 'experience' becomes intrinsically meaning-laded, adventurous, deep. There is also a sense of 'going somewhere' that wasn't there before. Like, wow, you can change your mind in fundamental ways; that is fun. And like, the quality of your experience can become truly better and better over time. Before my experience was one of 'make more money, more status, more things, more EA status'. But somehow none of that mattered at all.
And it doesn't, none of that can ever solve the issues meditation is trying to solve. And yet, also, getting all of those things 'more money, more status, more EA status, more beautiful people in your life', is kinda the point of meditation. Or, it should lead there, or, it should lead to a place where that doesn't matter.
Meditation is something I would both recommend as one of the most important thing a human being can do. And also something as most people shouldn't do.
The reason is maybe explained in the writing of this post and my own writing. The rest of society has an entirely different way of 'being' than you do, and it becomes very hard to express yourself. It becomes much harder to fit in and to do 'normal' things. And you often naturally find yourself on the edges of society because of that (for good, or for ill). And this can be stressful and painful.
It takes a tremendous amount of 'extra cognition' to deal with this 'awake' mind within society (whatever that is).
And most people don't have that extra slack. They will already be overwhelmed by the results of meditation; let alone trying to have a job, and all those other things. And then this can naturally lead to actually being way worse off (in terms of like your experience of consciousness, and like material/social things) than when you started meditating. But, also, I do believe that eventually most meditators do find those things again, but in a very different way than before.
For what it's worth, I cannot confidently distinguish this post from GPT-3 output. Which is to say, each sentence kind of makes sense, but then I get to the end and go "...what was that trying to say?"
In general I do want LW to be a place where people can brainstorm unfinished ideas at each other, but given the context that, as you mention, "meditation is a bit contentious in parts of the LessWrong community, out of concern for individuals and the wider community", I feel kinda icky about this particular post. Especially because "sounding more like GPT-3" is exactly the kind of thing that I am worried prolonged meditation might do to people.
For what it's worth, as someone with a lot of meditation experience and a longstanding interest in the topic, I didn't get a GPT-3 vibe at all. To me, the whole thing registered as meaningful communication on a poorly-understood topic, with roughly appropriate levels of tentativeness and epistemic caution.
I'm left wondering if "sounding more like GPT-3" might be a common feature of attempts to communicate across large inferential distances with significant amounts of nonshared referents. How could one distinguish between "there's no there there" and "there's a there there but it's inaccessible from my current vantage point"?
How could one distinguish between "there's no there there" and "there's a there there but it's inaccessible from my current vantage point"?
By having experienced meditators independently rate anonymous articles, some of them written by actual meditators, other written by people who don't meditate but believe that they can generate the same kind of GPT-3-ish output?
Yeah, that'd work for investigating the hypothesis and is interestingly similar to the theoretical ideal of the peer review process.
I was personally more curious about how to distinguish in the general case where reliable domain experts may not be recognizable or readily available but that question may be rationality-complete.
Something like: there are dozen groups of people believing they are experts on meditation, each of them believing the other groups are wrong -- how do find the right ones?
In that case, perhaps instead of talking about "meditation", we could define "meditation1" as whatever the group1 believes, "meditation2" as whatever the group2 believes... and test independently which groups can be easily fooled by GPT-3.
In my head I call this the "word salad" phenomenon, where one can read something and be like "that's just word salad..."
I think about this a lot because inferential distance makes "calling word salad" more likely, and it's maybe especially common in pre-paradigmatic domains.
Honestly, I see it a lot between meditators at all skill/knowledge levels, and, imo, it's often a good call. Sometimes it's a good call for part of someone's body of work and a mis-read for another part.
In general, I see it especially between "same-level" experts or autodidacts (again in pre-paradigmatic domains), as well as between non-experts and experts, in both directions. (The "expert" thinks "word salad" when a non-expert tries to convey something, and/or vice versa.) "Expert," here, could be replaced by advanced practitioner, unsocialized autodidact, crackpot...
I think the principle of charity is helpful, but also there's only so much time to evaluate claims.
When I'm evaluating something, I sort of run through a list of referents, concepts, relations, jargon/terms-of-art, equivocation, causality, implication in no particular order.
Have I ever encountered ANY of the referents before, as best I can tell? Are words being used in non-standard ways? Is the "language game" ostensive, assertoric, logical, mechanistic and/or all the above? Do the concepts and relations feel like they're "sufficiently high quality"?--how blurry are the edges? Are they of relatively small number? How elegant is the thing, overall? Do words change referents? Is referent-switching "doing useful work" or driven by lack of good vocabulary? What's the degree of causality or mechanism that inheres in the referent, or the degree of implication or argument that inheres in the writing?
Sometimes one has to eject or short-circuit the evaluation process. I use the above questions to do that. But, if I have time or there are outside-view reasons to give something a longer look, I try not to drop it until (a) I have an explanation of the generating process that gave rise to the statements or artifact I'm encountering (what is the sociological/epistemic causal history of this?), or (b) I have a more general explanation for which what I'm encountering is a limit case or edge case.
Because writing and speaking are "correlated with reality," even if tangled/confused, I think it's really powerful to "give word salad a chance," because people are exposed to different patches and trajectories of reality, and, modulo bullshit, it's never word salad from the inside. I think there's often net-alpha to be had, for the work put in, when someone is trying to communicate in good faith, and even when not.
(And it can create affordances to correct errors on both sides, can create a feedback loop for dispelling the curse of knowledge, etc., etc., etc.)
But, yeah, sometimes it's better to disengage or to put up a communal wall.
This post parses to meaning pretty fully for me, but I'm somewhat familiar with Mark's writings.
In case it helps anyone else, as I read it the key points are:
hypothesis: the mind is highly neuroplastic in the long term, capable of arbitrarily "large" error corrections. Also there are momentary moves it can do to encode more than one piece of information in the same bit of network. Inference: while maybe arbitrary neuroplasticity and error correction is possible in the limit, locally this looks like doing as series of highly-constrained changes like a sliding puzzle. We probably have some particular neural mechanism handling these updates 1b) these updates look like going through changing layers of encodings one at a time, to be more faithful/useful: "deconvolution" was a helpful word for me here. (I'm not quite capturing the meaning of this point but it's also somewhat hard to without drawing diagrams)
hypothesis: the mind doesn't have noise in it, only mis-encoded signal - error. (with 1, this makes it possible to "error-correct", locally and eventually globally with enough repeated local work)
hypothesis: there aren't "separate representations" for memories, ontologies, models etc: it's the same kind of network (perception=>action) all the way down with just lots and lots of layers (which contain information abstractions) in the middle. Inference: you can use the same kinds of "moves" to 1b) error-correct the whole thing, eventually. You don't need different "moves" for different kinds of "stuff"
some papers associated with the popular articles above, for precision and posterity:
Someone noted through a backchannel that the thing I'm trying to say in the model/error/noise section is maybe just "rather than treating some neural phenomenon statistically, it's better to engage with that phenomenon as concrete mechanism [or meaning-laden]." That's maybe super hard for high-complexity, hard to measure, or pre-paradigmatic contexts—which is why people often start or finish with statistics. Anyway, there's a map/territory thing I'm circling, here, in not the best way. Or it's hard because of friction between mechanism versus telos frames. They may chime in; mistakes mine.
You may enjoy this old paper (functionnal properties of somatosensory maps are not only plastic but actively constructed):
[cross-posted to my blog]
[About me: I'm the meditationstuff guy and the "folding" guy. I have a PhD in bioengineering, during which I did work with human clinical EEG (and also audited neuroscience and machine learning classes), but I'm not a neuroscientist, and I didn't run this by any practicing researchers before hitting publish.]
As an introduction, I wanted to just mention up front that I'm not going to talk about predictive processing, artificial neural networks, GPT-N, neural annealing, the free energy principle, jhanas, amygdalae, cerebella, agent-based models of mind, and a bunch of other stuff. Good stuff, though!
What follows are a sprinkling of popular and academic neuroscience references that have been resonant to me as a long-term meditator and meditation writer/teacher. I don't write a lot about neuroscience because it's paradigmatically changing too fast. So, instead, in other places, I try to to talk about meditation in a contemporary-yet-timeless way, hopefully future-proofed (using philosophical, folk, and mathy language). But, in the conclusion, here, I talk a very tiny bit about my vision for meditation instructions that are rendered in a neuroscientific ontology "all the way down."
I think meditation is a bit contentious in parts of the LessWrong community, out of concern for individuals and the wider community. I personally don't recommend that most people meditate, but I would generally like people to know that (I personally believe that) "there's a there, there," in the spirit of truth-seeking and because, for some subset of people (I claim!), it's really valuable. In the conclusion, I warn against "naively operationalizing" neuroscientific writing about meditation, and I list a bunch of peer-reviewed studies on meditative risks. Finally, I've left out a few possible sections that I thought were particularly tempting to "operationalize."
Finally, I wrote this all in one go, so that it would definitely ship, and it becomes progressively less coherent through to the end (I think).
A meta-representational level of organization and computation
According to wikipedia, "somatotopy" is the point-for-point correspondence of an area of the body to a specific point on the central nervous system. We know that amputation or deafferentation changes an organism's somatotopy and that these changes progress over time. It's relatively uncontroversial that parts of the brain can "rewire" themselves, to some degree, after organismal insult, so this is maybe not surprising.
It might be a little surprising how reversible these changes can be.
A 2008 paper [1] is titled, "Chronically Deafferented Sensory Cortex Recovers a Grossly Typical Organization after Allogenic Hand Transplantation." The abstract says,
It's just one paper, but there are nearby papers in "paper space." These results suggest that at least the sensory cortex maintains significant plasticity throughout one's entire life, or at least maintains latent plasticity.
Losing and reattaching a (new) hand is a very large change. What about the "changes" of daily life? The Atlantic reports on the phenomenon of "representational drift." [2] (There are links to the relevant papers, in the article.) For example, neurons that represent particular odors change over time ("in mice")—within a month, representations of the odor remain, but the representing neurons are completely different. This sort of phenomenon is found in several brain regions. Also, from older research, it's sort of a truism that expert skill execution recruits less of the brain (and fewer muscles). Together, these results suggest that, representations in the brain are simultaneously both continually in motion (copied, transferred) and continually being sculpted. (When I say "continually," perhaps this is during REM sleep, or perhaps it's when awake behavior partially recruits those representations, or maybe it's actually "continuously.")
For some possibly relevant mechanism, according to wikipedia, in 2007, evidence was first found for so-called "didactic organization," though this was predicted at least as early as 2001. [3] "Didactic organization is the ability of neurons within a network to impart their pattern of synaptic connectivity and/or response properties to other neurons."
Relatedly, a popular article [4] reports in vivo results ("in mice") showing that "the brain rotates memories to save them from new sensations." In summary, "some populations of neurons simultaneously process sensations and memories. [...T]he brain rotates those representations to prevent interference [n.b. via 'orthogonal coding']." The article also reports on similar work with monkeys, but in this case it's orthogonal activity in the motor cortex, to separate motor planning and motor output. To emphasize, they claim it's the same neurons: "Experimentally sifting through the possibilities, they ruled out the possibility that different subsets of neurons in the auditory cortex were independently handling the sensory and memory representations."
These days, I think rapid switching network configurations are uncontroversial, and ditto for storing information in the same same networks at different frequencies, and I think the "darwinian nature" of the brain is fairly well accepted at the "neural level" and possibly the "direct representational level," i.e. representations in local competition for neural substrate.
But, to my mind, all the above suggests that there is a sort of "meta-representational level" that can shunt around and transform representations somewhat independently of particular neural substrate. (versus, say, "substrate-locked representations" Hopefully the distinction is clear. And also I haven't super stress-tested this idea.)
Meditation Interlude 1
These popular articles and this research caught my eye, because, as a long-term meditator, I sometimes have the visceral experience of almost a "raft" of representation (or a reusable computation) "detaching from somewhere" and "drifting" until it bumps into stable structure. This is hardly an argument for anything, and phenomenology doesn't have to provide intuition about neural structure and function, and in fact phenomenology is often quite misleading, but, for what it's worth, these experiences are what made the above material stand out to me, over the past few years.
There's also something important that the above leaves out--I haven't come across any nicely exemplifying research--and it's something like "substrate contention" or just "constraint." While, above, I suggested that there is a "meta-representation layer" that can ship around representations somewhat independent of neural substrate, the available "shipping routes" at any given time seem finite and non-arbitrary. To be a bit more clear, through analogy, meditation, to me as a long-term meditator, has much in common with sliding puzzles [5] and Tower of Hanoi-like problems [6]. Over weeks and months, there is tremendous flexibility. But, locally, far into meditative progress, there's often very few degrees of freedom (after using up some degree of "local slack.")
(And this implies local maxima and so one of the main functions of meditation and other self-transformative practices are for stepping backwards out of local maxima.)
The Brain is Meaning-Laden and Erroneous but not Noisy
I think "active perception" [8] and "active inference" [9] are fairly well known. In these paradigms, the organism is actively sampling the world and deliberately altering the sensory apparatus to more efficiently maximize information and falsify hypotheses (and perhaps minimize free energy and etc.).
I want to very loosely combine the above ideas with some interesting experimental results to make a bold, hand-wavey to counter a vague popular intuition that people often have about brains. That's all pretty vague. Let's begin.
[Content warning for several subsequent paragraphs: invasive animal research] A lot of sensory neural coding experiments are done on anesthetized animals. The animal is perfectly still and "unconscious" (or perhaps barely-conscious, with drugs titrated to get a requisite level of brain activity). And the animal is stimulated, somehow, for example visually or aurally, with a mathematically parameterized stimulus, while neural recordings are taken.
But, more recently, relatively more data is being collected from awake, behaving animals. Reported popularly [10], researchers found that, for example, "The visual cortex knew exactly what the animal was doing, down to the details of its individual movements." The article quotes a researcher not affiliated with the study who says, “Everywhere in the brain, it’s the same story. The movement signals are just really unmistakable.”
In a different article, the same popular publication reports on "aperiodic background noise." [11] In my own words, and this is more a speculative interpretation and hardly a summary at all, this "noise" may indirectly subserve representation or inference or it may directly represent, but, in any case, this "noise" may not be "noise" at all.
What do I mean by "noise," then?
Before writing this section, I went briefly looking for a very concise statement about the difference between error and noise with respect to a model. I found something good enough in a somewhat random post by an individual named Adriaan Peens-Hough. Thank you, Adriaan. [13 [sic]]
Adriaan says the following (bolding mine):
Given all that terminology, I want to first-pass vaguely claim something like, "there is no noise in the brain; there's only model and error." I will expand on this a bit in the meditative interlude.
Meditation Interlude 2:
I think people sometimes experience themselves as sort of "messy" or "haphazard," say in behavior or belief. And popularly, we think of the brain as "wet and messy," a hunk of Jell-O. We talk of brain farts.
Say, Internal Family Systems and some other contemporary modalities (and old-school psychoanalysis--Freudian slips--to be fair). That is, in the popular consciousness, we have some conception that sometimes minds/brains/people do things for no reason, or at least no good reason, yet there's also the conception that we sometimes do things for "deep reasons."
After thousands of hours of meditation, I'm mostly on the "deep reasons" side. Or at least perhaps the "always reasons if not always deep ones" side. (Importantly, though, these reasons aren't necessarily first-pass or n-pass explicitly articulable; and I don't think "reasons" are a natural kind).
Instead of "reasons," maybe it's better to claim that sensory data is always "interpreted" or that almost all neural activity is nearly synonymous with interpretation.
I mentioned slider puzzles and Tower of Hanoi problems above.
To be sure, phenomenology can be buzzy, shimmery, sweeping. Attention can be choppy. "Thought" can seem fragmented, repetitive, loopy.
But the impression that emerges, at least for me, over thousands of hours of meditation, is of something the opposite of "mush" and instead something of "thresholds, criticalities, steel cables," something far closer to a Babbage difference engine than, say, a monkey mind or electrical impulses rattling around in Jello-O. Not springs and rubber bands, but gears and rods. The monkey mind is instantiated by the Babbage difference engine.
This is getting a big vague, be degrees, as I run out of steam. But: model error, not noise.
Source Localization and Epistemics
[This section is copied from a facebook comment I made [12]]
This is very terse/schematic, but I predict improvements in:
(1) attribution and provenance
(2) reasoning
(3) transitive closure and de-contradicting of current beliefs
*(1) attribution and provenance*
(a) better attribution of the causal chain of an belief update and
(b) improved concepts/ontologies,
because (a-b) are causally downstream of improved source localization and separation/deconvolution of sensory phenomena [1].
[1] https://en.wikipedia.org/wiki/Cocktail_party_effect
*(2) reasoning*
Improvements in sensory processing are also improvements in reasoning, even though high-level reasoning processes might seem remote from low-level sensory processing. Here are some additional claims, with some loose argumentation:
*Past* sensory impressions/memory, in a sense, become/condition/sculpt the structure of *current and future* sensory processing *which includes* the reasoning process itself, which itself is *partially* a sort of "sensory processing" of "inner experiences."
So sensory processing, at the neurological level, is deeply entangled with high-level reasoning.
And so even high-level epistemic errors can be traced back to past or current sensory processing errors.
*(3) transitive closure and de-contradicting of current beliefs*
Claim: Past (and contemporary) sensory processing failures, even ones from decades ago, can be corrected, which can cause a cascade of conceptual/belief improvements, to old beliefs, as well as the bleeding edge of belief, in the present.
This process of correction involves re-processing old sensory experiences, which includes reprocessing prior experiences of the reasoning process itself, and thereby a refactoring of the reasoning process itself (again because of how experience becomes structure/process).
Because of "compression" and a strange sort of quasi-losslessness, it's possible in meditation to go all the way back to childhood traumas, very old epistemic errors, and so forth, in less total durational time than it took to live them. (cf. also the so-called memory reconsolidation literature) And that reprocessing cascades through the rest of one's belief system causing further improvements, upgrades, and so on.
I analogize meditation to paying off technical debt. [More sections in my doc:]
https://meditationbook.page/#181
I made a lot of inferential jumps in the above, for brevity!
(Miscellaneous) Interlude 3:
Finally, I want to talk briefly about representations, as such. Above I refer to representations as if they're "real things that exist that directly represent other things." I'm taking a strawman interpretation of my own writing above. In that vein, one could imagine an organism with a brain as something like:
perception --> update representations (beliefs/goals) --> action --> perception --> action...
In place of representation, I like "functional (stateless) computation."
That is, the "state" of the system is stored in "that which continuously, waterfall-style, without feedback loops, computes motor outputs from perceptual inputs." (In that scheme, the "flow of information" through the computation directly and continuously shapes the computational substrate to optimize the computation.
But of course we know there's feedback and feedforward in the brain, so it's not quite like this. But I want to strawman caution against inferring the existence of "representations as such" from symbolic behavior such as speaking, writing, and even thinking.
Conclusion and Caution:
So the goal of this piece was to highlight some relatively more contemporary results in neuroscience that have piecemeal stood out to me as a long-term meditator. In my main meditation writing, I don't talk much about neuroscience because neuroscience is in its infancy and it's currently difficult to write meditation instructions in a way that "directly operationalizes neuroscientific concepts." If I wrote a lot about neuroscience and meditation, it'd all be more and more obsolete with each passing year. So I've tried to future proof my writing by using folk, philosophical and math-adjacent concepts, in both gestural and precise ways.
Maybe neuroscientific concepts will never be the right ontology. Of course stroke victims and TMS research subjects can sort of learn to differentiate when precise local substrate is or isn't working. And I will say that, when I started meditating, I "didn't feel like a brain," but now the long-run shimmer and sweep of meditation, at the finest phenomenological grain, feels suspiciously like brainwaves (the frequencies are right) and subtle, very fine-grain aperiodic shimmering across the whole phenomenological field (pervasive during meditation but not while living life) feels suspiciously like the phenomenological correlates of synaptic potentiation and depotentiation as such. But, anyway, still, I'm not sure, and maybe "neuroscience" will always be the wrong level of abstraction for meditation
And, anyway, in any case, I dream of meditation instructions that are "neuroscience all the way down."
Caution:
The above (and here) serve as a bit of a caution, too. Willoughby Britton [*] finds a non-negligible percentage of severe sequelae in meditators, even for those under the supervision or ostensibly qualified teachers.
Neural state space is finite but vast. I currently guesstimate that it takes about 10,000 hours to walk the "whole thing" even just once, loosely speaking. I believe that being safe can be very counterintuitive and even "accidentally optimal" meditation can be a rough ride. For what it's worth, the above concepts and ontologies in this post aren't remotely like the ones I use in my writing and teaching about how to meditate. Please don't operationalize any of the above in terms of "bespoke meditation instructions" or please be careful if you do.
***
[*] Britton et al., and others, from her citations:
***
[1] Frey, Scott H., et al. "Chronically deafferented sensory cortex recovers a grossly typical organization after allogenic hand transplantation." Current Biology 18.19 (2008): 1530-1534.
[2] https://www.theatlantic.com/science/archive/2021/06/the-brain-isnt-supposed-to-change-this-much/619145/ [Last accessed: 2022-02-07]
[3] https://en.wikipedia.org/wiki/Didactic_organisation [Last accessed: 2022-02-07]
[4] https://www.quantamagazine.org/the-brain-rotates-memories-to-save-them-from-new-sensations-20210415/ [Last accessed: 2022-02-07]
[5] https://en.wikipedia.org/wiki/Sliding_puzzle [Last accessed: 2022-02-07]
[6] https://en.wikipedia.org/wiki/Tower_of_Hanoi [Last accessed: 2022-02-07]
[7] https://en.wikipedia.org/wiki/Self-organizing_map [Last accessed: 2022-02-07]
[8] https://en.wikipedia.org/wiki/Active_perception [Last accessed: 2022-02-07]
[9] https://en.wikipedia.org/wiki/Free_energy_principle [Last accessed: 2022-02-07]
[10] https://www.quantamagazine.org/noise-in-the-brains-vision-areas-encodes-body-movements-20191107/ [Last accessed: 2022-02-07]
[11] https://www.quantamagazine.org/brains-background-noise-may-hold-clues-to-persistent-mysteries-20210208/ [Last accessed: 2022-02-07]
[12] https://www.facebook.com/robbensinger/posts/10165449844265447?comment_id=10165449991010447&reply_comment_id=10165450085795447
[13] Peens-Hough, Adriaan. (2016). Re: What's the difference between noise and error in a dataset?. Retrieved from: https://www.researchgate.net/post/Whats_the_difference_between_noise_and_error_in_a_dataset2/56f2628340485479c609ec0b/citation/download.