There is good evidence for its use in ADHD but not using home kits. Practitioners take a 20 10 quantitative EEG and from there create a program that targets certain wave bands
Do you have any good review/meta papers about effectivenes in ADHD? I always find the neurofeedback literature full of small sample sizes, lacking appropriate controls (sometimes even none at all) and small effect sizes, which makes me generally distrust the field. However, I'm also from a different domain, so I'm not familiar with the landmark studies and google scholar is only so useful for finding the best meta/review.
Unless it's localized to a small region (partial seizures), epilepsy tends to look like strong hypersynchronous activity. Large numbers of neurons firing in unison without the harmonics and complexity of typical conscious states.
What I'm interested in regarding synchronized brain waves are interferences. I suspect that consciousness is the ability of humans to coordinate brain functions in such a way that they "compose" higher order patterns of brain waves that interfere with each other, creating the brain wave equivalent of a moiré pattern. When we evolved conscious minds, we basically learned to create those patterns.
I would say that consciousness is a name we give to some examples, but not others, of a certain style of organization.
The relationship between "you" and "your consciousness" is precisely identical to the relationship between "Amazon" and "Jeff Bezos". That's what consciousness is.
Jeff Bezos is the official representative of Amazon, notionally responsible for everything it does (and for making the relevant decisions), and exerting actual control over a small but significant portion of that.
I do not feel everything that happens in my body, but I do feel it if blood vessels are damaged by, say, a bat. Bezos does not feel anything in his company outside his body, unless he has new implants I was unaware of.
> Ingram notes that every sensation vibrates in and out of consciousness at a rate of between five and forty vibrations per second, sometimes speeding up or slowing down depending on your mental state. I’m a pathetic meditator and about as far from enlightenment as anybody in this world, but with enough focus even I have been able to confirm this to be true. And this is pretty close to the frequency of brain waves, which seems like a pretty interesting coincidence.
Can you go into any more detail about what it's like to experience sensations in this discrete way? Also, kinda off-topic, but what's up with the jhanas? If anyone can experience mega-bliss states with a few months of proper training, shouldn't that be, like, a major EA cause area?
Love the idea of EA helping people learn to trigger jhanas. I suspect what’s going on is a bunch of adults are unwilling to believe the jhanas are real.
I mean, you can also experience a mega bliss state by taking heroin. Not that they're at all the same (for one, I don't think jhanas destroy peoples lives through addictiveness) but what I'm trying to say is that humans have multiple options we already know about to achieve bliss like states, so is there a particular unique benefit to jhanas that would make then a higher priority than keeping people from dying?
This probably depends on your causal model. Does helping people feel good help them perform better, and thus make more money which can be donated to EA causes? Do people who feel better naturally want to donate more?
These pleasures can can be achieved without intense bodily harm or risk of addiction. If helping people induce harmless bliss states is only a terminal good, then sure it’s maybe not THAT important. But even then, when 50,000 Americans die each year from suicide, it seems like maybe this isn’t a bad place to start. Especially if the big barrier is simply creating the public knowledge that these states exist .
I agree. EA should be about reducing suffering, not just blindly preserving life. Anything that helps treat depression, or improves mental resilience, should be a huge EA cause.
As someone who's spent a couple thousand hours meditating (with amazing practical benefits) and never achieved a jhana who has many friends who have spent a couple thousand hours meditating (with amazing practical benefits) who have also never achieved a jhana, I think the premise that "anyone can experience mega-bliss states with a few months of proper training."
(I believe jhanas are a real brain phenomenon, I just think they're quite difficult to achieve.)
Aren't there different styles of meditation which lead to different results? Perhaps the approaches that you and your friends take aren't suited to jhana attainments?
I've generally studied in the "pragmatic dharma" tradition, which includes both Culadasa's "Mind Illuminated" and Ingram's MCTB, and seem to be the place where most rationalists start. These works both claim (more or less) that jhana and even awakening can be reached relatively quickly, but my experience is that it is widely known in these communities that the central works overclaim heavily regarding typical timelines (while still delivering massive benefits).
Many teachers of more traditional approaches will say that basically nobody except dedicated monastics will achieve jhana, and they'll only achieve it after many years of study. (The pragmatic dharma "response" is that we have much better methods now, and I think there's something to this, but I still think the overall claim is too strong.)
So is it possible I'm wrong, and "anyone can experience mega-bliss states with a few months of proper training?" Sure it's possible. But if we narrow the claim slightly to "a few months of proper training with mo more than 400 hours training total, including instruction," I would bet strongly against this. A specificish bet (making it very specific is too much work since the bets here aren't runnable) I'd take is "I'd bet $1,000 against $100 that no currently existing method could get 75%+ of no-substantial-previous-experience meditators to jhana within 400 hours of total training."
I guess it's also worth noting that different traditions / teachers sometimes use the word "jhana" differently, but I'm allowing Luke's original description ("mega-bliss") to restrict the word to very highly concentrated states (which is how I'd naturally use it).
(Some of the most interesting relevant data is probably from Jeffrey Martin et al.'s Finders Course: https://finderscourse.com/. Their current claims is they have a short-protocol that gets a lot of people to something they're calling "Fundamental Wellbeing", but spending a bit of time looking at the paper, I don't think they're making any claims about mega-bliss / jhana?)
Is there any evidence that the positive feelings from jhanas have anything to do with depression and suicide? While it seems like it has an obvious connection, it also seems like depression is just as prevalent in people who meditate as people who don’t, and enlightened monks aren’t immune from it.
That's almost certainly true, due to sampling bias. Suicide is overwhelmingly correlated with mental illness, whereas the prerequisites for achieving jhanas include strong discipline and mental stability, not exactly common among mentally ill.
But I agree--it doesn't make you immune. And it certainly doesn't make you immune to sadness or fear! To quote a favorite song of mine: "you can't change the feeling, but you can change the feeling about the feeling"
Well, if the options that you have in mind are all like heroin and cocaine, some of the most cripplingly addictive and physically devastating drugs, not to mention highly illegal, then I'm sure that most people would say that a widely available alternative with milder side effects would present plenty of unique benefits, like making anybody who'd be able to turn it into a easy to use marketable product the richest man alive. Maybe this isn't exactly EA's main prerogative, but I'm sure that they wouldn't be too unhappy if it just happens to fall into their laps.
> so is there a particular unique benefit to jhanas that would make then a higher priority than keeping people from dying?
Jhana only requiers interanl input. Heroin is an external input. TV is an external input. Sexual stimulation is an external input. This blog is an external input. They could all bring you to the blissful state, the difference is internal vs external. When it comes from the internal, then it is always within reach. Externalities are ephemeral and not always within reach.
Binding frequency of consciousness. It was also mentioned in The Mind Illuminated with a footnote to "Selfless Insight: Zen and the Transformation of Consciousness", which counter to its hippie title is all about meditation experiences mapped to specific neurological processes with mechanistic explanations. Cannot explain it, have not read it yet, as I want to learn more neurology first. But it will have your answer. (or possibly in one of Austin's other two big books of hardcore meditation neuroscience with deceptively hippie-ish titles)
Not sure if he does Jhanas. There's a Leigh Brasington thing on them, but that seemed outdated and still crude. I have had my own theories about them, but they're too half-baked to even remember, let alone state.
1. Meditation teacher Rob Burbea was (and I agree with him) somewhat critical of the notion that "all experience is discrete mind moments coming one after the other". Not that this is not how you will start to perceive things if you really tune into the quick changing of experience. But rather that this is an artefact of this specific way of looking, rather than an universal aspect of experience. It's not that we fail to recognise the discrete mind moments in daily life (or when meditating in a different way), its that they are not there.
2. Jhanas are more finicky than that. Many people will not be able to get them at all no matter how much they practice. For many others, it takes months of intensive practice in the right conditions and jhana becomes unavailable once the intensive practice stops. Most theravada teachers I've heard said something to the effect that you should get as much concentration as you can reasonably get (including jhanas) and then just do insight practice.
> Many people will not be able to get them at all no matter how much they practice.
Is this something that your teachers said? I haven't come across this assertion before (and, I admit, I don't really want it to be true...). Is this one of those "hard jhana"/"soft jhana" things?
Basically yes. Both for weeklong jhana retreats as well as 3 months and longer intensive retreats jhana will onl happen for some people. I have also heard numerous teachers say that they themselves were unable to experience jhanas (the question of jhanas gets asked in like every second retreat I've ever been on and I'd say about half the time the teacher responds that they can't really comment on that because they never got there, sometimes not for lack of trying). I believe the harder the jhana the less people are likely to be able to get it, but even "Mr. soft Jhana" Leigh Brasington states in his book that only some people will be able to get his version of Jhana in his workshops (though the caveat may be: maybe if they practiced for a longer period of time...?).
I could imagine that if you increase the practice time enough and go for a very soft jhana most people would be able to get there eventually (if sporadically). But at that point it makes more sense to frame the whole thing as "concentration practice" rather than jhana practice and here the usual advice is to just go for as much concentration (shamatha) as you can reasonably get without significant strain. And it is assumed that the outcome will be very different for different people (as in all things in meditation).
I mean a more general point is that any time you consider a concrete goal in meditation, the timeline to reach it will be orders of magnitude different for different people. From getting it immediately or with minimal practice to "not in this life".
Yeah, I think there's loads of confusion with the one-after-another assertion. If I focus on lower-level vision percepts (especially in low-light conditions), I can see individual color sensations vibrating in the form of millions of little dots at once.
I am not seeing millions of discrete things one after another in the span of a second. Instead I am seeing them all at once.
Maybe once could claim, that I am just seeing binding moments sequentially and each binding moment gives me a picture of the whole complex visual field (combining millions of lowel-level mind moments? Combining some of them and then automatically using another part of the brain that generates an algorhithmic illusion to seemingly "fill out" the rest of the map?)?
I still think that the mind-moments model is a great starting point for investigation. Just gotta adjust it quite a bit, as one learns more about how things actually are.
Integrating fuzzy logic, parralel computation, fame-in-the-brain, the actual neurological layout etc.
Yeah. My take on this is that if we don't try to meddle with our experience, that is don't try to fit our experience to some kind of implicit standard, belief, need etc. we already know exactly what experience is like. Any time we try to find a conceptual framework to describe it we need to acknowledge that the framework cannot provide a full description. And that if our model makes very concrete claims (like discrete mind moments) it is probably reductionary in some way.
Sidenote: This meddling can get very very subtle. On the other hand non-meddling is enlightened looking. Its like what the Taoists mean with Non-action.
"all experience is discrete mind moments coming one after the other" - this seems like a statement with essentially no evidence. What little there could be is purely introspection, and seems super unreliable.
Until someone manages to design some experiment that could tell whether experience is made up of discrete moments or is continuous, it strikes me as an empirically empty statement.
Event-related potentials (ERPs) are like single-shot brain waves that occur in response to stimuli; had to take a whole section on them in grad school (with the guy who wrote the ERP textbook), and I still was not sure how they were supposed to be important. Most of it was used in research on attention (e.g., this spike would occur if you noticed X but not Y); I think they are supposed to be informative because you can measure them without having the subject respond at all (no button pressing responses, just pure noticing). At least they don't bring out some of the "woo" or just speculation like brainwaves tend to do. Was it Crick & Koch who thought consciousness was (possibly) a 60 hz oscillation of the frontal cortex or something like that?
In my undergrad biology program we visited a brain research lab near Moscow. The brain scientist gave us a brief intro to Fourier transforms, which made me understand how beautiful they are - something that 2 years of undergrad math classes didn't manage to do.
Then he explained the brain waves to us like this:
"Imagine you are standing outside the football stadium. You don't see what's happening inside, but you hear the chatter of the crowd. All the individual words blend together into indistinct mess and although there's definitely a local information transfer going on, from the outside you can't make out anything specific.
Then imagine one of the teams scored a goal. The crowd behavior is now very different! The fans of the winning team start to cheer and sing. You can easily pick this up from outside and infer what's happening. This is because the individuals behave in a globally coordinated manner, so their signals amplify each other in tune.
From this perspective, brain waves are a byproduct of globally coordinated neuronal activity, and it's the first one we historically learned to pick up. They appear when neurons stop chatting with each other and start chanting in unison."
Then he plopped some probes on my head and announced I have beautiful epileptic spikes (I'm not an epileptic)
I was in a workshop with some people who studied "Grandmother neurons." The story is that a particular neuron only fires when you see your grandmother. This is a false. The true story goes something like this: Imagine encoding memories as combinations of 88 piano keys (neurons). If you used a binary encoding, you could store 2^88 memories, but if you got one note wrong, it would be a totally different memory. If you stored one memory per key, you could only store 88 memories. The brain actually uses sparse codes-- a memory encoded as a chord, only using up to 10 keys played at once (since we only have 10 fingers) This can store "88 choose 10" which is about 4 billion memories. So when the scientists measure a hundred randomly sampled neurons, it just so happens that one particular neuron only participates in the chord for the grandmother in the tests. Does it participate in other chords? Yes, but they are all for images the scientist didn't test. Are there other cells involved in the chord for the grandmother? Yes, but the scientists didn't stick a measuring device next to those neurons (it's hard to do and if you stick in too many probes they damage the tissue).
This also seems much more intuitive and efficient + it falls in line well with what apparently happens in Machine Learning for example in word encodings: Each word/concept will rather be represented by some vector in the space of activations, the "neurons"/dimensions themselves meaning different things.
The natural solution would be to have similar chords refer to neighboring areas in the cluster of thingspace, like encodings for grandma and grandpa to differ by a single note.
That would also explain some real-life failings of memory. Suppose I had two uncles named Alex and Ben. I was talking to/about Alex but mistakenly referred to him as Bob without any particular reason other than both running under the "uncles" category.
I feel like it's probably not going to be totally different, it's going to include all the lower-bit memories at the same time. If one bit is "the sea" and another is "gray" and another is "cross-section windows" and all those are in the pattern of "grandma", you end up with "grandma who lives in the gray house with the cross-section windows by the sea."
So if a bit misfires and you don't think "grandma", you'd still get "like someone who lives in a gray house with cross-section windows by the sea". Kind of like shopping lists, where you can remember three items, and you can remember there was a fourth item, but it's pretty random whether you can actually remember the fourth item without checking the list.
In the binary encoding, yes. Every combination of keys identifies a unique memory.
In the unary encoding, yes if and only if "one note different" means replacing one note with another note. Adding or subtracting a note give you an invalid code.
In the chord encoding, the behavior is the same as the unary encoding. Replacing one note with a different note gives you a valid chord representing a different memory. Adding or subtracting a note gives you an invalid chord.
You might be interested in the concept of "error-correcting codes".
Arguably, even AI models have "waves". It is just that with the currently dominant paradigm of transformer models there is only a single wave going from input to output and no oscillations or recurrences. This used to be different, recurrent models like LSTMs, Neural Turing Machines and other older ideas could be described as oscillating. It just turned out that these recurrent models are hard to train with current methods and to scale to large amounts of data. Interestingly, recurrent models are thought to be better at reasoning tasks and algorithmic learning, something that transformer models are bad at ("bad" here meaning in comparison to their other skills like memorization and language understanding).
Huh, I know that waves can cause issues with single-step implicit numerical schemes. The single step will amplify the wave, eventually leading to the numerical solution exploding. In atmospheric models, predictor-corrector time stepping schemes are used that analytically damp the waves. I wonder if there's anything like that for training recurrent models and it it actually helps at all.
This sounds a lot like filter stability in iir filters. It's a stable filter as long as all of the poles of the transfer function are within the unit circle in the z plane - in plain English if you look at the frequency response graph nothing is above 1, so all becomes dampened over time.
So yeah, part of the issue was vanishing or exploding gradients with classic RNNs. You can kind of imagine the "transfer function" of the net being either too dampening or too exciting. The gate in LSTMs was designed to help with this by stopping or opening the flow a bit automatically.
My complete layman's guess is that a lot of this stuff only really applies well to Linear systems, and when it becomes non-linear it gets way harder.
Any recurrent network definitely shows oscillatory behaviour - I used to train GRUs (simpler variant of lstm) a while back for sound generation. Part of the difficulty of using them was how unstable they could be - things would work really well for a few seconds, then just degrade into noisy or weird feedback loops..showing classic chaotic behaviour. I managed to get some small ones running in realtime on fft data, with live input bled in from a microphone and the state feeding back into itself. You could drive it in and out of different stable points with different inputs, much like microphone feedback in a room if you've ever played around with that.
I think this is prevelant in almost any recurrent system, which raises the question of how our brains don't just collapse into noise or get stuck in endless loops...they're kind of finely balanced to be able to skirt around the edges of various attractors, moving from place to place.
> What’s that tune you can’t get out of your head?
There's an episode of Star Trek: The Next Generation in which a character is psychically afflicted with a melody and goes insane, and then loses consciousness, from the inability to escape it.
But that doesn't happen to non-fictional people. The tune you "can't get out of your head" is in fact easy to get out of your head by doing pretty much anything other than thinking about the tune. Nobody gets stuck on it, in the sense of being unable to do other things. Rather, it's a phenomenon that occurs while the brain is idle, and goes away when the brain stops being idle.
Some deeply dysfunctional people can get stuck in loops, and perhaps obsessive/compulsive behavior is an example of that kind of thing, but the often-complained-about phenomenon of having a tune in your head is not.
Well, think of it as a milder form of the same phenomenon. The review talked about how the various oscillations can be brought into some kind of “harmony “ by altering the frequency of some of the waves, which we can do (meditation, bio-feedback, rhythmic excercise etc.)
Yes! So maybe OCD is like being stuck rolling around in a basin of strong attraction. Why isn't it more common given that I think that is the most common failure mode of complex recurrent systems. In fact why aren't other failure modes like seizures much more common?
That's an evolutionary explanation, a bit like responding to "how does this crazy looking bridge stay up?" with -"because otherwise it wouldn't be able to carry stuff over it".
What I want to know is how is the structure of the brain evolved in an arrangement so as to not cause these kinds of issues with sensitive complex systems.
Like everything else, by trying random things and selecting which happened to work. I suspect that what you really want is a description of a particular feature which (mostly) prevents these issues, but our understanding of brain design is likely far too incomplete for that.
Brain waves are weird to me. I think of neurons as having certain jobs. Let other neurons know when there's a stimulus, or when the sum of stimuli reach a thresh hold, or do whatever computation. How does it help if every now and then (by which I guess I mean ~7 times a second or whatever) they all throw that calculation away and do the wave? Brainwaves seem to me like having the music on real loud in a party and trying to hold a serious conversation at the same time. But... music is pretty much ubiquitous at parties, so maybe it's actually a good analogy?
I guess it makes sense to me as a clock. You have your neuronal system, you do a thought, the thought gets fed to the output of your cluster, and then you you briefly force all neurons to high so they get dampened to zero, refreshing the state for the next thought cycle.
Well, from Scott's description it sounds like they keep doing their job just fine, and being a part of a "wave" simply means that their job is temporarily involved in the workflow of another department.
Maybe that helps. My point is, if a neuron firing means, say, "I see blue" how come during a brain wave we don't see all these crazy flashes for color, ditto for sound, smell, memory, emotion?
I guess the answer is that a neuron firing doesn't mean anything so simplistic, but instead contributes to a model of blue, or sound, or Grandma. And the receptors "know" that out of sync firing indicates those things, but in sync firing doesn't. It's hard to get my brain around itself sometimes.
Suppose, for simplicity, it's a neuron in the visual cortex, its firing only means "I see blue", and whenever the corresponding part of the retina receives blue light, it fires. Most of the time it's in your peripheral vision and you're not paying attention, which corresponds to it being "out of sync", but once in a while you focus your consciousness there, and suddenly it's "a part of a wave", while doing the same job all the time.
Thanks I was considering whether or not to order this just the other night, and this review has tipped me into going for it.
Here's an interview about his latest book - which seems to get at the idea that the brain is a self-coherent system, which is stable in and of itself without needing to be "driven" by sensory events. Anyway the summary can do a better job than me, for the lazy its all summarised from about 1:02
I love the examples here. The concept of “a stimulus is only detectable if it hits at the wave Peak” made me think of some mechanism for tuning out annoying sounds.
Minor typo - in the section talking about waves of different frequencies getting into sync, you’ve got the frequencies backward. The 2.1 Hz wave will prompt the 2Hz one to fire early, and the 4.1 Hz wave will prompt the 2Hz one to fire early on its second pass, rather than the other way around. (Everyone always gets wavelength and frequency reversed!)
You've hit on my favorite topic! I too found his stuff on pink noise difficult to parse.
Brain waves become significantly more exciting when you do a deep dive into specific oscillators.
1. Sharp-wave ripples (SWRs, 80-140 Hz) play an essential role in memory recall.
> Girardeau et al (2009) provides loss of function evidence that suppression of sleep SWRs dramatically impairs subsequent memory recall for spatial tasks. This finding is nicely complemented by several gain of function experiments. Fernandez-Ruiz et al (2019) were able to prolong ripple duration, and showed that prolongation improved performance, and conversely that shortening impaired performance. This may explain why novel situations naturally evince longer duration ripples.
2. The SPEAR model, which associates peak of theta wave with past encoding vs trough with future predictions is one of the most extraordinary facts I know.
4. How gamma-beta implement feedforward and feedback signals (in shallow and deep layers of cortex respectively), fully consistent with active inference models. Very strong recommendation for Earl Miller's laboratory. https://www.youtube.com/watch?v=Kqyhr9fTUjs&t=1841s
When I was an undergrad (1993) I had to lead a seminar on something for a senior CSC class. I chose chaos-and-neural-networks because there had recently been some intriguing brain research that purported to demonstrate that oscillations of neural activity were chaotic and developed attractors that corresponded to what the neural activity was "about" (the specific example had to do with the olfactory bulb of some animal and different scents). I speculated that maybe the course of artificial neural network research might change and become less linear (input neurons ↣ more layers ↣ output neurons) and more feedbacky (input neurons ↣ bunches of neurons feeding back to one another perpetually kind of like in those game-of-life animations).
Anyway: I bring this up because after graduation I completely dropped this thread and haven't kept up on it, and now I'm suddenly curious. Does this cognitions-as-chaotic-attractors thing still get taken seriously? Does anyone do anything with artificial neural networks that maintain a constantly, chaotically oscillating state rather than designating some "output neurons" that terminate the activity?
> And the waves solve wetware-specific problems, like conduction delay (silicon chips operate at the speed of light) and synchronization (computers have internal clocks, or can synchronize with atomic clocks via the Internet).
Silicon chips definitely deal with conduction delay. The speed of light is very fast, but at 4 GHz electric signals (moving at roughly 2/3 the speed of light) can only travel 5 cm in one clock cycle. Traveling 1 cm uses 20% of the cycle, etc. Computers also have to deal with various types of synchronization issues for a variety of reasons.
However, the logic here was flawed from the start, because artificial neural networks are a mathematical model simulated on a computer and the properties of the computational substrate of a simulation don't carry through into the simulation. And so ANNs indeed do not have to deal with conduction delays and synchronization issues the same way that biological neural networks do, but the reason does not have anything to do with the low-level details of silicon chips.
In an ANN, there is simply a list of neurons, a record of which neurons are connected to which others, and an assignment of weights to these connections. There is normally not an assignment of lengths to the connections. Therefore there can be no conduction delay, for example. You could create a nonstandard ANN with lengths and simulated conduction delays if you wanted, as that freedom is the nature of simulation. The conduction delays in the wires within the silicon chip don't inherently create conduction delays in the ANN, because the connections in the ANN are not wires in the ANN, they are numerical data upon which the silicon chip performs calculations. Just like the flatness of the silicon chip doesn't prevent simulated 3D environments, etc.
This may seem piddly, but it's important to keep in mind that the properties of an ANN come from the properties of the mathematical model, not from the properties of wiring in silicon chips.
As someone who knows nothing about the subject, my first reaction is: How do we know brain waves aren't just random noise to which we wrongly impute meaningful patterns? The post mentions that "pink noise" exists in stock market behavior. Aren't waves in the stock market the kind of things technicians love to write books about but which ultimately have no value for traders?
Something about this vaguely bothered me, and after a while I came up with the following dumb analogy to attempt to explain my thoughts:
An alien scientist publishes a paper saying "humans appear to have natural patterns in their behavior, tending to become quiescent for around 8 hours then active for 16 in a 'wave' that sweeps across the planet", and so on. But another alien then, reading it, responds: "how do we know this isn't just random noise to which we wrongly impute a meaningful pattern?"
That is, depending on their aims, this may or may not be interesting or useful — is "well, they sleep when it's dark" surprising and cool, or does it destroy a fascinating hypothesis about humans possessing a planetwide vibrational "wave of consciousness" or something? — but it's certainly *a real observation.*
That is (redux): Random, meaningful, and "of no value to traders" are different things. Brain waves aren't *random* noise: the patterns exist. They're probably meaningful, at least in the sense that "we have two eyes" is meaningful (it describes a true fact and some sort of reason for it can probably be derived) and possibly in a deeper sense as well (like the hypotheses, about how the brain might use them, which are outlined in the essay).
Are they of value to traders? ...I'm not sure about that, either; it was one of my first thoughts too. Stock market behavior can be accurately described but remain useless for trading; a trend can be real but ultimately unprofitable, if e.g. it's widely-known and priced-in already, or otherwise not specifically actionable.
Brain waves, similarly, may be a real thing that doesn't tell us much except "this thing happens". Much like, in the end, this entire comment is accurate (I hope), but ultimately says nothing except "yeah you're right I dunno either but I autistically object to some of the word choices in there".
Gems like the donkey-kong transistor vs the space-invaders transistor on the motorola 6502 (lesion the transistor, observe which behavior is impacted). Different switching rates and synchrony between chip regions across different behavior (switching transistor as analogue of firing neuron).
They didnt attach a real oscilloscope, but that would probably have uncovered beautiful chip waves, from the slow background 50 Hz (bleed from the power grid) to chip clock, memory clock, a variety of bus speeds, etc.
edit ps: speed of light is not so fast. 5 ghz means 6 cm, or about 3 x diameter of a chip. Brain analogue 50 m/s across 20 cm would be 80 Hz, so signal conduction in chips is quite slow compared brains, in relation to clock and distance.
pps. This always gets me when neuro people are like "the brain is so weird, it rolls back time and retcons stuff during saccades. an ai would never do this, bio sucks". Have you lived under a rock during spectre? Your phones cpu does this constantly. When you ask the cpu for its state it literally rolls back time and retcons a state it could have been in that is compatible with assembly level understanding. As the quip goes, the architectural state is architected.
TBF those waves in computer chips point to some pretty important aspects of how the chips work. I mean the clock signal is doing basically the time slicing that was one of the proposed uses for these waves and the waves in the hippocampus do sound alot like the patterns that an oscilloscope would read thanks to the rythym of memory update and fetch.
I think you mix up some concepts here. AIs run on computers, which use oscillations as part of their functioning. That does not mean that the oscillations are part of the AI algorithm or design, or even strictly necessary. "AIs" are just implemented on computers, but are relatively independent of the actual hardware implementation, and therefore should be regarded separatedly from any oscillation going on inside e.g. the CPU. The oscillations of the CPU have no effect on the computation inside a neural network. You could arbitrarily slow down the computations, oscillations and factors, and probably think of a CPU implementation that does not rely on oscillations and would still be able to perform the same computations than a neural network.
I agree that it's not a direct implication but TBF the clock synchronization is pretty essential to the programming of an artificial nueral net. You need to sum the inputs to each nueron at each layer so how do you know that you've accumulated all the inputs and are ready to sum them and go on?
Well it's a bit opaque but if we really get down into the weeds it depends on the apparently sequential nature of instructions being processed in a core and, in a parralel application, the ability to rely on certain atomic updates and reads. At a really low level those are possible because of the clock signal coordinating the movement of data around the chip.
Yes, there has been some research into clockless chips in certain applications but for the general case clocks offer some pretty clear benefits and may even be necessary.
Still I think the comparison is not a good fit. For the brain, the oscillations are engrained in the workings of the "algorithm" that is computed. The oscillations happen on a pretty high level, at a mostly global or regional scale. The oscillations adapt to each other, and this adaptation is part of the algorithm. For the computer, the oscillations are merely just supporting the algorithm. They happen at the most basic, low-level. There is no strictly necessary feedback loop between the highlevel functioning and the low level oscillations (disregarding downclocking to save energy, something entirely different). You could arbitrarily halt and continue computations on the CPU and would still obtain the same output results. This would not hold true for oscillations in the brain. If that is enough to claim that "there are oscilliations in current AIs", then you could probably also claim that "there are oscillations in a traffic light" and it would be similarly meaningful.
Before I respond I agree with you that there are some differences but I don't think they are as substantial as you suggest. And we don't really have any evidence of feedback between the two in the brain (there is no experiment where you change the rate of brain waves and produce different functional effects in the brain). I break this up into a few points below all of which are meant to suggest that we don't really have much evidence that brain waves are really all that different than what we see in a computer running a complex piece of code in response to external inputs.
1) I think the OP here is (or if they aren't I will) making the argument that we don't actually have that good a reason to think that these brain oscillations are happening at a "pretty high level". I think one point of the observation is exactly that if you looked at a microprocessor implementing any kind of complex program without knowledge of how it worked the most salient signals would be a number of wave like patterns from the clock signal to the patterns of memory access to display updates. In a certain sense the clock signal is low level but you would see a lot more such oscillations reflecting memory and data access patterns and I'm not seeing how that really differs so much from the brain case.
For instance, if it's an AI taking input from a camera running it through some kind of neural net, producing output and then updating the weights on that neural net you'd expect to see a number of different patterns of oscillation reflecting the different memory/transfer patterns for copying the data from the input, preprocessing it for input into the neural net, propogating the data through the neural net and then backpropogating to update weights. Indeed, I wouldn't be that surprised if you wouldn't see different patterns of oscillation depending on the fan out of the level of the net being processed at a given time. Is that the more high level kind of thing you were thinking about?
2) The reason I focused on clock signals is that I suspect that synchronization is a pretty universal feature any system needs to achieve and that this almost inevitably requires some kind of system wide signal to coordinate that and that's one potential explanation for some of the brain waves. In a processor you tend to get one global clock signal (though we are moving past that and allowing different parts of the system to be clocked at different rates) but it's not clear to me that's how the brain would do it so maybe that's all we are seeing in the brain too (just different low level coordination signals).
Even if you suppose the only point of brain waves is to allow for basic synchronization like clocks in a computer wouldn't you expect different processes in the brain to potentially use different synchronization rates just as you have a different clock signal for your memory bus, your CPU, your PCIE etc..
I admit this point is pretty speculative and I'm not arguing that it's the most likely explanation for what's happening only that I don't think we can exclude it so it's hard to say the two cases are so different.
The paper i linked applies standard neuroscience techiques, the ones that discover brainwaves and grandmother neurons and connectomes, to a pretty simple cpu running donkey kong and space invaders. It then describes the results and compares them to an actual understanding of the system.
As such, it is intended as a scathing methodolical critique of the field of neuroscience.
My narrower point in linking this paper was: Scott has no leg to stand on, when comparing presence vs absence of certain phenomena (wave/oscillation, grandmother neurons) in brains vs ai / motorola 6502 running donkeykong. This is because the same analytical toolkit, when applied to the system we dont understand (brain) and the system we do understand (motorola 6502) yield similar phenomena (wave like patterns, task specific transistors/ neurons, etc).
My wider point is that our understanding of the brain, and the "neuroscience toolkit", is inadequate to support the kind of speculation here, especially given that this toolkit fails so hard when validated on systems that we actually do understand.
Super interesting post but raises so many questions!
> there are other even longer-period brain oscillations of unclear purpose.
What are these??
I understand that when neurons fire, by default they're firing at like 100Hz and 'increases in activity' are upticks in the rate. What's the relationship between these frequencies and brain-wave ones? Are some waves at this high frequency, and some of them talking about pulses of raised frequency?
Action potentials, the mechanism by which neurons activate, are not unique to neurons, so wave-like synchronous activity between cells could very well exist in animals without brains (and plants).
Hey all. - I'm a working neuroscientist who's spent some time thinking through some of these issues.
I would say that there's a significant proportion of the field that is, shall we say, "skeptical" of the Buzsaki-style view on oscillations. Everyone agrees that the oscillations exist, and that changes in the oscillations are often correlated with changes in brain state, but making *causal* statements about their impact has proven much more difficult. There are certainly some suggestive data in the hippocampus pointing towards the importance of things like theta-band power and sharp wave ripples, but even there you get inconvenient data points like the fact that bats have a completely normal, functioning hippocampus with no apparent theta band oscillations: https://www.sciencedirect.com/science/article/pii/S0092867418312297
More broadly, it's hard to shake the feeling that the brain has some natural resonances for various states - sleep/unconsciousness is low frequency, low activity is ~alpha/beta, high activity is ~gamma. A statement like:
"Ask someone to think about a certain topic, and cells representing that topic will form a neuronal assembly (ie they will start oscillating together at the same frequency) in gamma rhythm."
is really just saying that those cells are active, and neurons in an active brain area tends to follow a gamma rhythm. In fact, I'm reasonably confident that if you were to look at the internal activity of cells in the same area that do NOT represent that topic, they would also be following the same gamma rhythm, just with less tonic input and thus less spiking activity.
People like to study oscillations because they're brain-wide phenomena that are (relatively) easy to detect in humans with non-invasive methods like EEG. However, anyone who's looked at neural activity using more finely-tuned, invasive methods (e.g. single neuron electrophysiology) knows that individual neurons close by each other can be tuned to very different sets of variables, and that when you average signals from large populations of neurons you lose a lot of useful information.
To be clear, I think everyone in the field agrees that neurons act together in assemblies, that patterns of activity that are shared across neural populations are important, and that sometimes these population patterns of activity can oscillate. The problem is that there are many, many, many of these subnetworks active at any given time, and that when you average them all together you get a signal that's easy to measure, but is limited in its ability to explain anything beyond basic brain function. You have 86 billion neurons in your brain. You should be skeptical of theories that tell you just 4-5 frequency bands are the secret to understanding it.
> Everyone agrees that the oscillations exist, and that changes in the oscillations are often correlated with changes in brain state, but making *causal* statements about their impact has proven much more difficult.
One argument I heard against this that struck me went along the lines of: "People have this gripe about oscillations being causal and many intuitively see oscillations as epiphenomenal. However, the same people take, at face value, that increases in brain activity has causal effects. Why is this? There is no stronger philosophic argument for activations rather rather than oscillations."
I think it's just application of a reductionist principle in this case. Do oscillations tell you anything you wouldn't know from simply averaging spike trains? If not, then I think we should prefer the spikes.
I think the pro-oscillation account is that yes they do. It would argue that much information can be gleaned by considering how those spike trains give rise to oscillations. They would argue that there are several meaningful levels of analysis (e.g., neurotransmitters within a synapse, individual neuron firing, etc.), and oscillations are one of those levels.
Can you provide an example of information that can be gleaned by considering how spike trains give rise to oscillations?
Here's my intuition for why spikes are more meaningful than e.g. neurotransmitters for cognition. I can stimulate a brain region electrically or optogenetically, thereby by-passing incident neurotransmitters, and drive a behavior. I don't think it's even possible in principle to do that with oscillations (i.e. by-pass the firing of single neurons).
That's a good point. I think in principle this is somewhat testable, by seeing whether the effect of optogenetic stimulation depends on whether the stimulation gives rise to an oscillation. I don't know enough about oscillations to cite anything though.
I have no clue what it would even mean to cause an oscillation without bypassing the firing of individual neurons, unless you just rewire them somehow (I'm also not a single-neuron researcher).
So, it's relatively easy to show how activity in motor areas causally leads to changes in muscle activity / movement. And, more generally, if Neuron A is synaptically connected to Neuron B, we have a very rich understanding of the electrical and molecular mechanisms that explain why an increase in activity of Neuron A causally increases the probability of activity in Neuron B. In so far as a mind is the product of 100 trillion neural connections, we know that each of those individual connections represents a causal chain that propagates increases in brain activity. There's lots of other very suggestive evidence, but these basic facts are, I think, why most neuroscientists feel comfortable hanging their hats on the idea that changes in behavior are tied to changes in neural activity somewhere.
In contrast, the structures of neural oscillations are (in most cases) set by emergent properties of densely connected neural networks. There's been amazing progress in trying to model these kinds of networks, but in truth we're still in the pre-statistical-mechanics era of neuroscience. We know a lot about individual neurons and neural connections, but we still don't have a well-accepted theoretical framework that allows us to understand how large populations of neurons interact and compute. This ignorance makes it hard to make confident statements about the causal impact of oscillations, or really of any other network-level phenomenon.
The philosophic argument for neural activation is that we understand its causality in very simple networks. The argument against oscillations is simply a lack of evidence, pending better models of network activity.
Thank you for writing this so that I didn't have to.
One thing I might add: it's very difficult for me even to imagine how a network of interconnected neurons, some of which are spontaneously active, could fail to give rise to oscillations. This to me is the best reason to think that EEG signals don't have much explanatory power over and above the neuronal spikes upon which they supervene.
One last point of clarification that may be helpful to non-neuroscientists: it may be useful to distinguish "brain rhythms" at different levels of analysis. I can imagine someone reading this and thinking that hippocampal theta is pretty much like visual cortex event-related potentials (ERPs). These are completely different phenomena measured using completely different methods! Hippocampal theta is measured as a local field potential (LFP), which usually means actually sticking electrodes in the brain and then low-pass filtering the signal. One step up from that is intra-cranial EEG, in which electrodes are placed on the surface of the cortex (in neurosurgical contexts). And finally you have good old-fashioned EEG, where you wear a cap over the head, like in a psychology lab, and measure electrical signals that originate over many centimeters and are difficult/impossible to conclusively localize. This is where people see ERPs.
I'm not a reductionist at all costs by any means, but generally I think it's safe to say that the actual causal relevance of each of these "rhythms" decreases as you ascend the hierarchy. Hippocampal LFP stands out as potentially interesting because of phenomena like phase precession of place cells, but Justin L aptly points out that this is by no means a necessary feature of place coding in general.
Yeah this seems like taking a 50 megapixel image and collapsing it down to a color frequency histogram. You’d get very interesting correlations! You might get very good at picking out trees from fire trucks and find all kinds of insights, but it doesn’t change the fact that you have thrown out 99.99% of the useful information and are looking at an Nth order artefact and trying to reason about it instead of the real thing.
Brain waves are what we can map, so spend enough time looking at them and you can talk yourself into thinking they are the territory.
Can you explain this in more detail? or provide a cite? I would guess the opposite because the ratio of consecutive terms of the Fibonacci sequence converges to φ.
Every real number can be approximated by an ever-closer series of rationals, since the rationals are dense in the real numbers (e.g. the square root of two is approximated by the fractions 1/1, 14/10, 141/100, 1414/1000, 14142/10000, ...). The question is how good of an approximation you can get for rational numbers with denominators up to some bound.
This turns out to be related to the size of the terms in the continued fraction expansion of a number - larger terms in the continued fraction expansion yield better approximations of your chosen value. (The unusually-good-for-its-complexity rational approximation of pi as 355/113 exists because of an unusually large term early in the continued fraction expansion of pi as [3; 7, 15, 1, 292, 1, 1, 1, 2, 1, 3, ...]. If we round the 1/292 term portion of the continued fraction to zero, we get 3+1/(7+1/(15+1/1)) = 355/113.)
The golden ratio's continued fraction expansion is all 1's, so in this sense it has the worst rational approximations of any real number. In practice, people tend to care about the asymptotics of this approximation quality, which leads to the notion of https://mathworld.wolfram.com/IrrationalityMeasure.html. More concrete implications between the terms of the continued fraction and the badness of rational approximations in this sense are mentioned at this MathOverflow thread: https://mathoverflow.net/questions/89600/numbers-with-known-irrationality-measures
But that's actually saying phi is as easy to approximate as an irrational number can be. As an irrational algebraic number it has mu = 2 while transcendental numbers have all sorts of higher values.
Higher irrationality measure implies *better* rational approximations. The reason it's called an irrationality measure is that rational numbers are especially badly approximated by any rational numbers other than themselves, so having good approximations makes you less like a rational number in that regard.
My bad...I just assumed that it couldn't be that algebraic numbers were harder to approximate than transcendental ones. Still find that hard to believe.
Still, nothing special about psi...no easier than any other irrational algebraic.
Thanks for all the responses. I was a former math major but somehow missed out on continued fractions and irrationality measures.
Another possible analogy - the musical interval between 2 notes has its consonance/dissonance related to having small integer ratios, or degree of resonance (attributed to Pythagoras). So the "most dissonant interval" would have the most anti-resonant ratio. I will have to try a bit of research comparing what the ratio φ sounds like.
A book on continued fractions is a good place to learn more. Continued fractions give the sequence of best rational approximations to any given real number and it turns out that the continued fraction expansion of the golden ratio is [1, 1, 1, 1, 1, 1...]. This is what drives the result, which I believe is called Hurwitz’s Theorem.
That doesn't seem to be true. See discussion below but it has irrationality measure of 2 like all the other irrational algebraic numbers while lots of transcendental numbers like pi or ln(2) are way worse.
And why should that make it best for anti-resonance anyway? I can vaguely see why it might matter but don't really understand what you mean well enough to say.
As mentioned above though, pi is easily approximated as 355/113 even though it's transcendental. Here, "easily" means the error is small relative to the complexity of the denominator. For comparison, a similar-sized approximation of phi, 233/144, has an error 40 times greater.
Resonance is greatest when the ratio is close to one with a small denominator, so a ratio that's far from any small denominator will avoid resonance.
If anyone is interested in learning more about the Fourier Transform (IMO one of the most beautiful concepts in mathematics), there's an incredible Stanford course on it: https://see.stanford.edu/course/ee261
I actually had a big "aha!" moment while meditating during the months I was taking the course. It's hard to describe, but it was like suddenly seeing my internal rhythms in the frequency domain instead of the time domain.
> The hippocampus is traditionally associated with memory, but Buzsaki cites recent research showing that it also handles navigation (eg a rat navigating a maze, or a person walking through a familiar city). Why this combination of tasks? Rhythms sort of suggests that brain areas are less about specific tasks than about specific graph-theoretic arrangements, which are convenient for specific algorithms, which are convenient for specific tasks. The neural connections in the hippocampus are mostly local in a way that makes it a good “auto-associater”, ie you give it part of a pattern and it can recognize which pattern is being suggested. For complicated reasons this works well for both spatial tasks and memory. In fact, Buzsaki sketches out a way that episodic memory might have evolved from our navigation system, where eg remembering a list is equivalent to remembering a path through space, and remembering a specific episode is equivalent to remembering a landmark (he doesn’t specifically say so, but this is a fascinating match for the method of loci mnemonic device).
Also, changing location can make you forget things, the "doorway effect" .
According to Wikipedia, pink noise is defined by the power *per frequency interval* being inversely proportional to the frequency - in other words, power *over time* is constant, independent of frequency. I don't see what's supposed to be mysterious about that, though. It makes perfect sense to me that the brain would want to maintain a consistent power draw, and would attenuate the amplitude of high-frequency signals to make that happen.
Right...and if you consider the fact that there is probably something like a per neuron limit on power output per unit time (can fire once every blah Ms) this almost seems inevitable.
I think the only examples I will remember are the fake toy ones, because they are simple and memorable. And soon I will forget that they are fake and remember them as real. Well done!
> Brain waves provide “synchrony”, allowing a smallest granular unit of time and essentially converting life into a turn-based game. Suppose that a snake bites your foot. You see the snake with your eyes, and also get a pain signal from your foot. The pain signal has to travel a long way, nerves have conduction delays, and so it reaches your brain well after the visual signal. But your brain needs to be able to combine the visual and pain signals into a single story (snake bit my foot). Brain waves separate experience into short granular “turns” so that the brain can attribute both stimuli to the same “turn” and connect them.
I'm having a hard time getting past the example. It's true that the brain needs a way of combining the visual signal and the pain signal into a single story. But there is quite obviously no need for synchrony or a minimum quantum of time at any point in this process; if I feel pain and then, five full minutes later, see a snake slither away from where I was standing when I got hurt, I'll still combine those two signals into "a snake bit my foot" (where that combination is plausible).
It may be necessary for the sense of the whole self for the brain to integrate different stimuli in that way.
I seem to remember a syndrome where a person doesn’t feel as if they are really in their body, but rather that the body is a thing they are operating sort of by remote control, like a marionette. Probably unrelated like all just-so stories, but I can easily imagine that a delay between seeing your foot get hurt and actually feeling the pain could be part of that.
That said, we’ve probably all had the experience of doing something and realizing, “Oh, that’s going to hurt tomorrow.”
Delays that need to be combined into a coherent whole is the entirety of multiplayer game network coding. People clicking buttons, sending them with variable delays along with a "happened at time t" attached to a server which then has to retroactively figure out who hit and who missed and then send that back in time. Ask game network engineers for a books worth of problems and solutions over the last 40 years :)
Games also use frames, discreet moments in time, to make physics and simulations doable at all. Too low a framerate and you notice the choppiness. Also if your monitor frequency and game frame frequency are nice multiples, it looks good, if not, it's choppy as well. Many analogues!
Why not just say that brain waves are emergent? They don’t do anything in particular. They are just the result of something else that does something. So everything here will only be based on correlation, not causation.
One thing that helped me understand EEG better, and I have seen other commenters mention is that there is a big difference between what the neurons are actually doing and why and what we pick up from the EEG with electrodes on your scalp. Each individual neuron fires and makes a tiny electromagnetic pulse sure, but if all the neurons were aligned randomly all these pulses would sum to zero - no signal. If the neurons all fired at random times they would sum to zero - no signal. If it's a small group of neurons it's too weak to make it to the scalp and overcome noise - no signal. So we only 'hear' large groups of aligned, synchronised neurons. So any sort of brain activity that doesn't meet these requirements is invisible to us unless we use electrodes to record the activity of single neurons. We also can't distinguish between synchronous activity being 'used' by the brain for a purpose (like synchronising areas of the brain), with byproducts of activity (like to actually move a muscle some amplification of a small signal has to take place to have a bunch of neurons fire at the same time).
This helps to explain the pink noise thing - larger groups of aligned neurons make a stronger signal, but neurons are slow so unless these areas are oscillating slowly its hard to get large areas to line up. So stronger signals tend to be lower frequency, average that out across millions of areas of different sizes and you get something like the pink noise curve.
I also can't stress enough how terrible a signal EEG is - it's the worst biosignal by far. Brain activity is already basically random noise with slightly different non-random noise overlaid in it, then its smeared out by volume conduction as it moves through the head, then it picks up every bit of environmental noise and muscular activity, then the electrodes and cables move and skin conductance changes, then its all screwed up by referencing and grounding effects. My research was in brain computer interfaces and my advice is stay well away until either portable MEG systems or intra-cortical electrodes go mainstream.
It'a even more complicated than this since depending on the frequency the source of extracellular potentials is probably not mostly from spiking activity, but from membrane voltages associated with synaptic inputs and intrinsic electrical signaling. In other words it might not be the neural _output_ that is aligned or not aligned, but the inputs and single neuron transformation of the inputs towards spiking outputs that one mostly picks up extracellularly. This is (as a general rule) mostly going to be true for anything in lower frequencies, certainly theta. For scalp recordings in particular (ie. not for electrodes that actually go into the brain) I'd argue that the signal is probably mostly made up of non-spiking sources, for any frequency measured.
Not too surprisingly, awareness appears to correlate more with overall signal diversity than specific wave frequencies:
"The foregoing studies complement an older, yet largely unacknowledged, body of literature that has documented awake, conscious patients with high amplitude delta oscillations in clinical reports. . . a largely parallel body of recent work has reported convincing evidence that the complexity or entropy of EEG and magnetoencephalogram or MEG signals strongly relates to an individual’s level of consciousness"
Hyper-synchronous activity correlates with unconscious states like deep sleep and generalized seizures.
In Conway's Game of Life, people have made some patterns that act as computers. For example there's this pattern that calculates and prints the digits of π: https://conwaylife.com/wiki/Pi_calculator. Here's a plot of how many live cells it has over the first 8192000 generations: https://i.imgur.com/KyICAc0.png.
In an analogy to the origin of the names of white, pink, and brown noise, the laboratory technique to separate and study DNA of different lengths is named southern blotting for its developer, Ed Southern. The similar method of separating RNA is called northern blotting, and for protein it's called western blotting. Once, I accidentally reversed the electrodes when doing a western blot, so that the protein ended up lost in the buffer instead of on the blot paper, which I guess could be an eastern blot. But apparently there are many things called eastern blots now: https://en.wikipedia.org/wiki/Eastern_blot
I feel the speculation at the end doesn't take sufficient notice of the fact that we are observing the system from the inside. For instance, whether or not time *feels* discrete in meditation has no real necessary relation to how the brain actually processes events. You could easily imagine a brain which processes events in discrete blocks but where this blockiness was imperceptible to self-reflection. Alternatively, the brain could just process experiences as they came in so there was no discreteness but then, when you try to reflect on the experiences, batch them together in a way that feels discrete.
More generally, I don't like these grand attempts to explain meditation/enlightenment as they all seem to suffer from this kind of importance bias: since these states feel so important they must reflect important brain processes. But isn't the most likely explanation just that the brain is quite plastic and it shouldn't be that surprising that you can train/trick it into producing pleasureabld experiences?
But those experiences need not be caused by grand things like linking usually unconnected regions...but most likely by the same kind of fuckups and errors that drugs often work by just without the direct chemical intervention (so less vulnerable to tolerance).
> But isn't the most likely explanation just that the brain is quite plastic and it shouldn't be that surprising that you can train/trick it into producing pleasureabld experiences?
I suppose the counter argument to that would be that if many masters of meditation describe their most profound sensation ("enlightenment") as essentially the same (insofar as an inherently individual experience can be sufficiently communicated and compared), maybe there really is some fundamental, neurological basis for that? It sure does sound interesting enough to investigate further. So many people across so many centuries deluding themselves in basically the same way through meditation would be quite the coincidence, the "no true scotsman" fallacy notwithstanding.
> Pink noise is apparently omnipresent in natural systems for kind of mysterious reasons - see eg this Quanta article, which says pink noise “is found in all kinds of electrical noise, stock market activity, biological rhythms, and even pieces of music — and no one [knows] why.”
> The take-home message of this rant will be that the universe counts in powers rather than linear progressions, and thus in most cases a power law is not so much surprising as it is overwhelmingly expected
Also, a fun book about the key role of synchronization of oscillations throughout nature is Sync, by Steven Strogatz. I really enjoyed it a few years back, but it's not really my area, so if anyone who's more knowledgeable about this has some criticism, I'd love to hear it!
Tbf the message i got from parsing the rant is that the power law in itself is not surprising, it is the power law exponent that is interesting (as usual in critical system)
Two weeks back I had lucid dreaming experience. I knew that I was dreaming while in a dream and I felt my sense of self. The moment I woke up after that dream I was left with lots of energy to carry out the rest of the day. Now what state of consciousness was I at? In dream state. Did I lose my sense of self? No, because I was conscious of the fact that I was dreaming. Probably a dopamine surge followed up the dream. The feeling in the dream was more like decoupling from time or freeing from time such that I was able to easily float around space. What existed was the sense of self observing the dream environment. It was around 5-minute trip in wakeful state whereas what I experienced in the dream was actually lesser than that (This I am making up because I do not have a sense of time in the dream, yet it was till the end, and I was continuously processing the information. But when I compare with the events in the wakeful state, I end up processing very few information in that dream state which is opposed to what I felt because it felt as if I have been for more than an hour in the dream). That is, I was in the midst of listening to some 30-minutes Yoga Nidra script. How is this synchronization of sense of self or time related with states of consciousness? I am conscious of these events in wakeful state. Is it just an approximation I am coming up with!
I'd love to see what an AI set to analysing EEG recordings using Fourier analysis or whatever algorithm it decides to make up would say about them. Just as AIs embarrassingly deduce the race of CXR subjects, they might surprise us with foreseeable things (which aren't always so clear to human eyes) becoming vastly more reliable, like a diagnosis of epilepsy, but it's not beyond the bounds of possibility they could be quite good at some surprising things: ADHD, autism, Alzheimer's. We can't give them training data for that as yet, as we are not aware of EEG findings in those three examples. But if we quietly set an AI to do its thing and later supplied diagnoses as they were made, the patterns, if any, could be found and quickly refined to make a predictive diagnosis for each.
The quote "the natural logarithm 2.17" must be a typo, because he published a paper just a few years before the book presenting the same argument but with the correct value of e https://doi.org/10.1016/S1472-9288(03)00007-4
One of my big insights studying creative writing (and art in general) is that practically everything is vibratory / enacted by oscillation. Stories oscillate along a bunch of dimensions, and will feel "flat" if constrained. Imagine explaining the shape of a working metronome without using time: the hand is to the left and the hand is to the right. It's a paradox, until you introduce the time dimension for the two contradictory states to oscillate along. So whenever you find a paradox, it probably represents an oscillation along some dimension you haven't discovered yet. The activity of the first act of a story is to establish these paradoxes (or "tensions") which themselves establish the dimensionality of the story. They make it feel "round" instead of "flat".
A good example is Pride and Prejudice, which introduces three paradoxes (at least) in the first sentence.
It's fascinating how much this description of the brain sounds so much like one of the best ways I've found to describe an excellent novel. Like, writers talk about "resonance" between separate parts of the story, and it's considered kind of a mysterious, undefinable thing. When it happens, it feels much like how this post describes different parts of the brain tuning to each other's frequency.
I had their "Conga" stuck in my head, on and off, for literally *decades*, without knowing what it was. Finally last year I stumbled across it on YouTube, and now I am fully recovered.
I just want to put here this Asimov quote from 1950 'cos I think it's delicious (no spoilers, I think; it's from the Foundation "trilogy").
“The science of electroencephalography was at once new and old. It was old in the sense that the knowledge of the microcurrents generated by nerve cells of living beings belonged to that immense category of human knowledge whose origin was completely lost. It was knowledge that stretched back as far as the earliest remnants of human history—
And yet it was new, too. The fact of the existence of microcurrents slumbered through the tens of thousands of years of Galactic Empire as one of those vivid and whimsical, but quite useless, items of human knowledge. Some had attempted to form classifications of waves into waking and sleeping, calm and excited, well and ill—but even the broadest conceptions had had their hordes of vitiating exceptions.
Others had tried to show the existence of brainwave groups, analogous to the well-known blood groups, and to show that external environment was not the defining factor. These were the race-minded people who claimed that humanity could be divided into subspecies. But such a philosophy could make no headway against the overwhelming ecumenical drive involved in the fact of Galactic Empire—one political unit covering twenty million stellar systems, involving all of Man from the central world of Trantor—now a gorgeous and impossible memory of the great past—to the loneliest asteroid on the periphery. And then again, in a society given over, as that of the First Empire was, to the physical sciences and inanimate technology, there was a vague but mighty sociological push away from the study of the mind. It was less respectable because less immediately useful; and it was poorly financed since it was less profitable.
After the disintegration of the First Empire, there came the fragmentation of organized science, back, back—past even the fundamentals of atomic power into the chemical power of coal and oil. The one exception to this, of course, was the First Foundation where the spark of science, revitalized and grown more intense, was maintained and fed to flame. Yet there, too, it was the physical that ruled, and the brain, except for surgery, was neglected ground.
Hari Seldon was the first to express what afterwards came to be accepted as truth.
“Neural microcurrents,” he once said, “carry within them the spark of every varying impulse and response, conscious and unconscious. The brainwaves recorded on neatly squared paper in trembling peaks and troughs are the mirrors of the combined thought-pulses of billions of cells. Theoretically, analysis should reveal the thoughts and emotions of the subject, to the last and least. Differences should be detected that are due not only to gross physical defects, inherited or acquired, but also to shifting states of emotion, to advancing education and experience, even to something as subtle as a change in the subject’s philosophy of life.”
But even Seldon could approach no further than speculation.
And now for fifty years, the men of the First Foundation had been tearing at that incredibly vast and complicated storehouse of new knowledge. The approach, naturally, was made through new techniques—as, for example, the use of electrodes at skull sutures by a newly developed means which enabled contact to be made directly with the gray cells, without even the necessity of shaving a patch of skull. And then there was a recording device which automatically recorded the brainwave data as an overall total, and as separate functions of six independent variables.”
Part of the difficulty with doing any kind of direct investigation of EEG tracings is that the EEG is a tremendously distorted view of what the EM fields inside the brain are actually doing. Necessarily: you're recording an electric field at the surface of a pretty conductive material. You'd be much better off using SQUIDs to detect the magnetic fields, because that would give you a relatively undistorted picture of the 3D pattern of EM fields, and I would guess the wavevector structure (i.e. the result of a Fourier analysis in space) would be at least as important as the frequency structure (the result of the Fourier analysis in time). That's expensive, though.
> Gamma was the rhythm I found most interesting - it seems to map very well to “items in conscious awareness”. Ask someone to think about a certain topic, and cells representing that topic will form a neuronal assembly (ie they will start oscillating together at the same frequency) in gamma rhythm. I was pretty surprised that there was this clear of a neural correlate of consciousness.
I'm interested to know if anyone has taken e.g. Daniel Ingram or other expert meditatiors and tried to measure correlates to Jhana and other rarified meditative experiences. As the progression of experiences in Jhana are very clearly defined (at least in some aspects), measuring the difference in activity at each Jhana should be possible at some resolution of instrument.
Coming from another angle, when achiving single-pointedness in meditation, conscious awareness rests entirely on one object of attention, without any scattering or wavering. Presumably there's some strong Gamma there? But with extremely focused attention, are the Gamma oscillations localized in some way, relative to the object of attention that is being focused on? Would we see different activation patterns when focusing single-pointedly on different mental objects?
(The key thing here being, in most minds, attention scatters and wanders from second to second, so these sorts of localization would be impossible to do if the oscillations are on the order of Hz. But single-pointedness seems a quite unusually-stable state, which might make it amenable to objectively studying qualitative/subjective experience.)
I imagine this is something https://theeprc.org/ is into, but I don't know the meditation-neuroscience literature well.
A final take on this; in the formulation of The Mind Illuminated, we have various hinderances such as Dullness and Distraction. These should be measurable in some way; can we detect Dullness from the outside? (For example it might be the percent of Theta waves increases as Dullness is perceived? Can this help a meditator to get feedback on different states? It's said that some get fooled into thinking Dullness is a more advanced state like Jhana; can we objectively detect Dullness to help meditators avoid this confusion? Particularly Subtle Dullness, which can sneak up on you?)
It's an intriguing idea that navigational memory underlays episodic memory. To build on your reference to the method of loci, I guess one explanation for the reason our declarative memory sucks is that we are using a system that was designed for navigation for something that has nothing to do with navigation. It makes intuitive sense: it's hard to remember abstract facts because they do not look like anything, and they are not located anywhere. I had learned that the method of loci works because humans have a *better* navigational memory system, but this is a different explanation: all memory is at root navigational, and it's not optimized for facts unless hacked by picture things in palaces.
For more on this, I recommend The Embodied Mind. It is full of gems like: people can think more clearly when they move their hands, or, people think Chile is south of British Columbia because of something to do with how we assemble objects in our heads. Movement came first, and thought, emotion, and language bare the imprint of this legacy. The book is badly written, but the examples are great.
Apropos, a musician I play with who also happens to be a neuroscientist just published this paper on synchronous states invoked by psychedelics. He'll be presenting at the Society for Neuroscience conference in SD, CA next month. The basic gist is that there seems to be some sort of ~150Hz standing wave of activity that appears widely in overall cognitive processing centers when (a rat) is on LSD/etc.
Oh boy, I love noise. From a physics, electrical engineering point of view; White noise is also called Johnson noise, it is the thermal noise inherent in all resistors*. And first reported by Johnson at Bell labs. https://en.wikipedia.org/wiki/Johnson%E2%80%93Nyquist_noise#History. (It has a flat frequency spectrum (white) at least for low frequencies... it's a classical effect. It breaks down at higher frequencies where quantum effects have to be accounted for.) Brownian noise is the noise of Brownian motion and is understood to be the integral of white noise. https://en.wikipedia.org/wiki/Brownian_noise. It has a spectrum that goes as 1/f^2 (where f is the frequency.) Both white noise and brown noise are fairly well understood from a theoretical perspective. Pink noise is called 1/f (one over f) noise in the physics community. It has a power spectrum that goes as 1/f (not surprisingly) It is not at all well understood theoretically... which is why we find it interesting.
Speaking of machine learning and brain waves, I am surprised not to find any direct mentions of positional encoding; see e.g. «Attention is all you need» arXiv:1706.03762 where they literally say that to make relative word positions easier for the network to use, multiple dimensions of sine waves of different frequencies — independent of the real input — are added as input features.
"If brain waves are really responsible for things like attention and cross-talk between brain regions, then might a lack of brain-wave-equivalents make AIs worse at these things"
Has anyone looked to see if there emergent oscillatory phenomena across the layers of really deep networks? Might me an interesting weekend project.
"Suppose that - and again, this is a fake toy example - your unconscious is oscillating at 1 hertz and your default mode network is oscillating at 2.17 hertz. Then you meditate, you slow your default mode network to 2 hertz, all of a sudden consciousness and unconsciousness are nice multiples of each other, they reach oscillatory synchrony, and they achieve higher levels of communication with each other."
Wondering could external sound waves/rhythms with the right frequency cause something similar? Thinking about shamanic drumming causing trance states. I think I read somewhere that strongly rhythmic music with the correct frequency can cause heart-rate to entrain.
The below is a bit of an axe that I grind, but I think that even just using the word/concept "force," in relation to enlightenment will (surprisingly) cause some non-zero number of meditators to really, really hurt themselves; and I think some larger number of meditators will get sent down a rabbit hole that won't get them to where they want to go:
Scott may not mean "force" in some effortful, phenomenological, qualia-feel sense, but some people will take it that way. Sometimes the experience of "force" or "forcing" does come up in meditation. But, I believe that it's bad to "add force" to one's meditation practice.
Because, say that mind-state-space has a lot of local minima (minima bc lower is better, in this analogy). Some local minima look, at first, kinda like enlightenment (though with initially subtle and then unfolding downsides), and those local minima *can* be reached by forcing.
But, forcing sort of has "inertial momentum" ("karma"!), and many hours of forcing can take a long time to undo (thousands of hours). And, meanwhile this remaining forcing "persists," eventually running amok, causing lots of psychological, phenomenological, behavioral, and conceptual problems.
"Actual enlightenment," or whatever, is, say, at the "true global minimum" and it can only be reached by a preponderance of non-forcing and undoing latent or actualized forcing that's currently in the system.
Enlightenment state transitions can seem abrupt, and can happen spontaneously, but for most people, those "stateless states" are only going to be complete and stable after *thousands* of hours of (paradoxical) "work"--which involves tacking towards effortlessness.
(***Cryptic and prolix, understatement, but more here: https://meditationbook.page/ (Ctrl/command-f on "force" might be interesting.))
I was surprised by the discussion of AI at the end. You seemed to reach a conclusion that the fact that large regions of the brain having synchronized brain waves was either a necessary precondition of consciousness or a strong correlate of it, then moved on to say that because AI would probably not have brain waves in this sense they would be less likely to have consciousness.
My intuition is that this is backward. It seems like computationally different brain regions operating at the same frequency means either that they can communicate with each other efficiently or that they are updating on the same internal clock. But for most ML models all neurons are updating on the same internal clock and can communicate with one another with nearly perfect efficiency (i.e. little signal is being lost because a neuron is not "close to activation) It seems like AI's are more likely than human brains to have most neurons all simultaneously communicating with one another on the same internal clock.
A potential flaw in my understanding here is that in AI's neurons not firing because they are not "close to activation" sounds like a neuron that uses the activation function max(0, input). If it is currently receiving a very negative input then changing that slightly via some other neuron will not cause it to activate whereas if its input is close to zero or positive changing it slightly will change the output. Is there a proposal here somewhere that having some neurons in an ML system activate at say every 3 activations of the rest of the network and just average the inputs between activations could cause more interesting behavior or something?
"Tune in, turn on, drop out" was a phrase evidently coined by Marshall McLuhan and certainly popularized by Timothy Leary in 1966 to encourage people to use psychedelics. If the frequency synchronization idea mentioned here is correct, the expression is remarkably prescient. Perhaps people on psychedelics literally tune in to normally inaccessible parts of their brain activity by getting them to the same frequency.
Sometimes at 3:45 in the morning, as the party’s closing up and you are well tired, if you’re staring at a white wall you can perceive a flicker. I believe that’s alpha, not waiting for you to close your eyes for once.
If you’re not going to that kind of party much anymore, there’s also a few optical illusions that flicker around the 10hz range. The most familiar of these is the wagon-wheel effect. So just look at a car.
That’s probably as close to watching your own brain at work as it gets in everyday situations.
An idea I've toyed around with is the idea of "blockchain protocols as the clock rhythm for distributed consciousness"
Like, you've got a bunch of computers running the same algorithm. Every so often, a new block gets pushed to the chain, and becomes common knowledge. This constitutes one "frame" of distributed consciousness, and contains some kind of consolidation or average of the input of all the constituent nodes.
Inside those nodes, you can have arbitrary subroutines operating in parallel wavelengths, but the outermost blockchain is like a [conductor | yearly census | drummer boy's marching beat | pulse | bible]
I have not read the whole thing yet but I am not too surprised by the pink noise distribution? White noise is constant energy per ± X Hz, pink noise is constant energy per ± X %. It doesn't feel any more surprising to me that the noise distribution should be such that the energy between 1 and 2 Hz is the same as the energy between 100 and 200 Hz (pink) rather than 100 and 101 Hz (white).
That's not particularly explanatory but consider that the energy above refers to the "energy in the noise waveform itself", not the energy required to produce the waveform. Suppose that the energy required for a neuron to oscillate at frequency f is proportional to f, then if every neuron on average uses the same amount of energy, the output energy spectral density is going to be proportional to 1/f, i.e., pink.
I thought we had scientific studies with brain blood-flow analysis, showing that meditation greatly curtails blood flow to a region which helps us locate ourselves (slightly above and behind our eyes). The theory was that both out of body experiences and the "one with everything" phenomenon were likely associated with this. The one time I meditated to the point of, "OMFG What just happened," I would put it more in the line of feeling everywhere than feeling at "one with everything." Maybe that was just the OMFG component though.
Extremely vague speculation, but the golden ratio is the most irrational number, the furthest from any even ratio, and it sounds like that could make a representation of it more consistent? People always talk about it in aesthetics, maybe it's related
It is! Frequency doublings maximize phase alignments but the golden ratio minimizes phase alignments. We discuss it in the neuroscience section of this paper on the concept of harmony.
I'm reading through the section on the hippocampus and how it functions in both navigation and memory. Does this connection have anything to do with how the "Memory Palace" mnemonic aid works? The one where you think of your house and assign a memory or a task to each room, and this helps you keep it all in mind later in the day.
Edit: Hey, guess what? It's mentioned in the very next sentence after where I stopped to write this comment.
Among all the oscillating brain waves, each sending and receiving signals to and from multiple recipients, it seems likely there are some neurons for which the sum of incoming signals moreorless somehow cancel each other out, eigen-neurons one might call them, for any given pattern of brainwave.
Think of waves falling on a rocky sea shore, with water swirling around the rock pools, and some will be quiescent at least for a moment, like the centres of miniature gyres. Maybe a memory is formed as a snapshot of the combination of quiescent neurons.
Someone once asked the ancient Greek sculptor Phidias how he carved such wonderfully realistic horses. "Easy", he replied "I start with a big lump of marble, and chip away everything that doesn't look like a horse". I suggest likewise that memories are formed somehow in sets of neurons whose lack of net inward, and thus any outward, signals means that they "look like" a memory, in the same sense as Phidias's comical reply!
"If we observed closely enough, we would find that experience is actually divided into individual moments of consciousness. These conscious “mind moments” occur one at a time, in much the same way that a motion picture film is actually divided into separate frames…"
There is a passage in Ramachandran's Phantoms in the Brain where he describes a patient with some sort of brain damage that perceived the world as a series of still frames, unable to stitch them together into a continuous experience (which made crossing the street, for instance, quite hazardous).
A couple of thoughts: (1) local cortical gamma oscillations are most commonly associated with attention. I think most neuroscientists would define consciousness as being the collection of global brain states that are different from what happens in anesthesia or damaged brains (or potentially sleep)
(2) I think that the jury is still out about whether oscillations are the signal that an invasive measurements of electrical activity detects when certain types of neurons/neural circuits are active, and all the bit about synchrony and chunking follows from the synaptic delays of the neurons involved. But I think that most neuroscientists expect this to be true. Historically folks who've studied the brain noninvasively with EEG for example imagine that the oscillations are meaningful by themselves, but that probably is just measurement bias.
(3) I've not done the experiment, but I suspect that a small antenna placed in or above a densely multicore processor would detect "local field potential oscillations" associated with different pieces of code running, and with enough resolution, associated with particular chip regions. Konrad Kording has a paper on this idea whose conclusions I don't love, but which is interesting in concept. From that perspective, there's nothing special about "wetware" except that the frequencies are a bit lower. Modern neuroscience has revealed that the old idea that a neuron was like a transistor was quite underestimating complexity. A neuron is probably more like a professor core. So it's not unlikely that some sort of machine learning/AI running on millions of cores would exhibit signals much like LFP...
>All of this hints at some deep connection between brain waves, consciousness, and selfhood.
Neuroscience is data rich, but theory poor. You often hear this lament from neuroscientists. What's even more unfortunate is that we actually do have a wonderful, mechanistically precise theory that explains how a specific type of brain waves give rise to consciousness (and selfhood).
Stephen Grossberg, a tragically under-appreciated pioneer in computational neuroscience, has over sixty five years put together a magnificent framework that explains how a limited set of constraints and computational principles can explain consciousness and why it is needed. His work is largely unknown even within neuroscience because he was very early (early 1960s) in realizing that the complex nonlinear dynamics of brains needed a different type of math to capture rhythmic activity (differential equations). He was doing theoretical neuroscience before the term neuroscience entered official usage. Later advances in techniques (fMRI being the worst offender) meant that it was easy to publish research by generating more data than making sense of it using a good theoretical framework.
Coming back to why we need brain waves for consciousness. Here’s a very short and lossy summary (more in the links below) . Please note that the theory itself is mathematical, mechanistically precise, and explains a ton of neuroscientific data. The following is an attempt to translate it into accessible plainspeak that avoids the “the grundicular region modulates output from the parashmoxular cortex” thicket you mention.
We cannot ask what consciousness is, without also asking *who* is experiencing this consciousness. That is, where does the self that experience come from? How do 86 billion neurons (37 trillion cells in the body, really) give rise to selfhood and is consciousness necessary for this? So to explain consciousness, we must explain the self, and to explain the self we must explain, what? Is there an infinite regress?
Grossberg’s Adaptive Resonance Theory helps us understand how to break this up by recasting one in terms of the other.
The Self is simply a constellation of experiences.
Consciousness is the constellation of past experiences experiencing the present, assimilating it to act and prepare for future opportunities.
We now merely have to explain how an experience is constructed. Here’s where the brain waves come in. We are actually incomprehensibly large beings with billions of neurons and trillions of cells with data in some large multiples of those numbers streaming in incessantly. How do we survive this onslaught and make sense of it? Meaning must be collectively manufactured by a very large number of these neurons coming together. Unfortunately, communication is expensive and long-range communication is even more expensive and we have a finite number of calories that we can use up. Synchronizing brain waves – synchrony across any oscillators for that matter - are the most energy efficient way to do large scale communication. This is also why synchrony appears so often in dynamic systems (stadium waves, fireflies, etc). They reorganize to take advantage of this energy efficiency. These brain waves help experiences in different parts of the network (brain) talk to each other if necessary to decide what to do with the sensory data that is streaming in. Note that this is merely data. It only becomes information after this cross-talk within the brain across experiences. This is what is often referred to as top-down expectations. Top-down expectations (experiences) talk to bottom-up inputs (sensory data) and arrive at an agreement through synchronizing brain waves. This is what we consciously experience as experience. Here’s the magical part. Once forged, this new experience becomes a part of the constellation of experiences we recognize as the self, changing us. No man crosses the same river twice. The constellation of experiences changes, changing the next experience.
I would strongly encourage anyone reading this to attempt reading Grossberg’s article [1] (where he puts it all together) or his book [2], which can be a little daunting.
[1] Towards solving the hard problem of consciousness: The varieties of brain resonances and the conscious experiences that they support
And if you are really averse to wading through grundicular regions and parashmoxular cortices, you can start with our book [3], where we offer a gentle introduction to all this.
[3] Journey of the Mind: How Thinking Emerged from Chaos
That was a great review! You really went to the heart of the matter. Some of your questions are also brought up in this recent article on “Resonance as a Design Strategy for AI and Social Robots.” If you are interested in the brain as an oscillating system, you may appreciate the effort we took to clarify key concepts.
And, “Harmony in Design: A Synthesis of Literature from Classical Philosophy, the Sciences, Economics, and Design” has a section on brainwaves. It reviews the evidence for, e.g., brain waves being structured like octaves. Which they weirdly are—so as to support phase amplitude coupling.
What do you call the way of thinking that looks for 'purpose' of phenomena, as if emergent phenomena were a mechaism put there as a tool towards some goal? I feel like this approach is like looking at the waves to study fish.
This makes me think about mechanisms of ECT. What if it’s less of a blah blah BDNF story and truly a matter of brain rebooting itself after the post ictal suppression on EEG?
Can confirm: synchronization of brainwaves seems to be what's going on in meditation - well, at least once I had that experience.
I didn't really know anything about brainwaves or -rhythms, but I back then described the experience in terms of things being out-of-phase and slowly getting in sync. More precisely, I got into the rare (for me) state when my attention wouldn't drift from the sensory meditation object virtually at all. Nevertheless, I noticed that there seemed to be small, sub-second gaps between noticing/experiencing the sense data, a sort of frame rate of stimuli. I kept my attention on that "rhythm" of attention and the gaps between, of ~ 5 Hz (that's the number I guesstimated, but hard to really tell) while it slowly was getting faster. Finally it locked into phase with my attention. The result was weird and pleasurable "liquid" compared to "granular" sensation of all stimuli, like I could take in the sense data at the rate it was coming in. Keeping with the frame rate analogy; if you can only render every nth frame, things look and feel janky, while now every frame was rendered.
It's been some years since that experience and I hadn't had it since (among other things, I'm sure I kept trying too hard to repeat it). That time it lasted some hours without effort, I guess just by its "momentum". My instructor's response to my description didn't really help me understand this and I felt I was seeing and describing things from such a different angle that no knowledge could be shared. This review got me interested again in meditation as it seems to give some framework that seems to match my experience. From here, it's possible for me to derive an explanation of a certain kind of meditation that is so simple and concise it can be expressed in a sentence without being too proverbial or abstract:
syncing your attention's frequency with the frequency (or it's multiple) of the object of your attention.
There is good evidence for its use in ADHD but not using home kits. Practitioners take a 20 10 quantitative EEG and from there create a program that targets certain wave bands
Do you have any good review/meta papers about effectivenes in ADHD? I always find the neurofeedback literature full of small sample sizes, lacking appropriate controls (sometimes even none at all) and small effect sizes, which makes me generally distrust the field. However, I'm also from a different domain, so I'm not familiar with the landmark studies and google scholar is only so useful for finding the best meta/review.
Unless it's localized to a small region (partial seizures), epilepsy tends to look like strong hypersynchronous activity. Large numbers of neurons firing in unison without the harmonics and complexity of typical conscious states.
What I'm interested in regarding synchronized brain waves are interferences. I suspect that consciousness is the ability of humans to coordinate brain functions in such a way that they "compose" higher order patterns of brain waves that interfere with each other, creating the brain wave equivalent of a moiré pattern. When we evolved conscious minds, we basically learned to create those patterns.
Do you have any particular evidence for this? Or does it just sound nice?
You don't state this as a hyper-speculative hypothesis, you say you "suspect" it, as if this is the most likely thing.
pure speculation
I would say that consciousness is a name we give to some examples, but not others, of a certain style of organization.
The relationship between "you" and "your consciousness" is precisely identical to the relationship between "Amazon" and "Jeff Bezos". That's what consciousness is.
Can you elaborate on what you think the relationship between Jeff bezos and Amazon is? Because your last claim doesn't make sense to me
Jeff Bezos is the official representative of Amazon, notionally responsible for everything it does (and for making the relevant decisions), and exerting actual control over a small but significant portion of that.
So your consciousness is a little man inside you that tells the rest of you what to do?
"Jeff Bezos" is not to "Amazon" what "feeling like a bat" is to a "bat."
This is incorrect.
Jeff Bezos experience's Amazon's supply chains as a conscious experience?
Sure, just as much as you experience the flow of blood in your upper arm as a conscious experience.
I do not feel everything that happens in my body, but I do feel it if blood vessels are damaged by, say, a bat. Bezos does not feel anything in his company outside his body, unless he has new implants I was unaware of.
Scott, you wrote in your review of MCTB that:
> Ingram notes that every sensation vibrates in and out of consciousness at a rate of between five and forty vibrations per second, sometimes speeding up or slowing down depending on your mental state. I’m a pathetic meditator and about as far from enlightenment as anybody in this world, but with enough focus even I have been able to confirm this to be true. And this is pretty close to the frequency of brain waves, which seems like a pretty interesting coincidence.
Can you go into any more detail about what it's like to experience sensations in this discrete way? Also, kinda off-topic, but what's up with the jhanas? If anyone can experience mega-bliss states with a few months of proper training, shouldn't that be, like, a major EA cause area?
Love the idea of EA helping people learn to trigger jhanas. I suspect what’s going on is a bunch of adults are unwilling to believe the jhanas are real.
I mean, you can also experience a mega bliss state by taking heroin. Not that they're at all the same (for one, I don't think jhanas destroy peoples lives through addictiveness) but what I'm trying to say is that humans have multiple options we already know about to achieve bliss like states, so is there a particular unique benefit to jhanas that would make then a higher priority than keeping people from dying?
This probably depends on your causal model. Does helping people feel good help them perform better, and thus make more money which can be donated to EA causes? Do people who feel better naturally want to donate more?
These pleasures can can be achieved without intense bodily harm or risk of addiction. If helping people induce harmless bliss states is only a terminal good, then sure it’s maybe not THAT important. But even then, when 50,000 Americans die each year from suicide, it seems like maybe this isn’t a bad place to start. Especially if the big barrier is simply creating the public knowledge that these states exist .
I agree. EA should be about reducing suffering, not just blindly preserving life. Anything that helps treat depression, or improves mental resilience, should be a huge EA cause.
As someone who's spent a couple thousand hours meditating (with amazing practical benefits) and never achieved a jhana who has many friends who have spent a couple thousand hours meditating (with amazing practical benefits) who have also never achieved a jhana, I think the premise that "anyone can experience mega-bliss states with a few months of proper training."
(I believe jhanas are a real brain phenomenon, I just think they're quite difficult to achieve.)
Aren't there different styles of meditation which lead to different results? Perhaps the approaches that you and your friends take aren't suited to jhana attainments?
Perhaps.
I've generally studied in the "pragmatic dharma" tradition, which includes both Culadasa's "Mind Illuminated" and Ingram's MCTB, and seem to be the place where most rationalists start. These works both claim (more or less) that jhana and even awakening can be reached relatively quickly, but my experience is that it is widely known in these communities that the central works overclaim heavily regarding typical timelines (while still delivering massive benefits).
Many teachers of more traditional approaches will say that basically nobody except dedicated monastics will achieve jhana, and they'll only achieve it after many years of study. (The pragmatic dharma "response" is that we have much better methods now, and I think there's something to this, but I still think the overall claim is too strong.)
So is it possible I'm wrong, and "anyone can experience mega-bliss states with a few months of proper training?" Sure it's possible. But if we narrow the claim slightly to "a few months of proper training with mo more than 400 hours training total, including instruction," I would bet strongly against this. A specificish bet (making it very specific is too much work since the bets here aren't runnable) I'd take is "I'd bet $1,000 against $100 that no currently existing method could get 75%+ of no-substantial-previous-experience meditators to jhana within 400 hours of total training."
I guess it's also worth noting that different traditions / teachers sometimes use the word "jhana" differently, but I'm allowing Luke's original description ("mega-bliss") to restrict the word to very highly concentrated states (which is how I'd naturally use it).
(Some of the most interesting relevant data is probably from Jeffrey Martin et al.'s Finders Course: https://finderscourse.com/. Their current claims is they have a short-protocol that gets a lot of people to something they're calling "Fundamental Wellbeing", but spending a bit of time looking at the paper, I don't think they're making any claims about mega-bliss / jhana?)
Is there any evidence that the positive feelings from jhanas have anything to do with depression and suicide? While it seems like it has an obvious connection, it also seems like depression is just as prevalent in people who meditate as people who don’t, and enlightened monks aren’t immune from it.
I doubt anyone has done this study.
I'm also willing to place a large bet that people who can reliably enter jhana states are much, much less likely to commit suicide.
That's almost certainly true, due to sampling bias. Suicide is overwhelmingly correlated with mental illness, whereas the prerequisites for achieving jhanas include strong discipline and mental stability, not exactly common among mentally ill.
It is possible that meditating can improve discipline and mental stability though.
The jhanas are ill-defined and probably not (yet) worthy of scientific study.
But meditation in general absolutely helps with depression and anxiety. Plenty of scientific evidence here, for a broad overview see https://www.health.harvard.edu/mind-and-mood/how-meditation-helps-with-depression
But I agree--it doesn't make you immune. And it certainly doesn't make you immune to sadness or fear! To quote a favorite song of mine: "you can't change the feeling, but you can change the feeling about the feeling"
Well, if the options that you have in mind are all like heroin and cocaine, some of the most cripplingly addictive and physically devastating drugs, not to mention highly illegal, then I'm sure that most people would say that a widely available alternative with milder side effects would present plenty of unique benefits, like making anybody who'd be able to turn it into a easy to use marketable product the richest man alive. Maybe this isn't exactly EA's main prerogative, but I'm sure that they wouldn't be too unhappy if it just happens to fall into their laps.
> so is there a particular unique benefit to jhanas that would make then a higher priority than keeping people from dying?
Jhana only requiers interanl input. Heroin is an external input. TV is an external input. Sexual stimulation is an external input. This blog is an external input. They could all bring you to the blissful state, the difference is internal vs external. When it comes from the internal, then it is always within reach. Externalities are ephemeral and not always within reach.
Binding frequency of consciousness. It was also mentioned in The Mind Illuminated with a footnote to "Selfless Insight: Zen and the Transformation of Consciousness", which counter to its hippie title is all about meditation experiences mapped to specific neurological processes with mechanistic explanations. Cannot explain it, have not read it yet, as I want to learn more neurology first. But it will have your answer. (or possibly in one of Austin's other two big books of hardcore meditation neuroscience with deceptively hippie-ish titles)
Not sure if he does Jhanas. There's a Leigh Brasington thing on them, but that seemed outdated and still crude. I have had my own theories about them, but they're too half-baked to even remember, let alone state.
1. Meditation teacher Rob Burbea was (and I agree with him) somewhat critical of the notion that "all experience is discrete mind moments coming one after the other". Not that this is not how you will start to perceive things if you really tune into the quick changing of experience. But rather that this is an artefact of this specific way of looking, rather than an universal aspect of experience. It's not that we fail to recognise the discrete mind moments in daily life (or when meditating in a different way), its that they are not there.
2. Jhanas are more finicky than that. Many people will not be able to get them at all no matter how much they practice. For many others, it takes months of intensive practice in the right conditions and jhana becomes unavailable once the intensive practice stops. Most theravada teachers I've heard said something to the effect that you should get as much concentration as you can reasonably get (including jhanas) and then just do insight practice.
> Many people will not be able to get them at all no matter how much they practice.
Is this something that your teachers said? I haven't come across this assertion before (and, I admit, I don't really want it to be true...). Is this one of those "hard jhana"/"soft jhana" things?
Basically yes. Both for weeklong jhana retreats as well as 3 months and longer intensive retreats jhana will onl happen for some people. I have also heard numerous teachers say that they themselves were unable to experience jhanas (the question of jhanas gets asked in like every second retreat I've ever been on and I'd say about half the time the teacher responds that they can't really comment on that because they never got there, sometimes not for lack of trying). I believe the harder the jhana the less people are likely to be able to get it, but even "Mr. soft Jhana" Leigh Brasington states in his book that only some people will be able to get his version of Jhana in his workshops (though the caveat may be: maybe if they practiced for a longer period of time...?).
I could imagine that if you increase the practice time enough and go for a very soft jhana most people would be able to get there eventually (if sporadically). But at that point it makes more sense to frame the whole thing as "concentration practice" rather than jhana practice and here the usual advice is to just go for as much concentration (shamatha) as you can reasonably get without significant strain. And it is assumed that the outcome will be very different for different people (as in all things in meditation).
I mean a more general point is that any time you consider a concrete goal in meditation, the timeline to reach it will be orders of magnitude different for different people. From getting it immediately or with minimal practice to "not in this life".
re 1:
Yeah, I think there's loads of confusion with the one-after-another assertion. If I focus on lower-level vision percepts (especially in low-light conditions), I can see individual color sensations vibrating in the form of millions of little dots at once.
https://en.wikipedia.org/wiki/Visual_snow
I am not seeing millions of discrete things one after another in the span of a second. Instead I am seeing them all at once.
Maybe once could claim, that I am just seeing binding moments sequentially and each binding moment gives me a picture of the whole complex visual field (combining millions of lowel-level mind moments? Combining some of them and then automatically using another part of the brain that generates an algorhithmic illusion to seemingly "fill out" the rest of the map?)?
I still think that the mind-moments model is a great starting point for investigation. Just gotta adjust it quite a bit, as one learns more about how things actually are.
Integrating fuzzy logic, parralel computation, fame-in-the-brain, the actual neurological layout etc.
Still working on that.
Yeah. My take on this is that if we don't try to meddle with our experience, that is don't try to fit our experience to some kind of implicit standard, belief, need etc. we already know exactly what experience is like. Any time we try to find a conceptual framework to describe it we need to acknowledge that the framework cannot provide a full description. And that if our model makes very concrete claims (like discrete mind moments) it is probably reductionary in some way.
Sidenote: This meddling can get very very subtle. On the other hand non-meddling is enlightened looking. Its like what the Taoists mean with Non-action.
"all experience is discrete mind moments coming one after the other" - this seems like a statement with essentially no evidence. What little there could be is purely introspection, and seems super unreliable.
Until someone manages to design some experiment that could tell whether experience is made up of discrete moments or is continuous, it strikes me as an empirically empty statement.
Event-related potentials (ERPs) are like single-shot brain waves that occur in response to stimuli; had to take a whole section on them in grad school (with the guy who wrote the ERP textbook), and I still was not sure how they were supposed to be important. Most of it was used in research on attention (e.g., this spike would occur if you noticed X but not Y); I think they are supposed to be informative because you can measure them without having the subject respond at all (no button pressing responses, just pure noticing). At least they don't bring out some of the "woo" or just speculation like brainwaves tend to do. Was it Crick & Koch who thought consciousness was (possibly) a 60 hz oscillation of the frontal cortex or something like that?
In my undergrad biology program we visited a brain research lab near Moscow. The brain scientist gave us a brief intro to Fourier transforms, which made me understand how beautiful they are - something that 2 years of undergrad math classes didn't manage to do.
Then he explained the brain waves to us like this:
"Imagine you are standing outside the football stadium. You don't see what's happening inside, but you hear the chatter of the crowd. All the individual words blend together into indistinct mess and although there's definitely a local information transfer going on, from the outside you can't make out anything specific.
Then imagine one of the teams scored a goal. The crowd behavior is now very different! The fans of the winning team start to cheer and sing. You can easily pick this up from outside and infer what's happening. This is because the individuals behave in a globally coordinated manner, so their signals amplify each other in tune.
From this perspective, brain waves are a byproduct of globally coordinated neuronal activity, and it's the first one we historically learned to pick up. They appear when neurons stop chatting with each other and start chanting in unison."
Then he plopped some probes on my head and announced I have beautiful epileptic spikes (I'm not an epileptic)
That’s a brilliant metaphor!
And the idea is that your consciousness needs a minimum threshold of synchronous activity, creating more powerful EM waves?
I was in a workshop with some people who studied "Grandmother neurons." The story is that a particular neuron only fires when you see your grandmother. This is a false. The true story goes something like this: Imagine encoding memories as combinations of 88 piano keys (neurons). If you used a binary encoding, you could store 2^88 memories, but if you got one note wrong, it would be a totally different memory. If you stored one memory per key, you could only store 88 memories. The brain actually uses sparse codes-- a memory encoded as a chord, only using up to 10 keys played at once (since we only have 10 fingers) This can store "88 choose 10" which is about 4 billion memories. So when the scientists measure a hundred randomly sampled neurons, it just so happens that one particular neuron only participates in the chord for the grandmother in the tests. Does it participate in other chords? Yes, but they are all for images the scientist didn't test. Are there other cells involved in the chord for the grandmother? Yes, but the scientists didn't stick a measuring device next to those neurons (it's hard to do and if you stick in too many probes they damage the tissue).
This also seems much more intuitive and efficient + it falls in line well with what apparently happens in Machine Learning for example in word encodings: Each word/concept will rather be represented by some vector in the space of activations, the "neurons"/dimensions themselves meaning different things.
I may be missing something - don’t all of these encoding schemes suffer from the “one note different means a totally different memory” problem?
The natural solution would be to have similar chords refer to neighboring areas in the cluster of thingspace, like encodings for grandma and grandpa to differ by a single note.
Ah, kind of like word embeddings, then. That makes sense.
That would also explain some real-life failings of memory. Suppose I had two uncles named Alex and Ben. I was talking to/about Alex but mistakenly referred to him as Bob without any particular reason other than both running under the "uncles" category.
You made a strikingly appropriate mistake
I feel like it's probably not going to be totally different, it's going to include all the lower-bit memories at the same time. If one bit is "the sea" and another is "gray" and another is "cross-section windows" and all those are in the pattern of "grandma", you end up with "grandma who lives in the gray house with the cross-section windows by the sea."
So if a bit misfires and you don't think "grandma", you'd still get "like someone who lives in a gray house with cross-section windows by the sea". Kind of like shopping lists, where you can remember three items, and you can remember there was a fourth item, but it's pretty random whether you can actually remember the fourth item without checking the list.
Depends what you mean by "one note different".
In the binary encoding, yes. Every combination of keys identifies a unique memory.
In the unary encoding, yes if and only if "one note different" means replacing one note with another note. Adding or subtracting a note give you an invalid code.
In the chord encoding, the behavior is the same as the unary encoding. Replacing one note with a different note gives you a valid chord representing a different memory. Adding or subtracting a note gives you an invalid chord.
You might be interested in the concept of "error-correcting codes".
Arguably, even AI models have "waves". It is just that with the currently dominant paradigm of transformer models there is only a single wave going from input to output and no oscillations or recurrences. This used to be different, recurrent models like LSTMs, Neural Turing Machines and other older ideas could be described as oscillating. It just turned out that these recurrent models are hard to train with current methods and to scale to large amounts of data. Interestingly, recurrent models are thought to be better at reasoning tasks and algorithmic learning, something that transformer models are bad at ("bad" here meaning in comparison to their other skills like memorization and language understanding).
Huh, I know that waves can cause issues with single-step implicit numerical schemes. The single step will amplify the wave, eventually leading to the numerical solution exploding. In atmospheric models, predictor-corrector time stepping schemes are used that analytically damp the waves. I wonder if there's anything like that for training recurrent models and it it actually helps at all.
This sounds a lot like filter stability in iir filters. It's a stable filter as long as all of the poles of the transfer function are within the unit circle in the z plane - in plain English if you look at the frequency response graph nothing is above 1, so all becomes dampened over time.
So yeah, part of the issue was vanishing or exploding gradients with classic RNNs. You can kind of imagine the "transfer function" of the net being either too dampening or too exciting. The gate in LSTMs was designed to help with this by stopping or opening the flow a bit automatically.
My complete layman's guess is that a lot of this stuff only really applies well to Linear systems, and when it becomes non-linear it gets way harder.
Any recurrent network definitely shows oscillatory behaviour - I used to train GRUs (simpler variant of lstm) a while back for sound generation. Part of the difficulty of using them was how unstable they could be - things would work really well for a few seconds, then just degrade into noisy or weird feedback loops..showing classic chaotic behaviour. I managed to get some small ones running in realtime on fft data, with live input bled in from a microphone and the state feeding back into itself. You could drive it in and out of different stable points with different inputs, much like microphone feedback in a room if you've ever played around with that.
I think this is prevelant in almost any recurrent system, which raises the question of how our brains don't just collapse into noise or get stuck in endless loops...they're kind of finely balanced to be able to skirt around the edges of various attractors, moving from place to place.
>which raises the question of how our brains don't just collapse into noise or get stuck in endless loops...
I would submit that this does happen with some people. Think obsessive/compulsive behavior and other kinds of neurosis.
What’s that tune you can’t get out of your head?
I have no scientific knowledge on this subject but in it’s broad terms it makes intuitive sense to me.
> What’s that tune you can’t get out of your head?
There's an episode of Star Trek: The Next Generation in which a character is psychically afflicted with a melody and goes insane, and then loses consciousness, from the inability to escape it.
But that doesn't happen to non-fictional people. The tune you "can't get out of your head" is in fact easy to get out of your head by doing pretty much anything other than thinking about the tune. Nobody gets stuck on it, in the sense of being unable to do other things. Rather, it's a phenomenon that occurs while the brain is idle, and goes away when the brain stops being idle.
Some deeply dysfunctional people can get stuck in loops, and perhaps obsessive/compulsive behavior is an example of that kind of thing, but the often-complained-about phenomenon of having a tune in your head is not.
Well, think of it as a milder form of the same phenomenon. The review talked about how the various oscillations can be brought into some kind of “harmony “ by altering the frequency of some of the waves, which we can do (meditation, bio-feedback, rhythmic excercise etc.)
Or not…
Yes! So maybe OCD is like being stuck rolling around in a basin of strong attraction. Why isn't it more common given that I think that is the most common failure mode of complex recurrent systems. In fact why aren't other failure modes like seizures much more common?
Because their presence gives a hefty fitness penalty?
That's an evolutionary explanation, a bit like responding to "how does this crazy looking bridge stay up?" with -"because otherwise it wouldn't be able to carry stuff over it".
What I want to know is how is the structure of the brain evolved in an arrangement so as to not cause these kinds of issues with sensitive complex systems.
Like everything else, by trying random things and selecting which happened to work. I suspect that what you really want is a description of a particular feature which (mostly) prevents these issues, but our understanding of brain design is likely far too incomplete for that.
Brain waves are weird to me. I think of neurons as having certain jobs. Let other neurons know when there's a stimulus, or when the sum of stimuli reach a thresh hold, or do whatever computation. How does it help if every now and then (by which I guess I mean ~7 times a second or whatever) they all throw that calculation away and do the wave? Brainwaves seem to me like having the music on real loud in a party and trying to hold a serious conversation at the same time. But... music is pretty much ubiquitous at parties, so maybe it's actually a good analogy?
I guess it makes sense to me as a clock. You have your neuronal system, you do a thought, the thought gets fed to the output of your cluster, and then you you briefly force all neurons to high so they get dampened to zero, refreshing the state for the next thought cycle.
Well, from Scott's description it sounds like they keep doing their job just fine, and being a part of a "wave" simply means that their job is temporarily involved in the workflow of another department.
Maybe that helps. My point is, if a neuron firing means, say, "I see blue" how come during a brain wave we don't see all these crazy flashes for color, ditto for sound, smell, memory, emotion?
I guess the answer is that a neuron firing doesn't mean anything so simplistic, but instead contributes to a model of blue, or sound, or Grandma. And the receptors "know" that out of sync firing indicates those things, but in sync firing doesn't. It's hard to get my brain around itself sometimes.
Suppose, for simplicity, it's a neuron in the visual cortex, its firing only means "I see blue", and whenever the corresponding part of the retina receives blue light, it fires. Most of the time it's in your peripheral vision and you're not paying attention, which corresponds to it being "out of sync", but once in a while you focus your consciousness there, and suddenly it's "a part of a wave", while doing the same job all the time.
Thanks I was considering whether or not to order this just the other night, and this review has tipped me into going for it.
Here's an interview about his latest book - which seems to get at the idea that the brain is a self-coherent system, which is stable in and of itself without needing to be "driven" by sensory events. Anyway the summary can do a better job than me, for the lazy its all summarised from about 1:02
https://brainsciencepodcast.com/bsp/2020/172-buzsaki
I love the examples here. The concept of “a stimulus is only detectable if it hits at the wave Peak” made me think of some mechanism for tuning out annoying sounds.
Minor typo - in the section talking about waves of different frequencies getting into sync, you’ve got the frequencies backward. The 2.1 Hz wave will prompt the 2Hz one to fire early, and the 4.1 Hz wave will prompt the 2Hz one to fire early on its second pass, rather than the other way around. (Everyone always gets wavelength and frequency reversed!)
You've hit on my favorite topic! I too found his stuff on pink noise difficult to parse.
Brain waves become significantly more exciting when you do a deep dive into specific oscillators.
1. Sharp-wave ripples (SWRs, 80-140 Hz) play an essential role in memory recall.
> Girardeau et al (2009) provides loss of function evidence that suppression of sleep SWRs dramatically impairs subsequent memory recall for spatial tasks. This finding is nicely complemented by several gain of function experiments. Fernandez-Ruiz et al (2019) were able to prolong ripple duration, and showed that prolongation improved performance, and conversely that shortening impaired performance. This may explain why novel situations naturally evince longer duration ripples.
https://kevinbinz.com/2021/12/30/sharp-wave-ripples-and-memory-retrieval/
https://www.youtube.com/watch?v=a8rbCJ8JSd0&t=41s There's a longer version of this video in supplementary materials of the paper.
2. The SPEAR model, which associates peak of theta wave with past encoding vs trough with future predictions is one of the most extraordinary facts I know.
https://kevinbinz.com/2021/12/09/the-theta-gamma-neural-code/
3. The link between respiration-entrained rhythms and volition.
https://kevinbinz.com/2022/07/24/the-rhythms-of-anxiety/
4. How gamma-beta implement feedforward and feedback signals (in shallow and deep layers of cortex respectively), fully consistent with active inference models. Very strong recommendation for Earl Miller's laboratory. https://www.youtube.com/watch?v=Kqyhr9fTUjs&t=1841s
When I was an undergrad (1993) I had to lead a seminar on something for a senior CSC class. I chose chaos-and-neural-networks because there had recently been some intriguing brain research that purported to demonstrate that oscillations of neural activity were chaotic and developed attractors that corresponded to what the neural activity was "about" (the specific example had to do with the olfactory bulb of some animal and different scents). I speculated that maybe the course of artificial neural network research might change and become less linear (input neurons ↣ more layers ↣ output neurons) and more feedbacky (input neurons ↣ bunches of neurons feeding back to one another perpetually kind of like in those game-of-life animations).
Anyway: I bring this up because after graduation I completely dropped this thread and haven't kept up on it, and now I'm suddenly curious. Does this cognitions-as-chaotic-attractors thing still get taken seriously? Does anyone do anything with artificial neural networks that maintain a constantly, chaotically oscillating state rather than designating some "output neurons" that terminate the activity?
> And the waves solve wetware-specific problems, like conduction delay (silicon chips operate at the speed of light) and synchronization (computers have internal clocks, or can synchronize with atomic clocks via the Internet).
Silicon chips definitely deal with conduction delay. The speed of light is very fast, but at 4 GHz electric signals (moving at roughly 2/3 the speed of light) can only travel 5 cm in one clock cycle. Traveling 1 cm uses 20% of the cycle, etc. Computers also have to deal with various types of synchronization issues for a variety of reasons.
However, the logic here was flawed from the start, because artificial neural networks are a mathematical model simulated on a computer and the properties of the computational substrate of a simulation don't carry through into the simulation. And so ANNs indeed do not have to deal with conduction delays and synchronization issues the same way that biological neural networks do, but the reason does not have anything to do with the low-level details of silicon chips.
In an ANN, there is simply a list of neurons, a record of which neurons are connected to which others, and an assignment of weights to these connections. There is normally not an assignment of lengths to the connections. Therefore there can be no conduction delay, for example. You could create a nonstandard ANN with lengths and simulated conduction delays if you wanted, as that freedom is the nature of simulation. The conduction delays in the wires within the silicon chip don't inherently create conduction delays in the ANN, because the connections in the ANN are not wires in the ANN, they are numerical data upon which the silicon chip performs calculations. Just like the flatness of the silicon chip doesn't prevent simulated 3D environments, etc.
This may seem piddly, but it's important to keep in mind that the properties of an ANN come from the properties of the mathematical model, not from the properties of wiring in silicon chips.
I wonder if someone has made an ANN with lengths and simulated connection delays like that then?
Thank you! I came here to take issue with the phrase that you did, but you have critiqued it so much better than I would have.
As someone who knows nothing about the subject, my first reaction is: How do we know brain waves aren't just random noise to which we wrongly impute meaningful patterns? The post mentions that "pink noise" exists in stock market behavior. Aren't waves in the stock market the kind of things technicians love to write books about but which ultimately have no value for traders?
Something about this vaguely bothered me, and after a while I came up with the following dumb analogy to attempt to explain my thoughts:
An alien scientist publishes a paper saying "humans appear to have natural patterns in their behavior, tending to become quiescent for around 8 hours then active for 16 in a 'wave' that sweeps across the planet", and so on. But another alien then, reading it, responds: "how do we know this isn't just random noise to which we wrongly impute a meaningful pattern?"
That is, depending on their aims, this may or may not be interesting or useful — is "well, they sleep when it's dark" surprising and cool, or does it destroy a fascinating hypothesis about humans possessing a planetwide vibrational "wave of consciousness" or something? — but it's certainly *a real observation.*
That is (redux): Random, meaningful, and "of no value to traders" are different things. Brain waves aren't *random* noise: the patterns exist. They're probably meaningful, at least in the sense that "we have two eyes" is meaningful (it describes a true fact and some sort of reason for it can probably be derived) and possibly in a deeper sense as well (like the hypotheses, about how the brain might use them, which are outlined in the essay).
Are they of value to traders? ...I'm not sure about that, either; it was one of my first thoughts too. Stock market behavior can be accurately described but remain useless for trading; a trend can be real but ultimately unprofitable, if e.g. it's widely-known and priced-in already, or otherwise not specifically actionable.
Brain waves, similarly, may be a real thing that doesn't tell us much except "this thing happens". Much like, in the end, this entire comment is accurate (I hope), but ultimately says nothing except "yeah you're right I dunno either but I autistically object to some of the word choices in there".
You're right. My comment didn't add anything.
It was a fascinating post yet for some reason I posted a knee-jerk response before even going back and re-reading it closely.
Maybe there's a brainwave-related explanation for why you did that!
AI have no brainwaves?
I beg you to consider the beautiful paper "could a neuroscientist understand a microprocessor" https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1005268
Gems like the donkey-kong transistor vs the space-invaders transistor on the motorola 6502 (lesion the transistor, observe which behavior is impacted). Different switching rates and synchrony between chip regions across different behavior (switching transistor as analogue of firing neuron).
They didnt attach a real oscilloscope, but that would probably have uncovered beautiful chip waves, from the slow background 50 Hz (bleed from the power grid) to chip clock, memory clock, a variety of bus speeds, etc.
edit ps: speed of light is not so fast. 5 ghz means 6 cm, or about 3 x diameter of a chip. Brain analogue 50 m/s across 20 cm would be 80 Hz, so signal conduction in chips is quite slow compared brains, in relation to clock and distance.
pps. This always gets me when neuro people are like "the brain is so weird, it rolls back time and retcons stuff during saccades. an ai would never do this, bio sucks". Have you lived under a rock during spectre? Your phones cpu does this constantly. When you ask the cpu for its state it literally rolls back time and retcons a state it could have been in that is compatible with assembly level understanding. As the quip goes, the architectural state is architected.
TBF those waves in computer chips point to some pretty important aspects of how the chips work. I mean the clock signal is doing basically the time slicing that was one of the proposed uses for these waves and the waves in the hippocampus do sound alot like the patterns that an oscilloscope would read thanks to the rythym of memory update and fetch.
I think you mix up some concepts here. AIs run on computers, which use oscillations as part of their functioning. That does not mean that the oscillations are part of the AI algorithm or design, or even strictly necessary. "AIs" are just implemented on computers, but are relatively independent of the actual hardware implementation, and therefore should be regarded separatedly from any oscillation going on inside e.g. the CPU. The oscillations of the CPU have no effect on the computation inside a neural network. You could arbitrarily slow down the computations, oscillations and factors, and probably think of a CPU implementation that does not rely on oscillations and would still be able to perform the same computations than a neural network.
I agree that it's not a direct implication but TBF the clock synchronization is pretty essential to the programming of an artificial nueral net. You need to sum the inputs to each nueron at each layer so how do you know that you've accumulated all the inputs and are ready to sum them and go on?
Well it's a bit opaque but if we really get down into the weeds it depends on the apparently sequential nature of instructions being processed in a core and, in a parralel application, the ability to rely on certain atomic updates and reads. At a really low level those are possible because of the clock signal coordinating the movement of data around the chip.
Yes, there has been some research into clockless chips in certain applications but for the general case clocks offer some pretty clear benefits and may even be necessary.
Still I think the comparison is not a good fit. For the brain, the oscillations are engrained in the workings of the "algorithm" that is computed. The oscillations happen on a pretty high level, at a mostly global or regional scale. The oscillations adapt to each other, and this adaptation is part of the algorithm. For the computer, the oscillations are merely just supporting the algorithm. They happen at the most basic, low-level. There is no strictly necessary feedback loop between the highlevel functioning and the low level oscillations (disregarding downclocking to save energy, something entirely different). You could arbitrarily halt and continue computations on the CPU and would still obtain the same output results. This would not hold true for oscillations in the brain. If that is enough to claim that "there are oscilliations in current AIs", then you could probably also claim that "there are oscillations in a traffic light" and it would be similarly meaningful.
Before I respond I agree with you that there are some differences but I don't think they are as substantial as you suggest. And we don't really have any evidence of feedback between the two in the brain (there is no experiment where you change the rate of brain waves and produce different functional effects in the brain). I break this up into a few points below all of which are meant to suggest that we don't really have much evidence that brain waves are really all that different than what we see in a computer running a complex piece of code in response to external inputs.
1) I think the OP here is (or if they aren't I will) making the argument that we don't actually have that good a reason to think that these brain oscillations are happening at a "pretty high level". I think one point of the observation is exactly that if you looked at a microprocessor implementing any kind of complex program without knowledge of how it worked the most salient signals would be a number of wave like patterns from the clock signal to the patterns of memory access to display updates. In a certain sense the clock signal is low level but you would see a lot more such oscillations reflecting memory and data access patterns and I'm not seeing how that really differs so much from the brain case.
For instance, if it's an AI taking input from a camera running it through some kind of neural net, producing output and then updating the weights on that neural net you'd expect to see a number of different patterns of oscillation reflecting the different memory/transfer patterns for copying the data from the input, preprocessing it for input into the neural net, propogating the data through the neural net and then backpropogating to update weights. Indeed, I wouldn't be that surprised if you wouldn't see different patterns of oscillation depending on the fan out of the level of the net being processed at a given time. Is that the more high level kind of thing you were thinking about?
2) The reason I focused on clock signals is that I suspect that synchronization is a pretty universal feature any system needs to achieve and that this almost inevitably requires some kind of system wide signal to coordinate that and that's one potential explanation for some of the brain waves. In a processor you tend to get one global clock signal (though we are moving past that and allowing different parts of the system to be clocked at different rates) but it's not clear to me that's how the brain would do it so maybe that's all we are seeing in the brain too (just different low level coordination signals).
Even if you suppose the only point of brain waves is to allow for basic synchronization like clocks in a computer wouldn't you expect different processes in the brain to potentially use different synchronization rates just as you have a different clock signal for your memory bus, your CPU, your PCIE etc..
I admit this point is pretty speculative and I'm not arguing that it's the most likely explanation for what's happening only that I don't think we can exclude it so it's hard to say the two cases are so different.
The paper i linked applies standard neuroscience techiques, the ones that discover brainwaves and grandmother neurons and connectomes, to a pretty simple cpu running donkey kong and space invaders. It then describes the results and compares them to an actual understanding of the system.
As such, it is intended as a scathing methodolical critique of the field of neuroscience.
My narrower point in linking this paper was: Scott has no leg to stand on, when comparing presence vs absence of certain phenomena (wave/oscillation, grandmother neurons) in brains vs ai / motorola 6502 running donkeykong. This is because the same analytical toolkit, when applied to the system we dont understand (brain) and the system we do understand (motorola 6502) yield similar phenomena (wave like patterns, task specific transistors/ neurons, etc).
My wider point is that our understanding of the brain, and the "neuroscience toolkit", is inadequate to support the kind of speculation here, especially given that this toolkit fails so hard when validated on systems that we actually do understand.
Super interesting post but raises so many questions!
> there are other even longer-period brain oscillations of unclear purpose.
What are these??
I understand that when neurons fire, by default they're firing at like 100Hz and 'increases in activity' are upticks in the rate. What's the relationship between these frequencies and brain-wave ones? Are some waves at this high frequency, and some of them talking about pulses of raised frequency?
I wonder if creatures with very small/simple brains like jellyfish have brain waves?
Fun chart here but I guess it’s total neurons in body, not brain: https://en.wikipedia.org/wiki/List_of_animals_by_number_of_neurons
Action potentials, the mechanism by which neurons activate, are not unique to neurons, so wave-like synchronous activity between cells could very well exist in animals without brains (and plants).
Interesting, thanks
Hey all. - I'm a working neuroscientist who's spent some time thinking through some of these issues.
I would say that there's a significant proportion of the field that is, shall we say, "skeptical" of the Buzsaki-style view on oscillations. Everyone agrees that the oscillations exist, and that changes in the oscillations are often correlated with changes in brain state, but making *causal* statements about their impact has proven much more difficult. There are certainly some suggestive data in the hippocampus pointing towards the importance of things like theta-band power and sharp wave ripples, but even there you get inconvenient data points like the fact that bats have a completely normal, functioning hippocampus with no apparent theta band oscillations: https://www.sciencedirect.com/science/article/pii/S0092867418312297
More broadly, it's hard to shake the feeling that the brain has some natural resonances for various states - sleep/unconsciousness is low frequency, low activity is ~alpha/beta, high activity is ~gamma. A statement like:
"Ask someone to think about a certain topic, and cells representing that topic will form a neuronal assembly (ie they will start oscillating together at the same frequency) in gamma rhythm."
is really just saying that those cells are active, and neurons in an active brain area tends to follow a gamma rhythm. In fact, I'm reasonably confident that if you were to look at the internal activity of cells in the same area that do NOT represent that topic, they would also be following the same gamma rhythm, just with less tonic input and thus less spiking activity.
People like to study oscillations because they're brain-wide phenomena that are (relatively) easy to detect in humans with non-invasive methods like EEG. However, anyone who's looked at neural activity using more finely-tuned, invasive methods (e.g. single neuron electrophysiology) knows that individual neurons close by each other can be tuned to very different sets of variables, and that when you average signals from large populations of neurons you lose a lot of useful information.
To be clear, I think everyone in the field agrees that neurons act together in assemblies, that patterns of activity that are shared across neural populations are important, and that sometimes these population patterns of activity can oscillate. The problem is that there are many, many, many of these subnetworks active at any given time, and that when you average them all together you get a signal that's easy to measure, but is limited in its ability to explain anything beyond basic brain function. You have 86 billion neurons in your brain. You should be skeptical of theories that tell you just 4-5 frequency bands are the secret to understanding it.
> Everyone agrees that the oscillations exist, and that changes in the oscillations are often correlated with changes in brain state, but making *causal* statements about their impact has proven much more difficult.
One argument I heard against this that struck me went along the lines of: "People have this gripe about oscillations being causal and many intuitively see oscillations as epiphenomenal. However, the same people take, at face value, that increases in brain activity has causal effects. Why is this? There is no stronger philosophic argument for activations rather rather than oscillations."
I think it's just application of a reductionist principle in this case. Do oscillations tell you anything you wouldn't know from simply averaging spike trains? If not, then I think we should prefer the spikes.
I think the pro-oscillation account is that yes they do. It would argue that much information can be gleaned by considering how those spike trains give rise to oscillations. They would argue that there are several meaningful levels of analysis (e.g., neurotransmitters within a synapse, individual neuron firing, etc.), and oscillations are one of those levels.
Can you provide an example of information that can be gleaned by considering how spike trains give rise to oscillations?
Here's my intuition for why spikes are more meaningful than e.g. neurotransmitters for cognition. I can stimulate a brain region electrically or optogenetically, thereby by-passing incident neurotransmitters, and drive a behavior. I don't think it's even possible in principle to do that with oscillations (i.e. by-pass the firing of single neurons).
That's a good point. I think in principle this is somewhat testable, by seeing whether the effect of optogenetic stimulation depends on whether the stimulation gives rise to an oscillation. I don't know enough about oscillations to cite anything though.
I have no clue what it would even mean to cause an oscillation without bypassing the firing of individual neurons, unless you just rewire them somehow (I'm also not a single-neuron researcher).
So, it's relatively easy to show how activity in motor areas causally leads to changes in muscle activity / movement. And, more generally, if Neuron A is synaptically connected to Neuron B, we have a very rich understanding of the electrical and molecular mechanisms that explain why an increase in activity of Neuron A causally increases the probability of activity in Neuron B. In so far as a mind is the product of 100 trillion neural connections, we know that each of those individual connections represents a causal chain that propagates increases in brain activity. There's lots of other very suggestive evidence, but these basic facts are, I think, why most neuroscientists feel comfortable hanging their hats on the idea that changes in behavior are tied to changes in neural activity somewhere.
In contrast, the structures of neural oscillations are (in most cases) set by emergent properties of densely connected neural networks. There's been amazing progress in trying to model these kinds of networks, but in truth we're still in the pre-statistical-mechanics era of neuroscience. We know a lot about individual neurons and neural connections, but we still don't have a well-accepted theoretical framework that allows us to understand how large populations of neurons interact and compute. This ignorance makes it hard to make confident statements about the causal impact of oscillations, or really of any other network-level phenomenon.
The philosophic argument for neural activation is that we understand its causality in very simple networks. The argument against oscillations is simply a lack of evidence, pending better models of network activity.
Thank you for writing this so that I didn't have to.
One thing I might add: it's very difficult for me even to imagine how a network of interconnected neurons, some of which are spontaneously active, could fail to give rise to oscillations. This to me is the best reason to think that EEG signals don't have much explanatory power over and above the neuronal spikes upon which they supervene.
One last point of clarification that may be helpful to non-neuroscientists: it may be useful to distinguish "brain rhythms" at different levels of analysis. I can imagine someone reading this and thinking that hippocampal theta is pretty much like visual cortex event-related potentials (ERPs). These are completely different phenomena measured using completely different methods! Hippocampal theta is measured as a local field potential (LFP), which usually means actually sticking electrodes in the brain and then low-pass filtering the signal. One step up from that is intra-cranial EEG, in which electrodes are placed on the surface of the cortex (in neurosurgical contexts). And finally you have good old-fashioned EEG, where you wear a cap over the head, like in a psychology lab, and measure electrical signals that originate over many centimeters and are difficult/impossible to conclusively localize. This is where people see ERPs.
I'm not a reductionist at all costs by any means, but generally I think it's safe to say that the actual causal relevance of each of these "rhythms" decreases as you ascend the hierarchy. Hippocampal LFP stands out as potentially interesting because of phenomena like phase precession of place cells, but Justin L aptly points out that this is by no means a necessary feature of place coding in general.
Yeah this seems like taking a 50 megapixel image and collapsing it down to a color frequency histogram. You’d get very interesting correlations! You might get very good at picking out trees from fire trucks and find all kinds of insights, but it doesn’t change the fact that you have thrown out 99.99% of the useful information and are looking at an Nth order artefact and trying to reason about it instead of the real thing.
Brain waves are what we can map, so spend enough time looking at them and you can talk yourself into thinking they are the territory.
Instead of 2.17, the golden ratio φ=1.618... is best for anti-resonance. This is because φ is maximally difficult to rationally approximate.
Can you explain this in more detail? or provide a cite? I would guess the opposite because the ratio of consecutive terms of the Fibonacci sequence converges to φ.
Every real number can be approximated by an ever-closer series of rationals, since the rationals are dense in the real numbers (e.g. the square root of two is approximated by the fractions 1/1, 14/10, 141/100, 1414/1000, 14142/10000, ...). The question is how good of an approximation you can get for rational numbers with denominators up to some bound.
This turns out to be related to the size of the terms in the continued fraction expansion of a number - larger terms in the continued fraction expansion yield better approximations of your chosen value. (The unusually-good-for-its-complexity rational approximation of pi as 355/113 exists because of an unusually large term early in the continued fraction expansion of pi as [3; 7, 15, 1, 292, 1, 1, 1, 2, 1, 3, ...]. If we round the 1/292 term portion of the continued fraction to zero, we get 3+1/(7+1/(15+1/1)) = 355/113.)
The golden ratio's continued fraction expansion is all 1's, so in this sense it has the worst rational approximations of any real number. In practice, people tend to care about the asymptotics of this approximation quality, which leads to the notion of https://mathworld.wolfram.com/IrrationalityMeasure.html. More concrete implications between the terms of the continued fraction and the badness of rational approximations in this sense are mentioned at this MathOverflow thread: https://mathoverflow.net/questions/89600/numbers-with-known-irrationality-measures
But that's actually saying phi is as easy to approximate as an irrational number can be. As an irrational algebraic number it has mu = 2 while transcendental numbers have all sorts of higher values.
Higher irrationality measure implies *better* rational approximations. The reason it's called an irrationality measure is that rational numbers are especially badly approximated by any rational numbers other than themselves, so having good approximations makes you less like a rational number in that regard.
My bad...I just assumed that it couldn't be that algebraic numbers were harder to approximate than transcendental ones. Still find that hard to believe.
Still, nothing special about psi...no easier than any other irrational algebraic.
Thanks for all the responses. I was a former math major but somehow missed out on continued fractions and irrationality measures.
Another possible analogy - the musical interval between 2 notes has its consonance/dissonance related to having small integer ratios, or degree of resonance (attributed to Pythagoras). So the "most dissonant interval" would have the most anti-resonant ratio. I will have to try a bit of research comparing what the ratio φ sounds like.
A book on continued fractions is a good place to learn more. Continued fractions give the sequence of best rational approximations to any given real number and it turns out that the continued fraction expansion of the golden ratio is [1, 1, 1, 1, 1, 1...]. This is what drives the result, which I believe is called Hurwitz’s Theorem.
That doesn't seem to be true. See discussion below but it has irrationality measure of 2 like all the other irrational algebraic numbers while lots of transcendental numbers like pi or ln(2) are way worse.
And why should that make it best for anti-resonance anyway? I can vaguely see why it might matter but don't really understand what you mean well enough to say.
As mentioned above though, pi is easily approximated as 355/113 even though it's transcendental. Here, "easily" means the error is small relative to the complexity of the denominator. For comparison, a similar-sized approximation of phi, 233/144, has an error 40 times greater.
Resonance is greatest when the ratio is close to one with a small denominator, so a ratio that's far from any small denominator will avoid resonance.
If anyone is interested in learning more about the Fourier Transform (IMO one of the most beautiful concepts in mathematics), there's an incredible Stanford course on it: https://see.stanford.edu/course/ee261
I actually had a big "aha!" moment while meditating during the months I was taking the course. It's hard to describe, but it was like suddenly seeing my internal rhythms in the frequency domain instead of the time domain.
> The hippocampus is traditionally associated with memory, but Buzsaki cites recent research showing that it also handles navigation (eg a rat navigating a maze, or a person walking through a familiar city). Why this combination of tasks? Rhythms sort of suggests that brain areas are less about specific tasks than about specific graph-theoretic arrangements, which are convenient for specific algorithms, which are convenient for specific tasks. The neural connections in the hippocampus are mostly local in a way that makes it a good “auto-associater”, ie you give it part of a pattern and it can recognize which pattern is being suggested. For complicated reasons this works well for both spatial tasks and memory. In fact, Buzsaki sketches out a way that episodic memory might have evolved from our navigation system, where eg remembering a list is equivalent to remembering a path through space, and remembering a specific episode is equivalent to remembering a landmark (he doesn’t specifically say so, but this is a fascinating match for the method of loci mnemonic device).
Also, changing location can make you forget things, the "doorway effect" .
https://en.m.wikipedia.org/wiki/Doorway_effect
According to Wikipedia, pink noise is defined by the power *per frequency interval* being inversely proportional to the frequency - in other words, power *over time* is constant, independent of frequency. I don't see what's supposed to be mysterious about that, though. It makes perfect sense to me that the brain would want to maintain a consistent power draw, and would attenuate the amplitude of high-frequency signals to make that happen.
Right...and if you consider the fact that there is probably something like a per neuron limit on power output per unit time (can fire once every blah Ms) this almost seems inevitable.
> In fact, Buzsaki sketches out a way that episodic memory might have evolved from our navigation system
My episodic memory is terrible, and so is my navigation ability. In other words: put that on the survey!
I think the only examples I will remember are the fake toy ones, because they are simple and memorable. And soon I will forget that they are fake and remember them as real. Well done!
> Brain waves provide “synchrony”, allowing a smallest granular unit of time and essentially converting life into a turn-based game. Suppose that a snake bites your foot. You see the snake with your eyes, and also get a pain signal from your foot. The pain signal has to travel a long way, nerves have conduction delays, and so it reaches your brain well after the visual signal. But your brain needs to be able to combine the visual and pain signals into a single story (snake bit my foot). Brain waves separate experience into short granular “turns” so that the brain can attribute both stimuli to the same “turn” and connect them.
I'm having a hard time getting past the example. It's true that the brain needs a way of combining the visual signal and the pain signal into a single story. But there is quite obviously no need for synchrony or a minimum quantum of time at any point in this process; if I feel pain and then, five full minutes later, see a snake slither away from where I was standing when I got hurt, I'll still combine those two signals into "a snake bit my foot" (where that combination is plausible).
It may be necessary for the sense of the whole self for the brain to integrate different stimuli in that way.
I seem to remember a syndrome where a person doesn’t feel as if they are really in their body, but rather that the body is a thing they are operating sort of by remote control, like a marionette. Probably unrelated like all just-so stories, but I can easily imagine that a delay between seeing your foot get hurt and actually feeling the pain could be part of that.
That said, we’ve probably all had the experience of doing something and realizing, “Oh, that’s going to hurt tomorrow.”
Delays that need to be combined into a coherent whole is the entirety of multiplayer game network coding. People clicking buttons, sending them with variable delays along with a "happened at time t" attached to a server which then has to retroactively figure out who hit and who missed and then send that back in time. Ask game network engineers for a books worth of problems and solutions over the last 40 years :)
Games also use frames, discreet moments in time, to make physics and simulations doable at all. Too low a framerate and you notice the choppiness. Also if your monitor frequency and game frame frequency are nice multiples, it looks good, if not, it's choppy as well. Many analogues!
Why not just say that brain waves are emergent? They don’t do anything in particular. They are just the result of something else that does something. So everything here will only be based on correlation, not causation.
One thing that helped me understand EEG better, and I have seen other commenters mention is that there is a big difference between what the neurons are actually doing and why and what we pick up from the EEG with electrodes on your scalp. Each individual neuron fires and makes a tiny electromagnetic pulse sure, but if all the neurons were aligned randomly all these pulses would sum to zero - no signal. If the neurons all fired at random times they would sum to zero - no signal. If it's a small group of neurons it's too weak to make it to the scalp and overcome noise - no signal. So we only 'hear' large groups of aligned, synchronised neurons. So any sort of brain activity that doesn't meet these requirements is invisible to us unless we use electrodes to record the activity of single neurons. We also can't distinguish between synchronous activity being 'used' by the brain for a purpose (like synchronising areas of the brain), with byproducts of activity (like to actually move a muscle some amplification of a small signal has to take place to have a bunch of neurons fire at the same time).
This helps to explain the pink noise thing - larger groups of aligned neurons make a stronger signal, but neurons are slow so unless these areas are oscillating slowly its hard to get large areas to line up. So stronger signals tend to be lower frequency, average that out across millions of areas of different sizes and you get something like the pink noise curve.
I also can't stress enough how terrible a signal EEG is - it's the worst biosignal by far. Brain activity is already basically random noise with slightly different non-random noise overlaid in it, then its smeared out by volume conduction as it moves through the head, then it picks up every bit of environmental noise and muscular activity, then the electrodes and cables move and skin conductance changes, then its all screwed up by referencing and grounding effects. My research was in brain computer interfaces and my advice is stay well away until either portable MEG systems or intra-cortical electrodes go mainstream.
Arne you familiar with the work of Thomas Collura and neurofeedback ?
Yours is the first post that discusses this group firing that is the basis of neurofeedback work.
It'a even more complicated than this since depending on the frequency the source of extracellular potentials is probably not mostly from spiking activity, but from membrane voltages associated with synaptic inputs and intrinsic electrical signaling. In other words it might not be the neural _output_ that is aligned or not aligned, but the inputs and single neuron transformation of the inputs towards spiking outputs that one mostly picks up extracellularly. This is (as a general rule) mostly going to be true for anything in lower frequencies, certainly theta. For scalp recordings in particular (ie. not for electrodes that actually go into the brain) I'd argue that the signal is probably mostly made up of non-spiking sources, for any frequency measured.
Not too surprisingly, awareness appears to correlate more with overall signal diversity than specific wave frequencies:
"The foregoing studies complement an older, yet largely unacknowledged, body of literature that has documented awake, conscious patients with high amplitude delta oscillations in clinical reports. . . a largely parallel body of recent work has reported convincing evidence that the complexity or entropy of EEG and magnetoencephalogram or MEG signals strongly relates to an individual’s level of consciousness"
Hyper-synchronous activity correlates with unconscious states like deep sleep and generalized seizures.
https://montilab.psych.ucla.edu/wp-content/uploads/sites/49/2021/03/Frohlich_Consciousness_Among_Delta_Waves_Final_Accepted_Version-1.pdf
In Conway's Game of Life, people have made some patterns that act as computers. For example there's this pattern that calculates and prints the digits of π: https://conwaylife.com/wiki/Pi_calculator. Here's a plot of how many live cells it has over the first 8192000 generations: https://i.imgur.com/KyICAc0.png.
In an analogy to the origin of the names of white, pink, and brown noise, the laboratory technique to separate and study DNA of different lengths is named southern blotting for its developer, Ed Southern. The similar method of separating RNA is called northern blotting, and for protein it's called western blotting. Once, I accidentally reversed the electrodes when doing a western blot, so that the protein ended up lost in the buffer instead of on the blot paper, which I guess could be an eastern blot. But apparently there are many things called eastern blots now: https://en.wikipedia.org/wiki/Eastern_blot
I feel the speculation at the end doesn't take sufficient notice of the fact that we are observing the system from the inside. For instance, whether or not time *feels* discrete in meditation has no real necessary relation to how the brain actually processes events. You could easily imagine a brain which processes events in discrete blocks but where this blockiness was imperceptible to self-reflection. Alternatively, the brain could just process experiences as they came in so there was no discreteness but then, when you try to reflect on the experiences, batch them together in a way that feels discrete.
More generally, I don't like these grand attempts to explain meditation/enlightenment as they all seem to suffer from this kind of importance bias: since these states feel so important they must reflect important brain processes. But isn't the most likely explanation just that the brain is quite plastic and it shouldn't be that surprising that you can train/trick it into producing pleasureabld experiences?
But those experiences need not be caused by grand things like linking usually unconnected regions...but most likely by the same kind of fuckups and errors that drugs often work by just without the direct chemical intervention (so less vulnerable to tolerance).
> But isn't the most likely explanation just that the brain is quite plastic and it shouldn't be that surprising that you can train/trick it into producing pleasureabld experiences?
I suppose the counter argument to that would be that if many masters of meditation describe their most profound sensation ("enlightenment") as essentially the same (insofar as an inherently individual experience can be sufficiently communicated and compared), maybe there really is some fundamental, neurological basis for that? It sure does sound interesting enough to investigate further. So many people across so many centuries deluding themselves in basically the same way through meditation would be quite the coincidence, the "no true scotsman" fallacy notwithstanding.
Thanks for finding the time to do this!
> Pink noise is apparently omnipresent in natural systems for kind of mysterious reasons - see eg this Quanta article, which says pink noise “is found in all kinds of electrical noise, stock market activity, biological rhythms, and even pieces of music — and no one [knows] why.”
Since pink noise = power law, the caveats expressed e.g. in https://scottbot.net/networks-demystified-3-the-power-law-rant/ apply:
> The take-home message of this rant will be that the universe counts in powers rather than linear progressions, and thus in most cases a power law is not so much surprising as it is overwhelmingly expected
Also, a fun book about the key role of synchronization of oscillations throughout nature is Sync, by Steven Strogatz. I really enjoyed it a few years back, but it's not really my area, so if anyone who's more knowledgeable about this has some criticism, I'd love to hear it!
Tbf the message i got from parsing the rant is that the power law in itself is not surprising, it is the power law exponent that is interesting (as usual in critical system)
I still have to read it carefully
Two weeks back I had lucid dreaming experience. I knew that I was dreaming while in a dream and I felt my sense of self. The moment I woke up after that dream I was left with lots of energy to carry out the rest of the day. Now what state of consciousness was I at? In dream state. Did I lose my sense of self? No, because I was conscious of the fact that I was dreaming. Probably a dopamine surge followed up the dream. The feeling in the dream was more like decoupling from time or freeing from time such that I was able to easily float around space. What existed was the sense of self observing the dream environment. It was around 5-minute trip in wakeful state whereas what I experienced in the dream was actually lesser than that (This I am making up because I do not have a sense of time in the dream, yet it was till the end, and I was continuously processing the information. But when I compare with the events in the wakeful state, I end up processing very few information in that dream state which is opposed to what I felt because it felt as if I have been for more than an hour in the dream). That is, I was in the midst of listening to some 30-minutes Yoga Nidra script. How is this synchronization of sense of self or time related with states of consciousness? I am conscious of these events in wakeful state. Is it just an approximation I am coming up with!
So I've seen you've taken to using AI Image generation for your cover photos
I'd love to see what an AI set to analysing EEG recordings using Fourier analysis or whatever algorithm it decides to make up would say about them. Just as AIs embarrassingly deduce the race of CXR subjects, they might surprise us with foreseeable things (which aren't always so clear to human eyes) becoming vastly more reliable, like a diagnosis of epilepsy, but it's not beyond the bounds of possibility they could be quite good at some surprising things: ADHD, autism, Alzheimer's. We can't give them training data for that as yet, as we are not aware of EEG findings in those three examples. But if we quietly set an AI to do its thing and later supplied diagnoses as they were made, the patterns, if any, could be found and quickly refined to make a predictive diagnosis for each.
The quote "the natural logarithm 2.17" must be a typo, because he published a paper just a few years before the book presenting the same argument but with the correct value of e https://doi.org/10.1016/S1472-9288(03)00007-4
One of my big insights studying creative writing (and art in general) is that practically everything is vibratory / enacted by oscillation. Stories oscillate along a bunch of dimensions, and will feel "flat" if constrained. Imagine explaining the shape of a working metronome without using time: the hand is to the left and the hand is to the right. It's a paradox, until you introduce the time dimension for the two contradictory states to oscillate along. So whenever you find a paradox, it probably represents an oscillation along some dimension you haven't discovered yet. The activity of the first act of a story is to establish these paradoxes (or "tensions") which themselves establish the dimensionality of the story. They make it feel "round" instead of "flat".
A good example is Pride and Prejudice, which introduces three paradoxes (at least) in the first sentence.
It's fascinating how much this description of the brain sounds so much like one of the best ways I've found to describe an excellent novel. Like, writers talk about "resonance" between separate parts of the story, and it's considered kind of a mysterious, undefinable thing. When it happens, it feels much like how this post describes different parts of the brain tuning to each other's frequency.
Thanks to the title of this post and book, now I have the earworm of Gloria Estefan and Miami Sound Machine:
https://www.youtube.com/watch?v=CZkjeJKBI0M
Oh well, makes a change from the previous earworm of "Hail, Bright Cecilia"!
I had their "Conga" stuck in my head, on and off, for literally *decades*, without knowing what it was. Finally last year I stumbled across it on YouTube, and now I am fully recovered.
I just want to put here this Asimov quote from 1950 'cos I think it's delicious (no spoilers, I think; it's from the Foundation "trilogy").
“The science of electroencephalography was at once new and old. It was old in the sense that the knowledge of the microcurrents generated by nerve cells of living beings belonged to that immense category of human knowledge whose origin was completely lost. It was knowledge that stretched back as far as the earliest remnants of human history—
And yet it was new, too. The fact of the existence of microcurrents slumbered through the tens of thousands of years of Galactic Empire as one of those vivid and whimsical, but quite useless, items of human knowledge. Some had attempted to form classifications of waves into waking and sleeping, calm and excited, well and ill—but even the broadest conceptions had had their hordes of vitiating exceptions.
Others had tried to show the existence of brainwave groups, analogous to the well-known blood groups, and to show that external environment was not the defining factor. These were the race-minded people who claimed that humanity could be divided into subspecies. But such a philosophy could make no headway against the overwhelming ecumenical drive involved in the fact of Galactic Empire—one political unit covering twenty million stellar systems, involving all of Man from the central world of Trantor—now a gorgeous and impossible memory of the great past—to the loneliest asteroid on the periphery. And then again, in a society given over, as that of the First Empire was, to the physical sciences and inanimate technology, there was a vague but mighty sociological push away from the study of the mind. It was less respectable because less immediately useful; and it was poorly financed since it was less profitable.
After the disintegration of the First Empire, there came the fragmentation of organized science, back, back—past even the fundamentals of atomic power into the chemical power of coal and oil. The one exception to this, of course, was the First Foundation where the spark of science, revitalized and grown more intense, was maintained and fed to flame. Yet there, too, it was the physical that ruled, and the brain, except for surgery, was neglected ground.
Hari Seldon was the first to express what afterwards came to be accepted as truth.
“Neural microcurrents,” he once said, “carry within them the spark of every varying impulse and response, conscious and unconscious. The brainwaves recorded on neatly squared paper in trembling peaks and troughs are the mirrors of the combined thought-pulses of billions of cells. Theoretically, analysis should reveal the thoughts and emotions of the subject, to the last and least. Differences should be detected that are due not only to gross physical defects, inherited or acquired, but also to shifting states of emotion, to advancing education and experience, even to something as subtle as a change in the subject’s philosophy of life.”
But even Seldon could approach no further than speculation.
And now for fifty years, the men of the First Foundation had been tearing at that incredibly vast and complicated storehouse of new knowledge. The approach, naturally, was made through new techniques—as, for example, the use of electrodes at skull sutures by a newly developed means which enabled contact to be made directly with the gray cells, without even the necessity of shaving a patch of skull. And then there was a recording device which automatically recorded the brainwave data as an overall total, and as separate functions of six independent variables.”
So in the moments where it feels like time slows down would that mean your conscious brain rhythm is beating faster?
This is blowing my mind. Also do we all have different clock speeds? I’m not sure I’m thinking about this right
Part of the difficulty with doing any kind of direct investigation of EEG tracings is that the EEG is a tremendously distorted view of what the EM fields inside the brain are actually doing. Necessarily: you're recording an electric field at the surface of a pretty conductive material. You'd be much better off using SQUIDs to detect the magnetic fields, because that would give you a relatively undistorted picture of the 3D pattern of EM fields, and I would guess the wavevector structure (i.e. the result of a Fourier analysis in space) would be at least as important as the frequency structure (the result of the Fourier analysis in time). That's expensive, though.
> Gamma was the rhythm I found most interesting - it seems to map very well to “items in conscious awareness”. Ask someone to think about a certain topic, and cells representing that topic will form a neuronal assembly (ie they will start oscillating together at the same frequency) in gamma rhythm. I was pretty surprised that there was this clear of a neural correlate of consciousness.
I'm interested to know if anyone has taken e.g. Daniel Ingram or other expert meditatiors and tried to measure correlates to Jhana and other rarified meditative experiences. As the progression of experiences in Jhana are very clearly defined (at least in some aspects), measuring the difference in activity at each Jhana should be possible at some resolution of instrument.
Coming from another angle, when achiving single-pointedness in meditation, conscious awareness rests entirely on one object of attention, without any scattering or wavering. Presumably there's some strong Gamma there? But with extremely focused attention, are the Gamma oscillations localized in some way, relative to the object of attention that is being focused on? Would we see different activation patterns when focusing single-pointedly on different mental objects?
(The key thing here being, in most minds, attention scatters and wanders from second to second, so these sorts of localization would be impossible to do if the oscillations are on the order of Hz. But single-pointedness seems a quite unusually-stable state, which might make it amenable to objectively studying qualitative/subjective experience.)
I imagine this is something https://theeprc.org/ is into, but I don't know the meditation-neuroscience literature well.
A final take on this; in the formulation of The Mind Illuminated, we have various hinderances such as Dullness and Distraction. These should be measurable in some way; can we detect Dullness from the outside? (For example it might be the percent of Theta waves increases as Dullness is perceived? Can this help a meditator to get feedback on different states? It's said that some get fooled into thinking Dullness is a more advanced state like Jhana; can we objectively detect Dullness to help meditators avoid this confusion? Particularly Subtle Dullness, which can sneak up on you?)
It's an intriguing idea that navigational memory underlays episodic memory. To build on your reference to the method of loci, I guess one explanation for the reason our declarative memory sucks is that we are using a system that was designed for navigation for something that has nothing to do with navigation. It makes intuitive sense: it's hard to remember abstract facts because they do not look like anything, and they are not located anywhere. I had learned that the method of loci works because humans have a *better* navigational memory system, but this is a different explanation: all memory is at root navigational, and it's not optimized for facts unless hacked by picture things in palaces.
For more on this, I recommend The Embodied Mind. It is full of gems like: people can think more clearly when they move their hands, or, people think Chile is south of British Columbia because of something to do with how we assemble objects in our heads. Movement came first, and thought, emotion, and language bare the imprint of this legacy. The book is badly written, but the examples are great.
Apropos, a musician I play with who also happens to be a neuroscientist just published this paper on synchronous states invoked by psychedelics. He'll be presenting at the Society for Neuroscience conference in SD, CA next month. The basic gist is that there seems to be some sort of ~150Hz standing wave of activity that appears widely in overall cognitive processing centers when (a rat) is on LSD/etc.
https://www.biorxiv.org/content/10.1101/2022.09.27.509527v1
Oh boy, I love noise. From a physics, electrical engineering point of view; White noise is also called Johnson noise, it is the thermal noise inherent in all resistors*. And first reported by Johnson at Bell labs. https://en.wikipedia.org/wiki/Johnson%E2%80%93Nyquist_noise#History. (It has a flat frequency spectrum (white) at least for low frequencies... it's a classical effect. It breaks down at higher frequencies where quantum effects have to be accounted for.) Brownian noise is the noise of Brownian motion and is understood to be the integral of white noise. https://en.wikipedia.org/wiki/Brownian_noise. It has a spectrum that goes as 1/f^2 (where f is the frequency.) Both white noise and brown noise are fairly well understood from a theoretical perspective. Pink noise is called 1/f (one over f) noise in the physics community. It has a power spectrum that goes as 1/f (not surprisingly) It is not at all well understood theoretically... which is why we find it interesting.
*other random processes also give you white noise, shot noise is one other example. https://en.wikipedia.org/wiki/Shot_noise
Speaking of machine learning and brain waves, I am surprised not to find any direct mentions of positional encoding; see e.g. «Attention is all you need» arXiv:1706.03762 where they literally say that to make relative word positions easier for the network to use, multiple dimensions of sine waves of different frequencies — independent of the real input — are added as input features.
"If brain waves are really responsible for things like attention and cross-talk between brain regions, then might a lack of brain-wave-equivalents make AIs worse at these things"
Has anyone looked to see if there emergent oscillatory phenomena across the layers of really deep networks? Might me an interesting weekend project.
There’s this book, Psychedelic Information Theory, which is an interesting source to fold in with this book
"Suppose that - and again, this is a fake toy example - your unconscious is oscillating at 1 hertz and your default mode network is oscillating at 2.17 hertz. Then you meditate, you slow your default mode network to 2 hertz, all of a sudden consciousness and unconsciousness are nice multiples of each other, they reach oscillatory synchrony, and they achieve higher levels of communication with each other."
Wondering could external sound waves/rhythms with the right frequency cause something similar? Thinking about shamanic drumming causing trance states. I think I read somewhere that strongly rhythmic music with the correct frequency can cause heart-rate to entrain.
Non-brainwave related side note:
> "[...] if you force it hard enough."
The below is a bit of an axe that I grind, but I think that even just using the word/concept "force," in relation to enlightenment will (surprisingly) cause some non-zero number of meditators to really, really hurt themselves; and I think some larger number of meditators will get sent down a rabbit hole that won't get them to where they want to go:
Scott may not mean "force" in some effortful, phenomenological, qualia-feel sense, but some people will take it that way. Sometimes the experience of "force" or "forcing" does come up in meditation. But, I believe that it's bad to "add force" to one's meditation practice.
Because, say that mind-state-space has a lot of local minima (minima bc lower is better, in this analogy). Some local minima look, at first, kinda like enlightenment (though with initially subtle and then unfolding downsides), and those local minima *can* be reached by forcing.
But, forcing sort of has "inertial momentum" ("karma"!), and many hours of forcing can take a long time to undo (thousands of hours). And, meanwhile this remaining forcing "persists," eventually running amok, causing lots of psychological, phenomenological, behavioral, and conceptual problems.
"Actual enlightenment," or whatever, is, say, at the "true global minimum" and it can only be reached by a preponderance of non-forcing and undoing latent or actualized forcing that's currently in the system.
Enlightenment state transitions can seem abrupt, and can happen spontaneously, but for most people, those "stateless states" are only going to be complete and stable after *thousands* of hours of (paradoxical) "work"--which involves tacking towards effortlessness.
(***Cryptic and prolix, understatement, but more here: https://meditationbook.page/ (Ctrl/command-f on "force" might be interesting.))
Agreed
feels like this has some super interesting implications for dissociative identity disorder that i dont understand this well enough to speculate about
I was surprised by the discussion of AI at the end. You seemed to reach a conclusion that the fact that large regions of the brain having synchronized brain waves was either a necessary precondition of consciousness or a strong correlate of it, then moved on to say that because AI would probably not have brain waves in this sense they would be less likely to have consciousness.
My intuition is that this is backward. It seems like computationally different brain regions operating at the same frequency means either that they can communicate with each other efficiently or that they are updating on the same internal clock. But for most ML models all neurons are updating on the same internal clock and can communicate with one another with nearly perfect efficiency (i.e. little signal is being lost because a neuron is not "close to activation) It seems like AI's are more likely than human brains to have most neurons all simultaneously communicating with one another on the same internal clock.
A potential flaw in my understanding here is that in AI's neurons not firing because they are not "close to activation" sounds like a neuron that uses the activation function max(0, input). If it is currently receiving a very negative input then changing that slightly via some other neuron will not cause it to activate whereas if its input is close to zero or positive changing it slightly will change the output. Is there a proposal here somewhere that having some neurons in an ML system activate at say every 3 activations of the rest of the network and just average the inputs between activations could cause more interesting behavior or something?
"Tune in, turn on, drop out" was a phrase evidently coined by Marshall McLuhan and certainly popularized by Timothy Leary in 1966 to encourage people to use psychedelics. If the frequency synchronization idea mentioned here is correct, the expression is remarkably prescient. Perhaps people on psychedelics literally tune in to normally inaccessible parts of their brain activity by getting them to the same frequency.
Sometimes at 3:45 in the morning, as the party’s closing up and you are well tired, if you’re staring at a white wall you can perceive a flicker. I believe that’s alpha, not waiting for you to close your eyes for once.
If you’re not going to that kind of party much anymore, there’s also a few optical illusions that flicker around the 10hz range. The most familiar of these is the wagon-wheel effect. So just look at a car.
That’s probably as close to watching your own brain at work as it gets in everyday situations.
An idea I've toyed around with is the idea of "blockchain protocols as the clock rhythm for distributed consciousness"
Like, you've got a bunch of computers running the same algorithm. Every so often, a new block gets pushed to the chain, and becomes common knowledge. This constitutes one "frame" of distributed consciousness, and contains some kind of consolidation or average of the input of all the constituent nodes.
Inside those nodes, you can have arbitrary subroutines operating in parallel wavelengths, but the outermost blockchain is like a [conductor | yearly census | drummer boy's marching beat | pulse | bible]
Same thing holds for New York Times headlines, Nobel Prize winners, really any public list with distributed inputs.
I have not read the whole thing yet but I am not too surprised by the pink noise distribution? White noise is constant energy per ± X Hz, pink noise is constant energy per ± X %. It doesn't feel any more surprising to me that the noise distribution should be such that the energy between 1 and 2 Hz is the same as the energy between 100 and 200 Hz (pink) rather than 100 and 101 Hz (white).
That's not particularly explanatory but consider that the energy above refers to the "energy in the noise waveform itself", not the energy required to produce the waveform. Suppose that the energy required for a neuron to oscillate at frequency f is proportional to f, then if every neuron on average uses the same amount of energy, the output energy spectral density is going to be proportional to 1/f, i.e., pink.
I thought we had scientific studies with brain blood-flow analysis, showing that meditation greatly curtails blood flow to a region which helps us locate ourselves (slightly above and behind our eyes). The theory was that both out of body experiences and the "one with everything" phenomenon were likely associated with this. The one time I meditated to the point of, "OMFG What just happened," I would put it more in the line of feeling everywhere than feeling at "one with everything." Maybe that was just the OMFG component though.
Extremely vague speculation, but the golden ratio is the most irrational number, the furthest from any even ratio, and it sounds like that could make a representation of it more consistent? People always talk about it in aesthetics, maybe it's related
It is! Frequency doublings maximize phase alignments but the golden ratio minimizes phase alignments. We discuss it in the neuroscience section of this paper on the concept of harmony.
https://doi.org/10.1016/j.sheji.2022.01.001
I'm reading through the section on the hippocampus and how it functions in both navigation and memory. Does this connection have anything to do with how the "Memory Palace" mnemonic aid works? The one where you think of your house and assign a memory or a task to each room, and this helps you keep it all in mind later in the day.
Edit: Hey, guess what? It's mentioned in the very next sentence after where I stopped to write this comment.
Scott: "... But as far as I know it’s the only book on brain waves ..."
As I mentioned in a comment on your "Highlights" post, I think neuroscientist William Calvin was working and writing on that some 20 years ago:
https://astralcodexten.substack.com/p/highlights-from-the-comments-on-brain/comment/10310008
Among all the oscillating brain waves, each sending and receiving signals to and from multiple recipients, it seems likely there are some neurons for which the sum of incoming signals moreorless somehow cancel each other out, eigen-neurons one might call them, for any given pattern of brainwave.
Think of waves falling on a rocky sea shore, with water swirling around the rock pools, and some will be quiescent at least for a moment, like the centres of miniature gyres. Maybe a memory is formed as a snapshot of the combination of quiescent neurons.
Someone once asked the ancient Greek sculptor Phidias how he carved such wonderfully realistic horses. "Easy", he replied "I start with a big lump of marble, and chip away everything that doesn't look like a horse". I suggest likewise that memories are formed somehow in sets of neurons whose lack of net inward, and thus any outward, signals means that they "look like" a memory, in the same sense as Phidias's comical reply!
"If we observed closely enough, we would find that experience is actually divided into individual moments of consciousness. These conscious “mind moments” occur one at a time, in much the same way that a motion picture film is actually divided into separate frames…"
There is a passage in Ramachandran's Phantoms in the Brain where he describes a patient with some sort of brain damage that perceived the world as a series of still frames, unable to stitch them together into a continuous experience (which made crossing the street, for instance, quite hazardous).
A couple of thoughts: (1) local cortical gamma oscillations are most commonly associated with attention. I think most neuroscientists would define consciousness as being the collection of global brain states that are different from what happens in anesthesia or damaged brains (or potentially sleep)
(2) I think that the jury is still out about whether oscillations are the signal that an invasive measurements of electrical activity detects when certain types of neurons/neural circuits are active, and all the bit about synchrony and chunking follows from the synaptic delays of the neurons involved. But I think that most neuroscientists expect this to be true. Historically folks who've studied the brain noninvasively with EEG for example imagine that the oscillations are meaningful by themselves, but that probably is just measurement bias.
(3) I've not done the experiment, but I suspect that a small antenna placed in or above a densely multicore processor would detect "local field potential oscillations" associated with different pieces of code running, and with enough resolution, associated with particular chip regions. Konrad Kording has a paper on this idea whose conclusions I don't love, but which is interesting in concept. From that perspective, there's nothing special about "wetware" except that the frequencies are a bit lower. Modern neuroscience has revealed that the old idea that a neuron was like a transistor was quite underestimating complexity. A neuron is probably more like a professor core. So it's not unlikely that some sort of machine learning/AI running on millions of cores would exhibit signals much like LFP...
>All of this hints at some deep connection between brain waves, consciousness, and selfhood.
Neuroscience is data rich, but theory poor. You often hear this lament from neuroscientists. What's even more unfortunate is that we actually do have a wonderful, mechanistically precise theory that explains how a specific type of brain waves give rise to consciousness (and selfhood).
Stephen Grossberg, a tragically under-appreciated pioneer in computational neuroscience, has over sixty five years put together a magnificent framework that explains how a limited set of constraints and computational principles can explain consciousness and why it is needed. His work is largely unknown even within neuroscience because he was very early (early 1960s) in realizing that the complex nonlinear dynamics of brains needed a different type of math to capture rhythmic activity (differential equations). He was doing theoretical neuroscience before the term neuroscience entered official usage. Later advances in techniques (fMRI being the worst offender) meant that it was easy to publish research by generating more data than making sense of it using a good theoretical framework.
Coming back to why we need brain waves for consciousness. Here’s a very short and lossy summary (more in the links below) . Please note that the theory itself is mathematical, mechanistically precise, and explains a ton of neuroscientific data. The following is an attempt to translate it into accessible plainspeak that avoids the “the grundicular region modulates output from the parashmoxular cortex” thicket you mention.
We cannot ask what consciousness is, without also asking *who* is experiencing this consciousness. That is, where does the self that experience come from? How do 86 billion neurons (37 trillion cells in the body, really) give rise to selfhood and is consciousness necessary for this? So to explain consciousness, we must explain the self, and to explain the self we must explain, what? Is there an infinite regress?
Grossberg’s Adaptive Resonance Theory helps us understand how to break this up by recasting one in terms of the other.
The Self is simply a constellation of experiences.
Consciousness is the constellation of past experiences experiencing the present, assimilating it to act and prepare for future opportunities.
We now merely have to explain how an experience is constructed. Here’s where the brain waves come in. We are actually incomprehensibly large beings with billions of neurons and trillions of cells with data in some large multiples of those numbers streaming in incessantly. How do we survive this onslaught and make sense of it? Meaning must be collectively manufactured by a very large number of these neurons coming together. Unfortunately, communication is expensive and long-range communication is even more expensive and we have a finite number of calories that we can use up. Synchronizing brain waves – synchrony across any oscillators for that matter - are the most energy efficient way to do large scale communication. This is also why synchrony appears so often in dynamic systems (stadium waves, fireflies, etc). They reorganize to take advantage of this energy efficiency. These brain waves help experiences in different parts of the network (brain) talk to each other if necessary to decide what to do with the sensory data that is streaming in. Note that this is merely data. It only becomes information after this cross-talk within the brain across experiences. This is what is often referred to as top-down expectations. Top-down expectations (experiences) talk to bottom-up inputs (sensory data) and arrive at an agreement through synchronizing brain waves. This is what we consciously experience as experience. Here’s the magical part. Once forged, this new experience becomes a part of the constellation of experiences we recognize as the self, changing us. No man crosses the same river twice. The constellation of experiences changes, changing the next experience.
I expand on this here: https://saigaddam.medium.com/understanding-consciousness-is-more-important-than-ever-7af945da2f0e
I would strongly encourage anyone reading this to attempt reading Grossberg’s article [1] (where he puts it all together) or his book [2], which can be a little daunting.
[1] Towards solving the hard problem of consciousness: The varieties of brain resonances and the conscious experiences that they support
https://www.sciencedirect.com/science/article/pii/S0893608016301800
[2] Conscious Mind, Resonant Brain: How Each Brain Makes a Mind
https://www.amazon.com/Conscious-Mind-Resonant-Brain-Makes/dp/0190070552/
And if you are really averse to wading through grundicular regions and parashmoxular cortices, you can start with our book [3], where we offer a gentle introduction to all this.
[3] Journey of the Mind: How Thinking Emerged from Chaos
https://www.amazon.com/Journey-Mind-Thinking-Emerged-Chaos/dp/1324006579/
That was a great review! You really went to the heart of the matter. Some of your questions are also brought up in this recent article on “Resonance as a Design Strategy for AI and Social Robots.” If you are interested in the brain as an oscillating system, you may appreciate the effort we took to clarify key concepts.
https://www.frontiersin.org/articles/10.3389/fnbot.2022.850489/full
And, “Harmony in Design: A Synthesis of Literature from Classical Philosophy, the Sciences, Economics, and Design” has a section on brainwaves. It reviews the evidence for, e.g., brain waves being structured like octaves. Which they weirdly are—so as to support phase amplitude coupling.
https://doi.org/10.1016/j.sheji.2022.01.001
What do you call the way of thinking that looks for 'purpose' of phenomena, as if emergent phenomena were a mechaism put there as a tool towards some goal? I feel like this approach is like looking at the waves to study fish.
This makes me think about mechanisms of ECT. What if it’s less of a blah blah BDNF story and truly a matter of brain rebooting itself after the post ictal suppression on EEG?
Can confirm: synchronization of brainwaves seems to be what's going on in meditation - well, at least once I had that experience.
I didn't really know anything about brainwaves or -rhythms, but I back then described the experience in terms of things being out-of-phase and slowly getting in sync. More precisely, I got into the rare (for me) state when my attention wouldn't drift from the sensory meditation object virtually at all. Nevertheless, I noticed that there seemed to be small, sub-second gaps between noticing/experiencing the sense data, a sort of frame rate of stimuli. I kept my attention on that "rhythm" of attention and the gaps between, of ~ 5 Hz (that's the number I guesstimated, but hard to really tell) while it slowly was getting faster. Finally it locked into phase with my attention. The result was weird and pleasurable "liquid" compared to "granular" sensation of all stimuli, like I could take in the sense data at the rate it was coming in. Keeping with the frame rate analogy; if you can only render every nth frame, things look and feel janky, while now every frame was rendered.
It's been some years since that experience and I hadn't had it since (among other things, I'm sure I kept trying too hard to repeat it). That time it lasted some hours without effort, I guess just by its "momentum". My instructor's response to my description didn't really help me understand this and I felt I was seeing and describing things from such a different angle that no knowledge could be shared. This review got me interested again in meditation as it seems to give some framework that seems to match my experience. From here, it's possible for me to derive an explanation of a certain kind of meditation that is so simple and concise it can be expressed in a sentence without being too proverbial or abstract:
syncing your attention's frequency with the frequency (or it's multiple) of the object of your attention.