I recently had a chance to talk with Scott Alexander of SSC and ACX fame at a Berkeley meetup this past summer. He’d been watching my videos and had some questions for me. In particular, he had questions about how literally we took the Symmetry Theory of Valence (STV), and whether the counter-examples he had in mind really went against the theory, or were perhaps also explained by it in non-obvious ways. Afterwards, he sent me a draft of his Jhanas and the Dark Room Problem post for me to preview before he published it. I had a look and offered some clarifications in case he wanted to discuss these ideas more deeply. Just a couple days ago he published it. Seeing that the topic could be explored much more deeply, I then asked him if he was ok with me posting (a lightly edited version of) the email I sent him. He said, “of course”. Thus, you can find it below, which I recommend reading after you read his post in ACX.

Note: For people mostly curious about how STV deals with harsh artforms e.g. death metal and Japanoise, we recommend reading Harmonic Society, an article we published in Art Against Art which explains how all kinds of exotic artforms ultimately cash out in (often counter-intuitive) valence effects via messing with the energy parameter and kick-starting cycles of neural annealing (see also “worldview annealing” to make sense of the often outsized effect of transformative festivals in people’s conception of the world). For those extra-curious about psychedelics and the free energy principle, I recommend watching our video specifically on that topic.

Without further ado….

---

Hi Scott!

Thank you for reaching out! And thank you for the conversation on Saturday. […] Please feel free to post your excerpt, but also I am sharing below information that you can use to edit it so that it is a more accurate portrayal of what we are up to (feel free to quote me below or quote any article or video we have online).

I’ll structure this email in the following way: (1) general clarifications about STV, (2) addressing your excerpt specifically, and (3) some of the meeting notes from our conversation in case you find it valuable to remember what we discussed (or what I can remember of it anyhow).

(1) General Clarifications

The first thing to mention is that the Symmetry Theory of Valence (STV) is really easy to strawman. It really is the case that there are many near enemies of STV that sound exactly like what a naïve researcher who is missing developmental stages (e.g. is a naïve realist about perception) would say. That we like pretty symmetrical shapes of course does not mean that symmetry is at the root of valence; that we enjoy symphonic music does not mean harmony is “inherently pleasant”; that we enjoy nice repeating patterns of tactile stimulation does not mean, well, you get the idea…

The truth of course is that at QRI we really are meta-contrarian intellectual hipsters (you know this link of course). So the weird and often dumb-sounding things we say are already taking into account the criticisms people in our people-cluster would make and are taking the conversation one step further. For instance, we think digital computers cannot be conscious, but this belief comes from entirely different arguments than those that justify such beliefs out there. We think that the “energy body” is real and important, except that we interpret it within a physicalist paradigm of dynamic systems. We take seriously the possible positive-sum game-theoretical implications of MDMA, but not out of a naïve “why can’t we all love each other?” impression, but rather, based on deep evolutionary arguments. And we take seriously non-standard views of identity, not because “we are all Krishna”, but because the common-sense view of identity turns out to, in retrospect, be based on illusion (cf. Parfit, Kolak, “The Future of Personal Identity“) and a true physicalist theory of consciousness (e.g. Pearce’s theory) has no room for enduring metaphysical egos. This is all to say that strawmanning the paradigms explored at QRI is easy; steelmanning them is what’s hard. I trust you can make a Titanium Man out of them! 🙂

Now, I am indeed happy to address any mischaracterization of STV. Sadly, to my knowledge very few people outside of QRI really “get it”, so I don’t think there is anyone other than us (and possibly you!) who can make a steelman of STV. My promise is that “there is something here” and that to “get it” is not merely to buy into the theory blindly, but rather, it is what happens when you give it enough benefit of the doubt, share a sufficient number of background assumptions, and have a wide enough experience base that it actually becomes a rather obvious “good fit” for all of the data available.

For a bit of history (and properly giving due credit), I should clarify that Michael Johnson is the one who came up with the hypothesis in Principia Qualia (for a brief history see: STV Primer). I started out very skeptical of STV myself, and in fact it took about three years of thinking it through in light of many meditation and high-energy/high-valence experiences to be viscerally convinced that it’s pointing in the right direction. I’m talking about a process of elimination where, for instance, I checked if what feels good is at the computational level of abstraction (such as prediction error minimization) or if it’s at the implementation level (i.e. dissonance). I then developed a number of technical paradigms for how to translate STV into something we could actually study in neuroscience and ultimately try out empirically with non-invasive neurotech (in our case, light-sound-vibration systems that produce multi-modally coherent high-valence states of consciousness). […]

For clarification, I should point out that what is brilliant (IMO) about Mike’s Principia Qualia is that he breaks down the problem of consciousness in such a way that it allows us to divide and conquer the hard problem of consciousness. Indeed, once broken down into his 8 subproblems, calling it the “hard problem of consciousness” sounds as bizarre as it would sound to us to hear about “the hard problem of matter”. We do claim that if we are able to solve each of these subproblems, that indeed the hard problem will dissolve. Not the way illusionists would have it (where the very concept of consciousness is problematic), but rather, in the way that electricity and lightning and magnets all turned out to be explained by just 4 simple equations of electromagnetism. Of course the further question of why do those equations exist and why consciousness follows such laws remains, but even that could IMO be fully explained with the appropriate paradigm (cf. Zero Ontology).

The main point to consider here w.r.t. STV is that symmetry is posited to be connected with valence at the implementation level of analysis. This squarely and clearly distinguishes STV from behaviorist accounts of valence (e.g. “behavioral reinforcement”) and also from algorithmic accounts (e.g. compression drive or prediction error minimization). Indeed, with STV you can have a brain (perhaps a damaged brain, or one in an exotic state of consciousness) where prediction errors are not in fact connected to valence. Rather, the brain evolved to recruit valence gradients in order to make better predictions. Similarly, STV predicts that what makes activation of the pleasure centers feel good is precisely that doing so gives rise to large-scale harmony in brain activity. This is exciting because it means the theory predicts we can actually observe a double dissociation: if we inhibit the pleasure centers while exogenously stimulating large-scale harmonic patterns we expect that to feel good, and we likewise expect that even if you activate the pleasure centers you will not feel good if something inhibits the large-scale harmony that would typically result. Same with prediction errors, behavior, etc.: we predict we can doubly-dissociate valence from those features if we conduct the right experiment. But we won’t be able to dissociate valence from symmetry in the formalism of consciousness.

Now, of course we currently can’t see consciousness directly, but we can infer a lot of invariants about it with different “projections”, and so far all are consistent with STV:

Of special note, I’d point you to one of the studies discussed in the 2020 STV talk: The Human Default Consciousness and Its Disruption: Insights From an EEG Study of Buddhist Jhāna Meditation. It shows a very tight correspondence between jhanas and various smoothly-repeating EEG patterns (including a seizure-like activity that unlike normal seizures (of typically bad valence) shows up as having a harmonic structure, but does not seem to have a direct conscious correlate – still worth mentioning in this context). Here we find a beautiful correspondence between (a) sense of peace/jhanic bliss, (b) phenomenological descriptions of simplicity and smoothness, (c) valence, and (d) actual neurophysiological data mirroring these phenomenological accounts. At QRI we have observed something quite similar studying the EEG patterns of other ultra-high-valence meditation states […]. I expect this pattern to hold for other exotic high-valence states in one way or another, ranging from quality of orgasm to exogenous opioids. 

Phenomenologically speaking, STV is not only capable of describing and explaining why certain meditation or psychedelic states of consciousness feel good or bad, but in fact it can be used as a navigation aid! You can introspect on the ways energy does not flow smoothly, or how the presence of blockages and pinch points make it reflect in discordant ways, or zone in on areas of the “energy body” that are out of sync with one another and then specifically use attention in order to “comb the field of experience”. This approach – the purely secular climbing of the harmony gradient – leads all on its own to amazing high-valence states of consciousness (cf. Buddhist Annealing). I’ll probably make a video series with meditation instructions for people to actually experience this by themselves first hand. It doesn’t take very long, actually. Also, apparently STV as a paradigm can be used in order to experience more pleasant trajectories along the “Energy X Complexity landscape” of a DMT trip (something I even talked about at the SSC meetup online!). In a simple quip, I’d say “there are good and bad ways of vibing on DMT, and STV gives you the key to the realms of the good vibes” 🙂

Another angle: we can find subtle ways of dissociating valence from e.g. chemicals: if you take stimulants but don’t feel the nice buzz that provides a “working frame” for your mental activity, they will not feel good. At the same time, without stimulants you can get that pleasant productivity-enhancing buzz with the right tactile patterns of stimulation. Indeed this “buzz” that characterizes the effects of many euphoric drugs (and the quality of e.g. metta meditation) is precisely a valence effect, one that provides a metronome to self-organize around and which can feel bad when you don’t follow where it takes you. Literally, one of the core reasons why MDMA feels better than LSD, which feels better than DOB, is precisely because the “quality of the buzz” of each of these highs is different. MDMA’s buzz is beautiful and harmonious; DOB’s buzz is harsh and dissonant. More so, such a buzz can work as task-specific dissonance guide-rails, if you will. Meaning that when you do buzz-congruent behaviors you feel a sense of inner harmony, whereas when you do buzz-incongruent behaviors you feel a sense of inner turmoil. Hence what kind of buzz one experiences is deeply consequential! All of this falls rather nicely within STV – IMO other theories need to keep adding epicycles to keep up.

Hopefully this all worked as useful clarifications. Now let me address your excerpt more specifically:

(2) The Excerpt

The Dark Room Problem in neuroscience goes something like this: suppose the brain is minimizing prediction error, or free energy, or whatever. You can minimize lots of things by sitting quietly in a dark room. Everything will be very, very predictable. So how come people do other things?

The usual workaround is inbuilt biological drives, considered as “set points”. You “predict” that you will be well-fed, so getting hungry registers as prediction error and brings you out of your dark room to eat. Et cetera.

Andrés Gómez Emilsson recently shared a perspective I hadn’t considered before, which is: actually, sitting quietly in a dark room is really great.

Indeed usually the Dark Room causes massive prediction errors (since our model of the world is one where being in a Dark Room is truly not expected!). But these prediction errors feel bad because of the dissonance they induce in our experience (which you can get rid of with drugs or meditation!). If you make the Dark Room an “expected” thing, then eventually it will start feeling great. In fact, something like this happens when you meditate a lot in a dark room and settle in. Alternatively, taking 5-MeO-DMT for the first 10 times can be very disconcerting, as it takes you to “the ultimate void of reality”. It’s surprising and dissonant to “find out” that the void is the ultimate truth (I’m not saying that’s true, just that it feels that way in that state!). But once you’ve done it enough times that you know what to expect, you can in fact receive with two open arms the void of ultimate reality. You learn to expect it and not code it as a prediction error, and then you can deeply, deeply “align” to it, which results in unfathomably positive valence that discharges tons of stored internal stress, the very source of low-level dissonance before the trip (again, STV here fits the data rather nicely). 

The Buddha discussed states of extreme bliss attainable through meditation:

> Secluded from sensual pleasures, secluded from unwholesome states, a bhikkhu enters and dwells in the first jhāna, which is accompanied by thought and examination, with rapture and happiness born of seclusion (Samyutta Nikaya)

I had always figured that “sensual pleasures” here meant things like sex. But I think maybe he just means stimuli, full stop. The meditator cuts themselves from all sensory stimuli, eg by meditating really hard on a single object like the breath and ignoring everything else, and as a result gets “rapture and happiness born of seclusion”.

The serious meditators I know say this is real, meaningful, and you can experience it after a few months of careful practice. You become really good at concentrating on one stimulus and ignoring all other stimuli, and eventually your brain kind of gets “in tune” with that stimulus and it’s really blissful. They say this seems to have something to do with the regularity or predictability of the stimulus; if you’re concentrating really hard on something, regularity/predictability/symmetry is just viscerally very good, better than anything you’ve felt before.

[….]

Something like regularity/predictability/symmetry is really good. So why doesn’t a metronome make you bliss out? Andrés says it’s because you can’t concentrate on it hard enough. It’s not engaging enough to occupy your whole brain / entire sensorium /whatever.

Exactly, this is true. The one twist I’ll add is that the regularity that matters is not, as I explained above, the regularity of the stimuli, but rather, the regularity of the inner state. In fact, I currently think that we can get a glimpse of the true shape of our consciousness precisely by studying how different meditation objects work better or worse for the purpose of meditative absorption! Indeed, more symmetrical objects are easier meditation objects (cf. QRI can steelman “sacred geometry” ꙮ). Likewise, the hallucinations one gets near or close to states of high-absorption are also reflections of our inner shape! (cf. Fire Kasina qualia). 

Symphonies are beautiful, and we intuitively feel like it’s because they have some kind of deep regularity or complicated pattern. But they’re less regular/predictable/symmetrical than a metronome. Andrés thinks this is because they hit a sweet spot: regular/symmetrical/predictable enough to be beautiful, but complex/unpredictable enough to draw and hold our attention. Compare to eg games, which are most fun when they’re hard enough to be challenging but easy enough to be winnable.

Indeed! We need unpredictability in order to disable the boredom mechanism, which prevents us from fully absorbing into patterns (or rather, prevents our experience from shaping itself in a way that perfectly predicts the stimuli – in a way when you reach absorption with a stimuli, you are in fact becoming its “complement” – a shape that can predict it perfectly). An important twist is that prediction errors give rise to energy spikes, and high-energy states of consciousness can give rise to pleasant resonance (think about the bodily euphoria that comes from eating spicy enough hot peppers). Plus, cooling down from high-energy states can lead to euphoric neural annealing (as explained here). In all cases, however, the thing that is the most closely related to valence is the regularity/smoothness of the internal (instantaneous) state, even though there might be other complex dynamics guiding the state from one configuration to another. 

But this sweet spot is the fault of your own inattentiveness. If you could really concentrate on the metronome, it would be even more blissful than the symphony. Emilsson says he’s achieved these levels of concentration and can confirm. I talked to another meditator who agrees metronomes can be pretty blissful with the right amount of (superhuman) focus, although – as per the Buddha quote above – total silence is best of all.

Agreed! An interesting note is that the first time this happened to me, it was not in meditation, but during a sleep paralysis! See: Dream Music where I discuss how a simple tone can sound amazing if you are in the right reverb-filled state of mind. See also: people with anhedonia often report feeling “back to normal” in dreams, and IMO that’s precisely because the neuroacoustic profile of dreams can be very reverb-filled and thus have significant valence effects (see below).

I find this to be an elegant explanation of what the heck is going on with jhanas, more convincing than my previous theory. It’s also a strong contender as a theory of beauty – a little different in emphasis from Schmidhuber’s theory, but eventually arriving at the same place: beauty is that which is compressible but has not already been compressed.

A brief comment here: your “going loopy” theory foreshadows our tracer tool and psychedelic cryptography, where psychedelics seem to activate a “delay overlay” of recent experiences on top of the current one. DMT gives rise to ~30hz loops, LSD to around ~18hz loops, and 2C-B closer to ~10hz loops. We hypothesize that there are a discrete number of serotonin-mediated metronomes that precisely modulate the degree to which experience is fed back to itself with a specific delay. Antidepressants may flatten affect by disrupting these loops, and thus eliminating sources of symmetry for the inner state. People describe the sense of “missing an echo”; quite literally having a more “flat” experience as a result!

Enhanced neuroacoustics (as with psychedelics) generally increase the range of valence because more loopy experiences are more intense and also more capable of pure dissonance or pure consonance. Dissociatives (nmda antagonism more generally) seem to instead do a low-frequency looping (around 8hz) together with a generalized reverb effect. Much as in music, adding reverb to almost literally anything makes it sound less harsh (like the baby crying vs. baby crying + reverb sounds I referenced in the presentation). And also much as in music, *compounding* delay and reverb effects gives rise to synergistic outcomes, often with crazy standing wave attractors (e.g. exactly what you see on LSD + nitrous or LSD + ketamine).

Importantly, STV is *not* a theory that lives at the computational or algorithmic level of analysis, which is unlike Schmidhuber’s theory. If I recall correctly, Schmidhuber’s theory doesn’t even care about phenomenal valence or consciousness. And it has no mechanism of binding or any sense of how the “reward” is implemented or who or what receives such reward. Its flavor is functionalist and concludes that beauty is to be found in the act of compression. But STV instead says that compression is merely correlated with valence: our brains are set up in such a way that making excellent compressions reduces dissonance! This is because there is (a) an inherent dissonance cost to complexity, and (b) there is a dissonance cost to prediction errors. But again, take the right drug, and all of a sudden you can experience high-valence while making tons of prediction errors or having models that are much more complex than the sensory data would suggest is necessary.

In particular, what makes good compressions feel good beyond reducing prediction errors is that they select for internal states that have simple sets of symmetries as the best generators which anticipate the stimuli. This is highly related to the concept of Harmonic Entropy (i.e. the entropy of the inner state, not of the stimuli). And here is where we find a stark and amazing difference between STV and compression drive: we in fact expect there to be a sort of “minimal construction” path where you get specific “complexity scores” for phenomenal objects based on the number of operations of the sort the brain can do that are needed to construct such phenomenal objects. The brain needs to explicitly render phenomenal objects, rather than merely encode them. So there is a harmonic entropy associated with each experience, which more-or-less correlates with Kolmogorov complexity but is different in that it uses resonance as the building block rather than arbitrary operations. We also predict that the valence associated with specific patterns of stimulation will be best correlated with a sort of “perceptual harmonic entropy” than with complexity in general: how well you can compress an input depends on what building blocks you have to reconstruct it. In the case of the brain, the building blocks seem to be patterns of resonance. So even if something is “highly compressible” but cannot be compressed with resonance (e.g. the prime numbers), you will not experience it as beautiful or “easy on the eye”.

Importantly, free energy minimization is a computational level analysis and we would say at QRI that it therefore is mistaken on “where to look”. Consequence: compressing information feels good *because* it often (but not always) reduces dissonance. But if your brain is set up in the wrong way, minimizing dissonance may not lead to good compressions, or doing good compressions may not in fact feel good. But reducing dissonance will always feel better, and having high-energy high-harmony patterns internally will always feel good. What this does at the algorithmic and computational level is tricky, but it generally implies that we can see “artifacts” of our resonance-based compression system all over the place when in exotic states of consciousness, which is what we observe (and at the end of the day this may explain why psychedelic fractals and Indra’s Net type experiences are so hedonically loaded! See: psychedelics and the free energy principle).

To sum it up: STV claims that what matters is the regularity of the conscious experience and not of the stimuli – the stimuli is only in a certain sense a “projection” of the inner state, but it can deviate from it in many ways. Prediction errors feel bad because our brain is set up in such a way that they cause dissonance. And compressions only feel good to the extent that they avoid prediction errors *and* minimize the internal dissonance cost of the internal representations used for prediction. In other words, STV explains the other theories, but not the other way around.

Finally…

(3) Meeting notes, in case you find them useful…

• You asked if I knew whether taking a lot of 5-MeO-DMT is compatible with sanity for most people: the answer is probably not. That said, we do know of some notable exceptions of very smart and sane individuals who have experimented heavily with the drug with no obvious cost to their sanity (e.g. see conversation with Ingram, Yang, McMullen, and Taft which touches upon the effects of daily use of 5-MeO-DMT).
• I brought up anti-tolerance drugs, of which black seed oil is promising (but a low-tier player). Most promising of all are ibogaine and proglumide. Opioids + anti-tolerance drugs are IMO the most promising long-term therapy for severe chronic pain. 
• Brought up Chanca Piedra as a promising highly cost-effective intervention to prevent kidney stones in the context of “hell must be destroyed” (other interventions we are fairly confident are in the category of “enormous if true, seem true, yet nobody is trying”: flumazenil for benzos, cooling gloves for MDMA neurotoxicity, DMT for cluster headaches, etc.)
• I shared that the three goals/pillars of QRI are to (1) reduce negative extremes, (2) increase baseline, and (3) achieve new heights.
• I gifted you a High-Entropy Alloy with unique material properties: much like looking for names of God, or amazing new perfumes, or indeed finding gems hidden in the state-space of consciousness, we must figure out clever ways of exploring large combinatorial spaces without going insane. You obviously have thought a lot about this 🙂
• We talked about “nausea annealing” (pro-tip: drinking ginger juice right when you take a psychedelic seems to drastically reduce how much nausea and body-load it causes).
• Overfitting.
• Information is in the coupling between harmonics; de-couple them and you can experience the “zero state” while still being awake.
• Meaning of the QRI logo (i.e. having a clear view of the entire state-space of consciousness; bridging quality/color and quantity/lines as a symbol for qualia formalism).
• You asked “can you tell me what are brainwaves in simple terms a child could understand?” (I answered with “they are the signature of resonance in the holistic field behavior of experience” and immediately realized I had miserably failed to “explain in simple terms”). 
• “Would listening to a pure tone be blissful?” Yes, you can absorb yourself into it. Second half of an orchestral song repeated could be better if it allows you to go deeper into absorption. Touched upon: Boredom mechanism. Harmonic entropy. Controlling for energy.
• 3D harmonics, STV presentation, symmetry of the mathematical object is what truly matters.
• Encoding vs. rendering.

Ok, that was rather long; I hope that you found it useful and clarifying! Please feel free to ask any questions and I promise I won’t send you another equally long email 🙂 Again, feel free to write about any and all of this.

Best of luck in your travels! 🙂

Infinite Bliss!