Opentheory.net

syndicated content powered by FeedBurner

FeedBurner makes it easy to receive content updates in My Yahoo!, Newsgator, Bloglines, and other news readers.

Learn more about syndication and FeedBurner...

A message from this feed's publisher:

Current Feed Content

A new theory of Open Individualism

Posted:Sat, 01 Sep 2018 16:30:53 +0000
My colleague Andrés recently wrote about various theories of personal identity, and how a lack of a clear consensus here poses a challenge to ethics. From his post:

Personal Identity: Closed, Empty, Open

In Ontological Qualia I discussed three core views about personal identity. For those who have not encountered these concepts, I recommend reading that article for an expanded discussion.

In brief:

1. Closed Individualism: You start existing when you are born, and stop when you die.

2. Empty Individualism: You exist as a “time-slice” or “moment of experience.”

3. Open Individualism: There is only one subject of experience, who is everyone. [Ed: this draws from Daniel Kolak‘s work]

Most people are Closed Individualists; this is the default common sense view for good evolutionary reasons. But what grounds are there to believe in this view? Intuitively, the fact that you will wake up in “your body” tomorrow is obvious and needs no justification. However, explaining why this is the case in a clear way requires formalizing a wide range of concepts such as causality, continuity, memory, and physical laws. And when one tries to do so one will generally find a number of barriers that will prevent one from making a solid case for Closed Individualism.

As an example line of argument, one could argue that what defines you as an individual is your set of memories, and since the person who will wake up in your body tomorrow is the only human being with access to your current memories then you must be it. And while this may seem to work on the surface, a close inspection reveals otherwise. In particular, all of the following facts work against it: (1) memory is a constructive process and every time you remember something you remember it (slightly) differently, (2) memories are unreliable and do not always work at will (e.g. false memories), (3) it is unclear what happens if you copy all of your memories into someone else (do you become that person?), (4) how many memories can you swap with someone until you become a different person?, and so on. Here the more detailed questions one asks, the more ad-hoc modifications of the theory are needed. In the end, one is left with what appears to be just a set of conventional rules to determine whether two persons are the same for practical purposes. But it does not seem to carve nature at its joints; you’d be merely over-fitting the problem.

The same happens with most Closed Individualist accounts. You need to define what the identity carrier is, and after doing so one can identify situations in which identity is not well-defined given that identity carrier (memory, causality, shared matter, etc.).

But for both Open and Empty Individualism, identity is well-defined for any being in the universe. Either all are the same, or all are different. Critics might say that this is a trivial and uninteresting point, perhaps even just definitional. Closed Individualism seems sufficiently arbitrary, however, that questioning it is warranted, and once one does so it is reasonable to start the search for alternatives by taking a look at the trivial cases in which either all or none of the beings are the same.

More so, there are many arguments in favor of these views. They indeed solve and usefully reformulate a range of philosophical problems when applied diligently. I would argue that they play a role in philosophy that is similar to that of conservation of energy in physics. The energy conservation law has been empirically tested to extremely high levels of precision, which is something which we will have to do without in the realm of philosophy. Instead, we shall rely on powerful philosophical insights. And in addition, they make a lot of problems tractable and offer a powerful lens to interpret core difficulties in the field.

Open and Empty Individualism either solve or have bearings on: Decision theory, utilitarianism, fission/fusion, mind-uploading and mind-melding, panpsychism, etc.

Andrés goes on to discuss David Benatar’s argument for antinatalism— the view that it is ethically preferable for individuals to never be born– and suggests arguments in this space tend to “rely implicitly on personal identity background assumptions. In particular, antinatalism is usually framed in a way that assumes Closed Individualism.” Furthermore, since Closed Individualism is on shaky philosophical ground, Benatar’s argument for antinatalism is likewise questionable.

There’s much more there, and I strongly endorse Andrés’s core theme, that you can’t get ethics right if you don’t get personal identity right, and that most ethical arguments right now assume a theory of identity (Closed Individualism) which breaks in illegible ways if we try to apply it in novel contexts.

At the same time, however, it feels like the distinction between Open Individualism (OI) and Empty Individualism (EI) is merely a distinction in name only; both theories of identity give identical answers for essentially all practical and ethical queries.

The following are some thoughts on one possible way to conceptualize this difference such that OI and EI would definitely point to different things, and give satisfyingly different answers to queries. (Status: exploratory, not strongly held.)

OI and EI are usually phrased in terms of identity— OI says everything is the same thing, and EI says everything is a different thing. But it’s not clear what this claim about identity *means*, how it cashes out in ontological and ethical senses. And it seems like the core issue is with the ambiguity involved in OI’s insistence that ‘everything is one’. What does it mean to say everything is one, that there is only one subject of experience? It’s not clear that this claim is phrased in a way that ‘pays rent‘.

Hypothesis:

One way of sharpening the definition of Open Individualism is to consider its core claim not (only) at the level of identity, but on the level of experience. That is: we could take Open Individualism to assert that phenomenal reality is, in the most literal sense, one huge qualia-bundle, and although it seems like this qualia-bundle has partitions or boundaries, these apparent partitions are illusions. Phenomenal binding, on the other hand, *is* real— but on only the *grandest* scale; absolutely everything is bound together. Everything is ontologically unitary, in all important senses. (This brings to mind Wheeler’s One Electron Universe hypothesis, where reality is ontologically unitary, but somehow an illusion of locality and diversity-of-forms arises.)

But why don’t we experience reality as this grand oneness-without-boundaries? Well, perhaps we do and we don’t know it. An important and unsolved question in qualia research is how to handle the reportability of qualia. A move often used by functionalists is that the apparent differences between the subjective experiences of people might *most precisely* boil down to differences in the substructure of the brain processes they use to report their internal states. This move is often used to try to explain qualia away, but perhaps the functionalists *and* the Open Individualists could be right– the universe could be one big global qualia-bundle, but individuals might only be able to report things about the local causal microstructure of the brain/mind. (I note this feels weird and counterintuitive, but not ruled out by the evidence, and as I mention later it might be indicated by certain elegance considerations.)

This definition of Open Individualism gives nearly everybody something that they want— OIs and Buddhists get everything-being-one (in the strongest sense!), functionalists get a heightened focus on the mechanisms in the brain that generate qualia reports (the ‘easy problems’ of consciousness), formalists get qualia formalism as ground-truth (albeit only at the largest scale), and so on.

But nobody gets everything that they want— in particular, functionalists don’t get to eliminate talk about qualia, and formalists have to worry a *lot* about the nature and specific mechanics of qualia reports, and might need to accept there’s no way to get to ‘local ground truth’ about qualia (epistemologically speaking).

Philosophical motivation:

To justify this way of contrasting OI and EI, we can look at their core difference in terms of monads (From Leibniz, “an indivisible and hence ultimately simple entity”). If monads do exist, then they probably exist at one scale; there probably aren’t two ‘flavors’ of monads. No hierarchy of indivisible things made up of other indivisible things. And so, if Empty Individualism is the case, monads, the-things-that-are-ontologically-unitary, the-things-that-are-indivisible, should be defined in terms of experience-slices. Those would *be* the monads. But if Open Individualism is the case, there’s exactly one monad, and it’s the universe.

In short, I’m led to the following definitions of Open Individualism and Empty Individualism: EI says the universe has lots of boundaries, is made of many monads. OI says the universe has no internal boundaries, is made of one monad. Either of these seems more elegant than the current haphazard definition of Open Individualism, that the universe is ontologically unitary but has distinct parts, is one monad made of smaller monads, is one big indivisible thing made up of a lot of smaller indivisible things. (I am still open to the logical possibility of OI with real partitions, but it does seem less elegant than the definition I’m offering, and elegance arguments seem really important here.)

Implications:

This difference between OI and EI would have deep implications for what kinds of knowledge we can expect qualia research to generate, and what sorts of methods might offer reliable data. Basically, EI would be a lot easier to manage; being able to divide and conquer is a key enabling factor for scientific progress. Easier to study the structure of reality if there are many small monads of bounded complexity to study and compare, rather than one big monad with only very fuzzy internal partitions.

In terms of ethics, if this version of Open Individualism is true, it’d be deep justification for utilitarianism: let’s be good to each other, because We Are The Same Thing.

But what’s true?

How do we pick between Closed Individualism, Empty Individualism, and Open Individualism? First, I think it’s important to note there seems to be a significant difference between truth & usefulness regarding theories of identity. It seems as though evolution has biased us toward Closed Individualism as sort of a ‘Goldilocks zone’ of selflessness vs selfishness:
- EIs have trouble coordinating because there aren’t compelling Schelling points for its momentary-slices-of-experience to cooperate with each other;
- OIs have trouble coordinating because their natural Schelling point is always cooperate with everything (since I Am You and You Are Me), which fails in the presence of possible defectors, and this ‘game-theoretic naivety’ prevents positive-feedback loops from forming;
- Closed Individualism provides an effective middle path between coordination and defection: it allows an organism to reliably coordinate with its past and future selves (thus constituting a temporal superorganism), while at the same time finding it natural to be able to strategically cooperate with or defect against other such temporal coalitions.
I think OIs and EIs who want to build effective coordination mechanisms should take note of how powerful of a strategy CI is. That said, just because something like CI has been evolutionarily useful as a coordination strategy doesn’t mean it’s metaphysically true as a theory of identity.

What is “metaphysically true”? I suspect we can’t use the traditional method of picking theories (judging them by their predictive power) so instead I think we have to rely on elegance arguments. As Andrés suggests, I think we can already disqualify Closed Individualism here: for CI to be crisply true, there’d need to be a crisp carrier of identity, which seems less and less likely the more we learn about reality. But how do we pick between EI and OI? Essentially we may need to wait until we’ve solved consciousness (determined the precise formalism for qualia) and see which seems simpler, which seems to naturally ‘pop out of the equations’.
A future for neuroscience

Posted:Tue, 14 Aug 2018 01:32:20 +0000
I think all neuroscientists, all philosophers, all psychologists, and all psychiatrists should basically drop whatever they’re doing and learn Selen Atasoy’s “connectome-specific harmonic wave” (CSHW) framework. It’s going to be the backbone of how we understand the brain and mind in the future, and it’s basically where predictive coding was in 2011, or where blockchain was in 2009. Which is to say, it’s destined for great things and this is a really good time to get into it.

I described CSHW in my last post as:

Selen Atasoy’s Connectome-Specific Harmonic Waves (CSHW) is a new method for interpreting neuroimaging which (unlike conventional approaches) may plausibly measure things directly relevant to phenomenology. Essentially, it’s a method for combining fMRI/DTI/MRI to calculate a brain’s intrinsic ‘eigenvalues’, or the neural frequencies which naturally resonate in a given brain, as well as the way the brain is currently distributing energy (periodic neural activity) between these eigenvalues.

This post is going to talk a little more about how CSHW works, why it’s so powerful, and what sorts of things we could use it for.

CSHW: the basics

All periodic systems have natural modes— frequencies they ‘like’ to resonate at. A tuning fork is a very simple example of this: regardless of how it’s hit, most of the vibration energy quickly collapses to one frequency- the natural resonant frequency of the fork.

All musical instruments work on this principle; when you change the fingering on a trumpet or flute, you’re changing the natural resonances of the instrument. In the video below you can sort of ‘see’ this resonance:

Here we see time-averaged standing waves (resonance) on the front plate of a guitar:

And here are some of the elegant mathematical relationships between the notes a guitar string is made to resonate at (assuming a ‘just temperament’ tuning):

CSHW’s big insight is that brains have these natural resonances too, although they differ slightly from brain to brain. And instead of some external musician choosing which notes (natural resonances) to play, the brain sort of ‘tunes itself,’ based on internal dynamics, external stimuli, and context.

The beauty of CSHW is that it’s a quantitative model, not just loose metaphor: neural activation and inhibition travel as an oscillating wave with a characteristic wave propagation pattern, which we can reasonably estimate, and the substrate in which they propagate is the the brain’s connectome (map of neural connections), which we can also reasonably estimate.

Rough map of a human connectome. (Source)

This means we can calculate (fairly) precisely which frequencies will be naturally resonant in a given brain. And Atasoy has done just that, as this fancy graphic shows:

(a) shows the Laplacian eigenfunctions of different shapes (where the natural standing waves form); (b) shows the process of calculating the brain’s ‘shape’ such that we can apply this equation to the brain; (c) is a list of the resulting harmonics. The workflow is basically three steps: first combine MRI and DTI to approximate a brain’s connectome, then with an empirically-derived wave propagation equation calculate what the natural harmonics are of this connectome, then estimate which power distribution between these harmonics would most accurately reconstruct the observed fMRI activity.

Why is a harmonic analysis of the brain so powerful?

In my last post I mentioned that:

[I]t seems a priori plausible that systems like brain with significant periodicity will self-organize around their eigenvalues; i.e. these eigenvalues will be functionally significant and ‘costless’ Schelling points. This implies that these harmonics will be a good place to start if we want to efficiently compress a lot of the brain’s (and mind’s) complexity.

I would add that harmonic analysis of the brain is particularly powerful because harmonics follow highly elegant mathematical rules, and insofar as the brain self-organizes around them, the rest of the brain will have a hidden elegance, a hidden simplicity, to it as well.

The problem facing neuroscience in 2018 is that we have a lot of experimental knowledge about how neurons work– and we have a lot of observational knowledge about how people behave– but we have few elegant compressions for how to connect the two. CSHW promises to do just that, to be a bridge from bottom-up neural dynamics – things we can measure – to high-level psychological/phenomenological/psychiatric phenomena – things we care about. And a bottom-up bridge like this should also allow continuous improvement as our understanding of the fundamentals improve, as well as significant unification across disciplines: instead of psychology, psychiatry, philosophy, and so on each having their own (slightly incompatible) ontologies, a true bottom-up approach can unify these different ways of knowing and serve as a common platform, a lingua franca for high-level brain dynamics.

(Also on the topic of building theoretical bridges, see my recent thoughts on how we could approach unifying CSHW with other paradigms like IIT & FEP.)

What types of new things could we do with CSHW?

CSHW is still a very young paradigm, and so far research has reasonably focused on the foundations: how it works and straightforward applications. Atasoy & coauthors have looked at how the natural power distribution between harmonics, how this changes between resting state brain activity vs psychedelic (LSD) brain activity, and how this pushes the brain towards self-organized criticality. All novel and important results. But if CSHW is as promising as it seems to me, this initial frame dramatically undersells the potential of the paradigm.

The following is a compilation of about a year’s worth of thoughts about what could be done with CSHW. It assumes a reasonable conceptual familiarity with the CSHW paradigm; for more background, see this video and these papers.

I. Proxy phenomenological structure.

The Holy Grail of various brain sciences (and a topic of great interest to QRI) is a clear, intuitive method for connecting what’s happening in the brain to what’s happening in the mind. CSHW is interesting here since the brain likely self-organizes around its characteristic frequencies, so changes in these frequencies (and the distribution of power between them) should ripple through phenomenology, likely in predictable ways. Insofar as this ‘harmonic scaffolding’ hypothesis is true, CSHW could form the seed to a formal science of phenomenology.

To me, the most obvious starting point is QRI’s Symmetry Theory of Valence (STV), that consonance between the natural harmonics of a brain will be a good proxy for that experience’s overall degree of pleasantness (Johnson 2016; extended by Gomez Emilsson 2017). Atasoy herself is interested in explaining psychedelic effects as an increase in criticality and as shifts in the power distribution between harmonics (Atasoy et al. 2016; 2017). A natural fusion of these approaches is to parametrize the effects (and ‘phenomenological texture’) of all psychoactive drugs in terms of their effects on the consonance, dissonance, and noise of a brain, both in overall terms and within different frequency bands (Gomez Emilsson 2017).

A “CDNS” (Consonance/Dissonance/Noise Signature) evaluation of brain harmonics, and its resultant valence (Atasoy et al. 2016 + Johnson 2016 + Gomez Emilsson 2017). Image credit: Andrés Gomez Emilsson

In the long term, we’ll want to move upstream and predict connectome-specific effects of drugs- treating psychoactive substances as operators on neuroacoustic properties, which produce region-by-region changes in how waves propagate in the brain (and thus different people will respond differently to a drug, because these sorts of changes will generate different types of results across different connectomes). Essentially, this would involve evaluating how various drugs change the internal parameters of the CSHW model, instead of just the outputs. Moving upstream like this might be necessary to predict why e.g. some people respond well to a given SSRI, while others don’t (nobody has a clue how this works right now).

II. Understand and improve mental health

The most important future theme in psychiatry is “What you can measure, you can manage.” If connectome harmonics are as tightly coupled with phenomenology as we think, then we may be able to identify a target emotional state for someone, identify what about their connectome harmonics is different from that state, and use this to inform us how to push the system toward that state.

I’m increasingly suspecting that many psychiatric illnesses will leave a semi-unique fingerprint on someone’s connectome harmonics. Furthermore, connectome harmonics may be sufficiently coupled to the causal structure of these illnesses that adjusting the system to eliminate this fingerprint may actually cure the illness. It would be difficult for me to overstate how important I think this is.

In Quantifying Bliss, my colleague Andrés discusses using CSHW to parametrize and measure mental health (with real-time debugging):

The “clinical phenomenologist” of the year 2050 might look into your brain harmonics, and try to find the shortest paths to nearby state-spaces with less chronic dissonance, fishing for high-consonance attractors with large basins to shoot for. The qualia expert would go on to provide you various options that may improve all sorts of metrics, including valence, the most important of them all. If you ask, your phenomenologist can give you trials for fully reversible treatments. You sample them in your own time, of course, and test them for a day or two before deciding whether to use these moods for longer.

Using CSHW (Atasoy et al. 2016) + CDNS (Gomez Emilsson 2017) + STV (Johnson 2016) to adjust brain activity to a more pleasant state. Image credit: Andrés Gomez Emilsson.

Likewise, if we build a suite of methods for parametrizing and replicating phenomenological states with CSHW, we could presumably use this to replicate the psychoactive (and psychedelic) effects of various drugs, without the drugs and without their broad-spectrum side-effects. Opioid painkillers without the chemical addictiveness; MDMA without the neurotoxicity. And new drug effects that current pharmaceuticals are unable to create. I call this class of interventions “patternceuticals” in Principia Qualia.

Finally, it’s not unreasonable to think that we may be able to use CSHW to better understand limited aspects of someone’s physical health, as well. For example, inflammation looks like it can cause depression, but perhaps it goes the other way too: depression causing inflammation, in a very precise and mechanistic way. If we’re right that harmony is the natural homeostatic state of our brain, then persistent dissonance would indicate a threat or injury, something to mobilize resources against and fight, and the dissonance itself may be sufficient to kickstart this defensive process. In this case we should expect to find mechanisms which activate both global hormones like cortisol, and local defense mechanisms like nearby microglia releasing cytokines, in response to irregular (dissonant) neural firings, perhaps with high-energy beat patterns as the specific trigger. The core intuition here is that connectome harmonics will have consequences at all levels of biology, and some of these consequences can be predicted a priori.

III. New science of psychometrics & psychodynamics

For over a hundred years now, we’ve been trying to figure out how to systematize the structure of variation between peoples’ minds and brains. This has led to a slew of psychometric frameworks, the two most robust being IQ and the Big 5 personality dimensions (Openness, Conscientiousness, Extroversion, Agreeableness, Neuroticism). Both are top-down frameworks built on observational data and factor analysis.

But with CSHW, if we’re able to more elegantly model the ‘neurological natural kinds’ which generate our cognitive, affective, and social, and phenomenological dynamics in bottom-up ways, we should be able to improve on these metrics, and generate new metrics for interesting dimensions of variation that have eluded formal measurement so far.

From IQ to NaQ

As an opening move, I’d suggest that we could reconceptualize intelligence as NaQ (neuroacoustic quotient), or ‘the capacity to cleanly switch between different complex neuroacoustic profiles.’ This would envision the brain as analogous to a musical instrument that can (nigh-instantly and perfectly) retune itself from one complex key signature (set of connectome harmonics) to another, where the key signature corresponds to the parameters of some problem domain and harmony in this key signature corresponds to solving the task at hand.

Higher intelligence would encompass quicker & cleaner transitions, more complex neuroacoustic ‘key signatures,’ more flexibility and adaptability in configuration, greater numbers of subpartitions (orthogonal sets of connectome harmonics) within the brain and lower amounts of leakage between these subpartitions, and better harmony engineering (making the computation ‘flow’ in the right direction). Some of these capacities will be improved by practice and life experience (‘crystallized intelligence’) and some won’t (‘fluid intelligence’).

Steve Lehar has speculated that this sort of harmony-based computation could be implemented by modeling perceptions as spatio-temporal constraints on frequencies which in turn constrain the possible resonances of the system, much like how putting a clamp on a chladni plate constraints its resonances. There’s much more that could be said here about merging Lehar’s intuition with Atasoy’s CSHW, but we need not speculate too much about implementation to say interesting things about psychometrics.

I see two research angles here: we could start with the current testing methodology for IQ, and try to reverse-engineer what neuroacoustic properties each subtest is effectively testing for (Verbal Comprehension, Perceptual Reasoning, Working Memory, Processing Speed). This seems easiest. We could also start from the literature on CSHW and self-organizing (harmonic) systems, identify core principles (e.g., conditional metastability, integrated information, symmetry/harmony as success condition, orthogonalization of harmonics) that seem relevant for intelligence, and try to remake NaQ from the bottom up. This seems better long-term.

What’s the point? I suspect that understanding the algorithmic implementation of intelligence could help us better see what IQ tests are actually measuring, how to improve them such that they reflect the contours of reality more closely, and perhaps eventually how to improve (or how to prevent modern life from degrading!) various subtypes of intelligence. The value of even incremental advances here could be large.

An interesting variable is how much external noise is optimal for peak processing. Some, like Kafka, insisted that “I need solitude for my writing; not ‘like a hermit’ – that wouldn’t be enough – but like a dead man.” Others, like von Neumann, insisted on noisy settings: von Neumann would usually work with the TV on in the background, and when his wife moved his office to a secluded room on the third floor, he reportedly stormed downstairs and demanded “What are you trying to do, keep me away from what’s going on?” Apparently, some brains can function with (and even require!) high amounts of sensory entropy, whereas others need essentially zero. One might look for different metastable thresholds and/or convergent cybernetic targets in this case.

A nice definition of psychological willpower falls out of this paradigm as well: the depletable capacity to adjust harmonics, perhaps against a gradient of increasing Free Energy.

From emotional intelligence to EnQ & MQ

EQ (emotional intelligent quotient) isn’t very good as a formal psychological construct- it’s not particularly predictive, nor very robust when viewed from different perspectives. But there’s clearly something there– empirically, we see that some people are more ‘tuned in’ to the emotional & interpersonal realm, more skilled at feeling the energy of the room, more adept at making others feel comfortable, better at inspiring people to belief and action. It would be nice to have some sort of metric here.

I suggest breaking EQ into entrainment quotient (EnQ) and metronome quotient (MQ). In short, entrainment quotient indicates how easily you can reach entrainment with another person. And by “reach entrainment”, I mean how rapidly and deeply your connectome harmonic dynamics can fall into alignment with another’s. Metronome quotient, on the other hand, indicates how strongly you can create, maintain, and project an emotional frame. In other words, how robustly can you signal your internal connectome harmonic state, and how effectively can you cause others to be entrained to it. Empirically, women seem to have a higher EnQ (and are generally more sensitive to the energy in a room), whereas MQ might be more similar on average, with men being slightly higher (especially on the tails). Most likely, these are reasonably positively correlated; in particular, I suspect having a high MQ requires a reasonably decent EnQ. And importantly, we can likely find good ways to evaluate these with CSHW.

The neuroscientific basis for the Big 5

The Big 5 personality framework (Openness, Conscientiousness, Extroversion, Agreeableness, Neuroticism) is one of the crown jewels of psychology; it’s robust, predictive, and intuitive. But it’s a top-down construct born of factor analysis, not fundamentally grounded in neuroscience.

I don’t think the Big 5 fits cleanly under any one neuroscientific paradigm. Instead, we can take a grab-bag approach. First, there’s an interesting paper that attempts to describing each dimension in terms of cybernetic control theory, where the brain tries to keep certain internal proxies within some range. I think this is a reasonable explanation for some dimensions but not others. There’s also interesting work from affective neuroscience where the Big 5 are generated by different combinations of ‘axiomatic’ primal emotional drives.

But if I had to spitball it from the perspective of CSHW, I’d suggest the following:
- Openness: degree of canalization of the brain’s initial harmonic key signature.
- Conscientiousness: strength of internal drive toward consistency across connectome harmonic partitions.
- Extroversion: generally formulated as reward sensitivity; perhaps we could also model it as ease of switching between high-entrainment & high-metronome modes in social interactions; alternatively, whether interaction with others adds or drains Free Energy.
- Agreeableness: multifactorial; part of it seems to just be EnQ. I suspect this might be the least-robust factor of the Big 5.
- Neuroticism: degree to which one’s harmonic key signature is biased toward dissonance.
Autism might be reconceptualized along two dimensions: first, most forms of autism would entail less general ability to reach interpersonal entrainment with another’s connectome harmonics- a lower EnQ. Second, most forms of autism would also entail a non-standard set connectome harmonics. I.e., the underlying substructure of core harmonic frequencies may be different in people on the autism spectrum, and thus they can’t effectively reach social entrainment with ‘normal’ people, but some can with other systems (e.g. video games, specific genres of music), and some can with others whose substructure is non-standard in the same way. We can think of this as a game-theoretic strategy: groups which have some ‘harmonic diversity’ might find some forms of emotional and semantic synchronization more difficult, but this would preserve cognitive, behavioral, and social diversity, and produce more net comparative advantage.

The mathematics of signal propagation and the nature of emotions

High frequency harmonics will tend to stop at the boundaries of brain regions, and thus will be used more for fine-grained and very local information processing; low frequency harmonics will tend to travel longer distances, much as low frequency sounds travel better through walls. This paints a possible, and I think useful, picture of what emotions fundamentally are: semi-discrete conditional bundles of low(ish) frequency brain harmonics that essentially act as Bayesian priors for our limbic system. Change the harmonics, change the priors and thus the behavior. Panksepp’s seven core drives (play, panic/grief, fear, rage, seeking, lust, care) might be a decent first-pass approximation for the attractors in this system.

On harmonic canalization and emotional key signatures

The core phenomena which self-organizing harmonic systems leverage to self-organize is the ratio between their internal frequencies. As such, initial conditions matter, and in particular since low-frequency harmonics tend to entrain and synchronize high-frequency activity, the specific frequencies of ‘base tone’ (emotional) harmonics will determine much about which higher-frequency harmonics are consonant vs dissonant, and may highly constrain the possibility space of a given brain’s harmony gradient. This calls to mind my favorite passage from Kahlil Gibran: “We choose our joys and sorrows long before we experience them.” Perhaps it is not so much us doing the choosing, as the emotional key signature generated by the coldly-beautiful mathematics of our base harmonics.

This is not to say our key signatures are completely static, however: an interesting thread to pull here may be that some brains seem to flip between a major key and a minor key, with these keys being local maximas of harmony. I suspect each is better at certain kinds of processing, and although parts of each can be compatible with the other, each has elements that present as defection to the internal logic of the other and so these attractors can be ‘sticky’. But there can also be a buildup of tension as one gathers information that is incompatible with one’s key signature, which gets progressively more difficult to maintain, and can lead to the sort of intensity of experience that drives an annealing-like process when the key signature flips. And in the case of repeated flips, the patterns which are compatible with both key signatures will be the most strongly reinforced.

Initial conditions in low-frequency harmonics may also constrain the possible ways subsets of brain harmonics can be separated– e.g. Pletzer et al. 2010 observed that the golden ratio (1:1.618) is the theoretically most effective ratio for preventing cross-frequency coupling, and this may drive observed phenomena across various EEG bands (Klimesch 2012). I suspect this orthogonalization process is enormously important and complex, and may be one of the strongest forces acting against a brain changing its ‘key signature’ once the initial patterns are established. (My friend Adam comments: “This seems crucially important for establishing desirable value cores in human and human-like artificial systems.”)

How much variation in this is there across brains, how much is nature vs nurture, in what practical ways does this influence one’s emotional dynamics and emotional attractors, how could one characterize a typology and build a metric for this- these are all unknown and will be interesting to explore.

On aesthetics

I tend to believe a person’s aesthetic– what they find beautiful, and what they find ugly– is the closest thing to a carrier of personal identity we have. I am not my physical body; I am not my memories; I’m not even my brain. But as a lossy approximation, I would be okay with saying I am my aesthetic taste.

The promise of CSHW is we might be able to characterize this, perhaps as connectome-specific harmony dynamics (the ’emotional key signature’ I mention above). If we could in fact measure this, it would open up a world of applications: more inclusive personality metrics; connectome harmonic based methods for evaluating romantic compatibility; ‘backing up’ peoples emotional patterns and aesthetics in the case of stroke or degenerative disease; even modeling how various pieces of neurotechnology which could inject information into the brain (looking at you, Kernel and Neuralink) could be implemented without changing personality.

IV. Model interpersonal dynamics

How should we understand the neuroscience of human sociality and interaction? There’s an enormous amount of qualitative commentary on this, and some toy models involving things like mirror neurons, oxytocin, and such, but by and large this question has been very resistant to quantification. If, however, CSHW does describe the ‘deep contours’ of brain dynamics, brain harmonics will likely covary with social dynamics and could be a good foundation for building quantitative models.

The lowest-hanging fruit might be this: interpersonal compatibility seems to be about harmony in some deep respect. Maybe we could simply add two peoples’ connectome harmonics together and evaluate the result for consonance vs dissonance. This probably wouldn’t give a great result– a lot of interpersonal harmony seems to be about subtle dynamics, not resting-state activity. But it’d be worth a try.

I mentioned in my last post that:

I suspect the way our brains naturally model other brains is through modeling their connectome harmonics! That in studying CSHW we’re tapping into the same shortcuts for understanding other minds, the same compression schemas, that evolution has been using for hundreds of millions of years. This is a big claim, to be developed later.

I’m still bullish on this; when we model other peoples’ emotions, I think we’re simulating their connectome harmonics. (If our brains aren’t doing this, it would be like evolution leaving $100 bills on the sidewalk, which it rarely does.) Humans have white eyes so other humans can see where they’re looking; honest signals like this facilitate trust and social coordination. It seems plausible that this is true in the CSHW frame too, that the information peoples’ bodies naturally telegraph- their facial expressions, their vocal tone, their body language- could be fairly high-fidelity proxies for important aspects of their internal harmonics. This might also suggest that during conspecific rivalry– fights with other humans– emotions such as anger or jealousy would act to decorrelate our bodily ‘tells’ and our harmonics, in both large and subtle ways.

But more generally, I’d like to pose the question: what happens when two sets of connectome harmonics dance? When two people talk, or dance, or laugh, or debate, or fight, or make love, we can think of it as their connectome harmonics interacting — so how should we understand the typology of possible interactions, and the internal syntax of each interaction?

First, I suspect (with a nod to Section III above) we should consider humans as doing a complex mix of signal emission, signal entrainment, and signal filtering, and in a significant way, these signals ‘cash out’ in terms of their effects on connectome harmonics. Signals which affect high frequency harmonics mostly act as perceptual constraints on cognitive processing; signals which affect low frequency harmonics mostly act as drivers on emotional state.

Anecdotes: We can see the purpose of smalltalk here: it’s nominally the exchange of logical information, but the actual result is to drive two harmonic systems into entrainment with each other. At a recent party, I noticed two very distinct strategies people were employing to raise the mood of those around them: one friend, S, obviously was in a great mood and through loud talk and exaggerated gestures, was blasting out his mind music for the people around him to enjoy– an emotional metronome. Another friend, A, was listening very closely to the thoughts and feelings of the person he was talking to, and then reflecting back a more-harmonious version of that same pattern, cleaned of dissonance and noise– a dissonance filter. Adversarial interactions can also exist; driving someone into a state of internal dissonance is simply a matter of reversing these processes. Finally, the trick to socially navigating Depression seems to be to avoid being an emotional metronome of bad feeling to those around you.

Second, as to the technical syntax and mechanics of psychosocial interactions, I suspect we should look at them through the lens of signal interactions within periodic systems. This would imply understanding interactions between connectome harmonics in terms of different types of attempted causal coupling, which could plausibly lead to a reasonably bounded typology of interactions. (There is more to say here, but it might be an information hazard and I suspect it’s better discussed in person.)

We might expect that some future social psychologist could use this CSHW frame to map out and characterize interactions at a social gathering: e.g., “That gentleman is chatting up that lady, trying to pull her into entrainment with his harmonics, and it looks like he’s succeeding; meanwhile over there, there’s a couple that’s psyching themselves up to go talk to that famous person over there, probably trying to ‘presynchronize’ with his harmonics, and over there …”

At any rate, if CSHW can be used to build a good quantitative model of human-human interactions, it might also be possible to replicate these dynamics in human-computer interactions. This could take a weak form, such as building computer systems with a similar-enough interactional syntax to humans that some people could reach entrainment with it; affective computing done right. It might also take a much stronger form: if we can cleanly map the actual behavioral state space and dynamics of an AI system to human-like connectome harmonic dynamics, we would have something intuitively predictable, in the same way that humans find other humans intuitively predictable– and perhaps even trustworthy, in the same way. Who knows– perhaps this could even lead to alignable AI.

V. New theory of language & meaning

Philosophers and linguists tend to understand the world in terms of map and the territory; words and objects; language games and reality (although the distinction is not always clear). Philosophers like Hoffstadter & Dennett (and arguably Friston) broaden this somewhat with attempting to ‘naturalize’ meaning and intentionality through mechanistic processes, describing how the territory might generate the map.

What is often missed is that language is a felt-sense phenomenon. When I say something to you, I am operating at the semantic level, but I’m also operating at the visceral level. My words are essentially reaching out and ‘plucking’ certain metaphorical strings of feeling. In a very real sense, my intention is to convey that “if you combine this feeling, with that feeling, you get such and such feeling” and success results in you realizing “oh yeah, when I combine those two feelings I do get such and such feeling!”

Different languages (and individuals) will have subtly different ways to do this, and may produce subtly different results; but there’s a lot of pressure for convergence too, both across individuals and across linguistic clusters. When a German says haus and an American says house, I think it’s pretty safe to say a very similar felt-sense is being channeled in each. (At least, this is very true for low-level phenomena everyone experiences regularly in their daily lives. For higher abstractions, not always.)

At any rate, instead of this:

I think we should be talking about this:

The ontology of words is clear. You’re reading them now. Philosophers squabble over the ontology of objects, but generally physics and folk psychology seem to be doing a good job dealing with things. But the ontology of the felt-senses attached to given word+object pairs is a lot less clear.

In the long run, perhaps a formal theory of qualia like IIT could offer something like objective data here, by generating a mathematical qualia-object that represents the characteristic ‘shape’ which arises in people when they say or hear a word. (Caveat: we might get better data with phrases, instead of single words.) But in the meantime, it might be possible to say some things about this with the CSHW framework: we might think of words as pointers to, or operators on, specific bundles of connectome harmonics; language is thus a network of such pointers&operators. Under this framing, I think we’d be able to explain a lot of things about a few words, and a few interesting things about most words.

Granted, it would be really hard to get good data here, and practically speaking there are better uses of research time. I think the real value is looking at what this might imply about the metaphysics of communication. E.g., it would strongly imply language as both a driver of, and dependent upon, the standardization of connectome harmonics, and might suggest some interesting things about how languages naturally lead to different ’emotional timbres’. (See Section III for comments on how this might interact with the autism spectrum.)

The theme I find most interesting here is ontology as a system of relations between felt senses, and metaphysics as the relation between this network of felt-senses and the world. Shakespeare’s Hamlet famously admonished, ”There are more things in heaven and earth, Horatio, than are dreamt of in your philosophy.” If Horatio has fewer felt-senses than there are things in the world, then his ontology will behave as a too-small blanket, able to cover any one piece of reality but unable to cover it all at the same time, and he’ll make all sorts of type errors. Perhaps we could reinterpret Wittgenstein’s Tractatus as essentially saying, ‘we should aim to have as many words as there are felt-senses, and as many felt-senses as there are things in the world, and all forms of confusion come from mismappings between these domains.’ Or maybe not; Wittgenstein is hard to pin down.

At any rate, there’s plenty of threads we could pull here: we could try to understand the ‘characteristic connectome harmonic typology’ of different languages; we could use the word+object+felt-sense model to try to improve brain-computer interfaces; we could try to enrich communication by sonifying connectome harmonics, allowing an alternative intuitive appreciation for mind states.

Finally, I think we could get deep insight into the nature of the culture war: we might productively model politics, and all ideological conflict, as a vicious battle over the valence of these ‘felt senses’ and the logic of how they combine. Competing memeplexes have deep incompatibilities between their internal ‘felt-sense arithmetic’, which effectively constitute their boundaries and axis of conflict. Ontological holy war is about degrading the other side’s ability to think complex and/or adversarial thoughts, by desecrating and warping the meanings of their words, fragmenting the ‘key signature’ of their connectome harmonics. Brutal stuff.

Unknowns:

Since CSHW is a young paradigm (2016!), there’s a lot we don’t know about it. Some important question marks are:
- How invariant are these harmonics? How much do they vary within people depending on activity, mood, age, etc? And how much do they vary between people? How stable are they if we measure a brain two times in a row?
- There seems to be two ways the brain could change its harmonics: it could shift energy between static harmonics, or it could shift the actual frequencies of its harmonics. What’s the balance between these two strategies, and what are the rules of thumb for when it does the former vs the latter?
- Can we proxy Atasoy’s fMRI+DTI+MRI method with much cheaper (and more temporally accurate) EEG devices? What about if we get a ‘baseline’ reading once, with Atasoy’s full method, then use that to build an EEG proxy? Would we have to do this on a per-connectome basis, or could we fudge things with a good machine learning model? (The ‘wavenumber problem’ may be relevant here…)
- Atasoy’s method essentially outputs a time-averaged snapshot of someone’s connectome harmonics. How do we measure dynamics? Is a real-time derivation of someone’s connectome harmonics possible, given current theory?
- How can we improve the internal model parameters of CSHW? Can we borrow from the predictive coding / FEP paradigms to improve CSHW’s model of wave propagation in different regions of the brain?
The Big Picture: investing in humans

Bryan Johnson (Kernel / OSfund) has A Plan For Humanity. (Don’t be alarmed, it’s pretty benevolent.) There are a lot of subtle details, but the core theme is that it’s safer for humanity to invest in the ‘human codebase’ and human-improvement ecosystem, rather than the AI codebase and ecosystem. I think this is both reasonable and important, and I endorse what he argues. But to intelligently and presciently invest in the human codebase & human-improvement ecosystem, you need a Paradigm. Maybe many paradigms, but at least one capital-P Paradigm. I think CSHW could be that.

Thanks to Andrés Gomez Emilsson, Sarah McManus, Adam Safron, and Romeo Stevens for comments on previous drafts of this work.
Seed ontologies

Posted:Mon, 04 Jun 2018 05:01:11 +0000

Chatting with people at a recent conference on consciousness (TSC2018), I had the feeling of strolling through an alchemist’s convention: lots of optimistic energy & clever ideas, but also a strong sense that the field is pre-scientific. In short, there was a lot of overly-confident hand-waving.

But there were also a handful of promising ideas that stood out, that seemed like they could form at least part of the seed for an actual science of qualia; something that could transform the study of mind from alchemy to chemistry. Today I want to list these ideas, and say a few things about their ecosystem.

I. Key pieces of the formalization puzzle

Giulio Tononi’s Integrated Information Theory (IIT) is the leading (and perhaps the only) fully formal theory of consciousness. It’s essentially a mathematical method for measuring how much “integrated information” is embedded in a system, or in other words, how much each part of a system ‘knows about’ the other parts. IIT argues this is an identity relation, such that the amount of integrated information a system encodes is equal to the amount of consciousness it has.

IIT is enormously controversial, and to some extent a work-in-progress, but the theory does three very clever things that I think are under appreciated: first, it uses this idea of integrated information to naturally determine the boundary of conscious systems (and thus the boundary of phenomenology): if including an element increases the total integrated information (Phi), then it’s inside the boundary; if it doesn’t, it’s not. I.e., IIT solves many different kinds of problems with a single underlying mechanic. Second, it introduces the idea of Qualia Formalism, that the correct goal for a theory of consciousness is to calculate a mathematical representation of a system’s phenomenology — a central insight that seems obvious once formally expressed, but was not obvious before IIT expressed it. Third, it actually makes falsifiable predictions that have been tested, and passed – something no other theory of consciousness has done.

Karl Friston’s Free Energy Principle (FEP) is the leading formal theory of self-organizing system dynamics, one which has (in various guises) pretty much taken neuroscience by storm. It argues that any self-organizing system which effectively resists disorder must (as its core organizing principle) minimize its free energy, that free energy is equivalent to surprise (in a Bayesian sense), and that this surprise-minimization drives basically all human behavior.

The FEP is notoriously difficult to understand*, but it’s also provided a unifying frame for systems neuroscience, where many of the things we’ve laboriously discovered about how the brain works- and common ways it malfunctions- just naturally ‘pop out of’ Friston’s equations.**

Credit to Adam Safron for the graphic; captions added by me.

*With a nod to Chalmers, I’d pose the meta-problem of free energy: why does trying to understand Friston’s argument that all systems try to minimize their free-energy, so reliably increase our free energy?

**There are some similarities between Tononi’s IIT and Friston’s FEP, which might be due to both being grad students together under Gerald Edelman, who got his Nobel for studying the adaptive immune system and then applied the same principles of self-organization to study the brain. In a substantial sense, IIT and FEP are different extrapolations of Edelman’s “neural darwinism” frame.

Selen Atasoy’s Connectome-Specific Harmonic Waves (CSHW) is a new method for interpreting neuroimaging which (unlike conventional approaches) may plausibly measure things directly relevant to phenomenology. Essentially, it’s a method for combining fMRI/DTI/MRI to calculate a brain’s intrinsic ‘eigenvalues’, or the neural frequencies which naturally resonate in a given brain, as well as the way the brain is currently distributing energy (periodic neural activity) between these eigenvalues. CSHW is a fairly young paradigm, but already a lot of big names (Kringelbach, Carhart-Harris) are onboard.

Why do I think CSHW will be so powerful for talking about phenomenology? First, it seems a priori plausible that systems like brain with significant periodicity will self-organize around their eigenvalues; i.e. these eigenvalues will be functionally significant and ‘costless’ Schelling points. This implies that these harmonics will be a good place to start if we want to efficiently compress a lot of the brain’s (and mind’s) complexity. Second, Atasoy and QRI’s own Emilsson have already offered compelling stories about how differences in the harmonic energy distribution might influence phenomenology; these models are ever-improving, and given someone’s connectome harmonics at two different points in time, I think we can make some reasonably good educated guesses on how the texture of their subjective experience has changed. No other neuroimaging paradigm can claim this at any significant degree of granularity. Third, I suspect the way our brains naturally model other brains is through modeling their connectome harmonics! That in studying CSHW we’re tapping into the same shortcuts for understanding other minds, the same compression schemas, that evolution has been using for hundreds of millions of years. This is a big claim, to be developed later.

A Chladni plate, showing the existence of eigenvalues (standing waves) at various frequencies. Atasoy’s work applies these same equations to the brain.

Atasoy’s work essentially decomposes brain activity (waves of excitation and inhibition) into its constituent harmonics, such as these. Note that each pattern perfectly repeats over time; they ‘wrap around’ the brain an integer number of times. Image credit: Andrés Gomez Emilsson & Selen Atasoy, et al.

QRI’s Symmetry Theory of Valence (STV) is plausibly the first line in the Rosetta Stone of consciousness: namely, if we have a mathematical representation of a subjective experience (such as the output of IIT), the symmetry of this representation will correspond to the emotional valence of the experience. To put it succinctly: harmony in the brain = pleasure, and this is an identity relation which is universally true, not a ‘mere correlation’.

STV is still a young theory, but it’s generating philosophical clarification (how ‘pleasure centers’ function; how various drugs change our mood; why music is so affectively compelling; why we seek out pleasure) as well as falsifiable predictions (the world’s first method for measuring emotional valence using fMRI built entirely from first principles; tinnitus likely causes hidden affective blunting) with more in the pipeline. I’m admittedly biased, but I’m excited for the road ahead, and view this as the pilot project for reverse-engineering all sorts of qualia.

For the core theory, see Principia Qualia; image credit Andrés Gomez Emilsson, pulled from his presentation on Quantifying Bliss.

II. Ecosystem thinking

I think all of the above theories are talking about things that are real and important. To slightly mix metaphors, if everybody doing consciousness research is blind, then at least these paradigms have found the elephant and are each describing a different part of it.

But the ultimate goal here is unification. Combining these different ways of ontologizing and understanding the mind, into new forms of knowledge that have all their collective strengths and none of their individual weaknesses; perceiving the whole elephant. This depends on understanding the relational ecosystem of such theories: what special insight each theory brings to the table, and what tools it might benefit from borrowing from the others.

Here’s my (circa June 2018) take on this:

IIT‘s core insights are that causal microstructure matters for phenomenology and defines a topology, phenomenal systems have boundaries, and knowledge about qualia requires a formalism about qualia. But aside from a bit of exploratory work on animats, it says little about the evolution of integrated information in systems, nothing about embodied minds, little about how large-scale brain dynamics might map to qualia dynamics, and little about what the ‘natural kinds’ of qualia are.

What could IIT borrow from other pieces?

FEP: Principles (multilevel constraints) by which all evolved systems self-organize (i.e. the system-level constraints by which high Phi is selected); also symbol grounding, via the inherent cybernetic/embodied nature of FEP (thanks to Adam Safron for this point).

CSHW: A semi-clean high-level of abstraction which may elegantly proxy certain aspects (emotion & integration?) of phenomenology. (And a much better paradigm than EEG for applying IIT?)

STV: A qualia natural kind, the other half of ethics, another way to test the formalism.

FEP‘s core insights are that surprise minimization is the absolute core directive of all successful self-organizing systems, action & intention & everything in between can be framed as surprise minimization, and that successful systems spend most of their time in a small subset of their possible set of states. However, despite these enormous insights, FEP fails as a theory of phenomenology as it says nothing about phenomenology. It’s also difficult to literally measure free energy, at least in a useful way. Finally, the way the FEP models emotions is clever, but admits of many counterexamples and edge-cases.

What could FEP borrow from other pieces?

IIT: A principled boundary for, and thus a bridge to, phenomenology.

CSHW: A semi-clean high-level of abstraction which tracks & constrains (provides Schelling points for) FE flows.

STV: A phenomenologically-supported expectation of how the brain implements FE gradients algorithmically (symmetry as computational success target); a better model for valence than FE-over-time.

CSHW‘s core insights are that the brain is a system which exhibits significant & semi-discrete resonances, has probably self-organized around its natural resonances, and that changes in the brain’s resonant profile likely track changes in phenomenology. However, CSHW says very little about the functional or phenomenological significance of specific resonances & dynamical interactions between resonances, all of which would be really interesting to know.

What could CSHW borrow from other pieces?

IIT: A place to put knowledge of phenomenology. Much more, if IIT figures out how to calculate Phi from brain harmonics.

FEP: Functional significance of certain CH dynamics (dynamo (free-energy-collecting) harmonics? constraints on what eigenvalues-in-a-FE-minimizing-system-mean?).

STV: A success condition for computation; a clean way to derive valence from CHs.

STV‘s core insights are that symmetry is the first place to look for qualia natural kinds, and most other theories are leaving value on the table by not directly addressing symmetry or valence (and each one of these will lead to the other). But STV alone doesn’t have as grand of a scope as the other theories mentioned and says little about their bailiwicks.

What could STV borrow from other pieces?

IIT: A place to put knowledge about qualia (like symmetry/valence).

FEP: A framework for understanding multi-level self-organization towards (and sometimes away from) symmetry.

CSHW: An empirical framework for proxying symmetry-in-phenomenology.

STV is a smaller theory, so it tends to be easy to combine with the others (especially IIT & CSHW). I’m not sure what would be necessary to unify FEP and IIT, FEP and CSHW, or IIT and CSHW. But I’d bet on an eventual Grand Unification of all four eventually, and I suspect this unification will usher in a new era of studying the mind (and thus, perhaps a new era of humanity).

III. Musings about the pursuit of the One True Ontology

One under appreciated thing about each of the architects above is how deeply they live and breathe their ontologies. Much as a primatologist who’s studied dominance hierarchies for 30 years sees everything through that lens, or how a preacher may see the hand of God in everything, the theorists above see reality in terms of their core ontologies. E.g., I have no doubt that Tononi sees everything through the lens of cause-effect clusters, each with their own Phi; Friston sees everything as a Markov blanket (including, I suspect, literally seeing people as walking Markov blankets). And so on. I call this being a ‘One True Ontologist’.

And I think it’s underappreciated how good, necessary, and generative this is. First, any ontology which can satisfyingly explain the brain and the mind might be powerful enough to be the last ontology we need; it might just be Truth-with-a-capital-T. It pays to aim high and look for this. Second, original work at the cutting-edge of these fields is very difficult, and people have to be intensely motivated and emotionally inspired to try. And the strongest creative motivation there is, is the feeling of explaining everything in terms of your Big Idea. Likewise, it takes a Big Idea to break out of current paradigms & equilibriums. So there are selection effects on the frontiers for Big All-Encompassing Ideas. Third, having a Big Idea about the mind that nobody else has is a substantial social advantage.

More on this latter point: we routinely and instinctively employ methods to obfuscate our inner states and intentions; to do so is to be a strategic human in an uncertain world. On the other hand, having a novel (and accurate) method to model minds, especially one that carves its ontology differently from how most people do, is very effective at circumventing most of these social obfuscation methods people instinctively toss up. You get a piercing understanding of people. I don’t want to give away any secrets, but I think all of the four theories above are worth studying even from just this perspective.

At any rate, I think chasing the One True Ontology is a worthwhile goal.

Of course, to truly win the ontology game, one has to eat (prove your ontology can maintain consistency with, and ideally derive & improve on the predictions of) the current king, which is physics. More on this later.
Why are humans good?

Posted:Mon, 19 Mar 2018 19:32:36 +0000
Are humans worthy of colonizing the universe? Are we particularly awesome and benevolent, moreso than a random mind sampled from mindspace?

The following isn’t a full argument, but I want to point toward two things humans seem to do:
- First, our brains are set up in such a way that we stochastically seek out pleasant states of mind. (I think this is a contingent fact about humans, not a universal for intelligent beings);
- Second, we model other beings through empathy (i.e. we’re ‘qualia resonators’ and take on aspects of the phenomenology of those around us, insofar as we can perceive it).
These two things combine in a very interesting way: left to our individual devices, we tend to adapt our environment such that it embodies many beings with positive valence, and their positive valence becomes our positive valence. The domestication and adaptation of wolves is a good example: dogs are basically four-legged happiness machines that we keep around because their fantastic qualia rubs off on us.

Now of course there are a million caveats: we’re often bad at these things, empathy-based behavior has a ton of failure-modes, this doesn’t address hedonic set-points or game-theoretic traps, etc. But the way these two things interact is a big reason I like humanity, and want to preserve it.
Rescuing Philosophy

Posted:Tue, 03 Oct 2017 05:04:39 +0000

I.

Philosophy has lost much of its energy, focus, and glamor in our modern era. What happened?

I’d suggest that five things went wrong:

1. Historical illegibility. Historically, ‘philosophy’ is what you do when you don’t know what to do. This naturally involves a lot of error. Once you figure out a core ontology and methodology for your topic, you stop calling it ‘philosophy’ and start calling it ‘science’, ‘linguistics’, ‘modal logic’, and so on. This is a very important, generative process, but it also means that if you look back at the history of philosophy, you basically only see ideas that are, technically speaking, wrong. This gives philosophers trying to ‘carry on the tradition’ a skewed understanding of what philosophy is, and how to do it.

2. Evaporative cooling. The fact that the most successful people, ideas, ontologies/methodologies, and tools tend to leave philosophy to found their own disciplines leads to long-term problems with quality. We can think of this as an evaporative cooling effect where philosophy is left with the fakest of problems, and the worst, most incoherent and confused ways to frame what real problems are left.

3. Lack of feedback loops. Good abstract philosophy is really hard to do right, it’s hard to distinguish good philosophy from the bad, and the value of doing philosophy well isn’t as immediately apparent as, say, chemistry. This leads to ‘monkey politics’ playing a large role in which ideas gain traction, which in turn drives a lot of top talent away.

4. Professionalization. Turning metaphysical confusion into something clear enough to build a new science on tends to be a very illegible process, full of false-starts, recontextualizations, and unpredictable breakthroughs. This is really hard to systematically teach students how to do, and even harder to plan an academic budget around. As philosophy became regularized and professionalized– something that you can have a career in– it was also pushed toward top-down legibility. This resulted in less of a focus on grappling with metaphysical uncertainty and more focus on institutionally-legible things such as scholarship, incremental research, teaching, and so on. Today, the discipline is often taught and organized academically as a ‘history of ideas’, based on how past philosophers carved various problem-spaces.

5. Postmodernism. Philosophy got hit pretty hard by postmodernism — and insofar as philosophy was the traditional keeper of theories of meaning, and insofar as postmodernism attacked all sources of meaning, philosophy suffered more than other disciplines. Likewise, academic philosophy has inherited all the problems of modern academia, of which there are many.

I’m painting with a broad brush here, and I should note that there are pockets of brilliant academic philosophers out there doing good, and even heroic, work in spite of these structural conditions. #notallphilosophers. But I don’t think many of these would claim they’re happy with modern academic philosophy’s structural conditions or trajectory.

And this matters, since philosophy is still necessary. There’s a *lot* of value in having a solid philosophical toolset, and having a healthy intellectual tradition of being mindful about ontological questions, epistemology, and so on. As David Pearce often points out, there’s no clean way to abstain from thorny philosophical issues: “The penalty of not doing philosophy isn’t to transcend it, but simply to give bad philosophical arguments a free pass.”

II.

So philosophy is broken. What do we do about it?

My friend Sebastian Marshall describes the ‘evaporative cooling’ philosophy has undergone, and suggests that we should try to rescue and reclaim philosophy:

So, this bastardized divorced left-behind philosophy will be here to stay in some form or fashion. We can’t get rid of it… but it’s also not necessary to get rid of it.

Turning to better news, even in mainstream philosophy, there are still sane and sound branches doing good work, like logic (which is basically math) and philosophy of mind (which is rapidly becoming neuroscience but which hasn’t yet evaporatively cooled out of philosophy).

It wouldn’t take very many people reclaiming the word philosophy as a love of wisdom to begin to turn things around.

Genuinely good philosophy is happening all over the place – though it’s rarely people in fields that don’t fight back at all. Indeed, you see computer programming and financiers doing some of the best philosophy now – Paul Graham, Eliezer Yudkowsky, Ray Dalio, Charlie Munger, Nassim Taleb. When the computer scientist get something wrong, their code doesn’t work. When the financier gets something wrong, they lose a lot of money. Excellent philosophers still come of the military – John Boyd and Hyman Rickover to name two recent Americans – and they come out of industrial engineering, like Eli Goldratt.

That these people are currently not classified as philosophers is simply an error– let the people doing uselessness in the towers call themselves “theoretical screwaroundists” or whatever other more palatable name they might come up with for themselves; genuine philosophy is alive and well, even as the word points to decayed and failing institutions.

There would clearly be enormous benefits to reclaiming the word “philosophy” for serious generative work. But I worry it’s going to be really hard.

Words have a lifecycle- often, they start out full of focus, wit, and life, able to vividly convey some key relationship. As time goes on, however, they lose this special something as they become normalized and regress toward the linguistic mean. Part of being a good writer is being in-tune with what words & phrases still have life, and part of being a great writer (like Shakespeare) is minting new ones. My sense is that “philosophy” doesn’t have much sparkle left, and it may be preferable to coin a new word.

Unfortunately, I don’t know of any better words that would encapsulate everything we’d want, and it may be very difficult to rally people behind a new term unless it’s really good. Even though academic philosophy is in terrible shape, the term ‘philosophy’ is still an effective Schelling point; still prime memetic real-estate. So, pending that better option, I think Sebastian’s right and we do need to do what we can to rescue & reclaim philosophy.

III.

How do we rescue philosophy? — I think we need to think about this both in terms of individual tactics and collective strategy.

Individual tactics: survival and value creation in an unfriendly environment

Essentially, those that wish to make a notable, real, and durable contribution to philosophy should understand that association with academia is a double-edged sword. On the plus side, it can give people credibility, access, and fellowship with other academics, apprenticeships with established thinkers, maybe a steady income, and a great excuse to engage deeply with philosophy. On the other hand, by going into academic philosophy someone is essentially granting an unhealthy, partially moribund system broad influence over their local incentives, memetic milieu, and aesthetic. That’s a really big deal.

A personal aside: I struggled with how to navigate this while writing Principia Qualia. Clearly a new philosophical work on consciousness should engage with other work in the space– and there’s a lot of good philosophy of mind out there, work I could probably use and build upon. At the same time, if philosophy’s established ways of framing the problem of consciousness could lead to a solution, it would’ve been solved by now, and by using someone else’s packaged ontology, I’d be at risk of importing their confusion into my foundation. With this in mind I decided that being aware of key landmarks in philosophy was important, but being uncorrelated with philosophy’s past framing was equally important, so I took a minimalist first-principles approach to building my framework and was very careful about what I imported from philosophy and how I used it.

Collective strategy: Schelling points & positive-feedback loops

The machinery of modern academic philosophy is going to resist attempts at reformation, as all rudderless bureaucratic entities do, but it won’t be proactively hostile about it, and in fact a lot of philosophers desperately want change. This means people can engage in open coordination on this problem. I.e., if we can identify Schelling points and plant rallying flags which can help coordinate with potential allies, we could probably make a collective push to fix certain problems or subfields (my sources say this sort of ‘benign takeover’ is already in motion in certain departments of bioethics).

Ultimately, though, fixing philosophy from within probably looks like a better option than it actually is, since (1) entryism is sneaky, always has a bad faith component, and is never as simple as it sounds (if nothing else, you have to fight off other entryists!), and (2) meme flow always goes both ways, and a plan to fix philosophy’s norms faster than its bad norms subvert us is inherently risky. Plenty of good people with magnificent intentions of fixing philosophy go into grad school, only to get lost in the noise, fail to catalyze a positive-feedback-loop, burn out, and give up years later. If you’re going into academic philosophy anyway, then definitely try to improve it, but don’t go into academic philosophy in order to improve it.

Instead, it may be better to build institutions that are separate from modern academic philosophy, and compete against it. Right now, academic philosophy looks “too big to fail”- a juggernaut that, for all its flaws, is still the go-to arbiter of success, authority, and truth in philosophy. And as long as academic philosophy can keep its people stably supplied with money and status, and people on the outside have to scramble for scraps, this isn’t going to change much. But nothing is forever & there are hints of a shift, the world needs better alternatives, and now is a great time to start building them.

In short, I think the best way to fix philosophy may be to to build new (or revive ancient) competing metaphors for what philosophy should be, to solve problems that modern philosophy can’t, to offer a viable refuge for people fleeing academia’s dysfunction, and to make academia come to us if it wants to stay relevant.

IV.

This is essentially what we’re working toward at the Qualia Research Institute: building something new, outside of academic philosophy in order to avoid its dysfunction, but still very much focused on a core problem of philosophy.

I see this happening elsewhere, too: LessWrong is essentially a “hard fork” of epistemology, with different problem-carvings, norms, and methods, which are collectively slowly maturing into the notion of executable philosophy. Likewise, Leverage Research may be crazy, but I’ve got to give them credit for being crazy in a novel and generative way, one which is uncorrelated with the more mundane, depressing ways modern academic philosophy & psychology are crazy. Honorable mentions include Exosphere, an intentional community I’m pretty sure Aristotle would have felt right at home in, and Alexandros Pagidas, a refugee from academic philosophy who’s trying to revive traditional Greek-style philosophical fight clubs (which, to be honest, sound kind of fun).

There are a lot of these little seeds around. Not all of them will sprout into something magnificent. But I think most are worth watering.
Why I think the Foundational Research Institute should rethink its approach

Posted:Thu, 20 Jul 2017 21:10:00 +0000
The following is my considered evaluation of the Foundational Research Institute, circa July 2017. I discuss its goal, where I foresee things going wrong with how it defines suffering, and what it could do to avoid these problems.

TL;DR version: functionalism (“consciousness is the sum-total of the functional properties of our brains”) sounds a lot better than it actually turns out to be in practice. In particular, functionalism makes it impossible to define ethics & suffering in a way that can mediate disagreements.

I. What is the Foundational Research Institute?

The Foundational Research Institute (FRI) is a Berlin-based group that “conducts research on how to best reduce the suffering of sentient beings in the near and far future.” Executive Director Max Daniel introduced them at EA Global Boston as “the only EA organization which at an organizational level has the mission of focusing on reducing s-risk.” S-risks are, according to Daniel, “risks where an adverse outcome would bring about suffering on an astronomical scale, vastly exceeding all suffering that has existed on Earth so far.”

Essentially, FRI wants to become the research arm of suffering-focused ethics, and help prevent artificial general intelligence (AGI) failure-modes which might produce suffering on a cosmic scale.

What I like about FRI:

While I have serious qualms about FRI’s research framework, I think the people behind FRI deserve a lot of credit- they seem to be serious people, working hard to build something good. In particular, I want to give them a shoutout for three things:
- First, FRI takes suffering seriously, and I think that’s important. When times are good, we tend to forget how tongue-chewingly horrific suffering can be. S-risks seem particularly horrifying.
- Second, FRI isn’t afraid of being weird. FRI has been working on s-risk research for a few years now, and if people are starting to come around to the idea that s-risks are worth thinking about, much of the credit goes to FRI.
- Third, I have great personal respect for Brian Tomasik, one of FRI’s co-founders. I’ve found him highly thoughtful, generous in debates, and unfailingly principled. In particular, he’s always willing to bite the bullet and work ideas out to their logical end, even if it involves repugnant conclusions.
What is FRI’s research framework?

FRI believes in analytic functionalism, or what David Chalmers calls “Type-A materialism”. Essentially, what this means is there’s no ’theoretical essence’ to consciousness; rather, consciousness is the sum-total of the functional properties of our brains. Since ‘functional properties’ are rather vague, this means consciousness itself is rather vague, in the same way words like “life,” “justice,” and “virtue” are messy and vague.

Brian suggests that this vagueness means there’s an inherently subjective, perhaps arbitrary element to how we define consciousness:

Analytic functionalism looks for functional processes in the brain that roughly capture what we mean by words like “awareness”, “happy”, etc., in a similar way as a biologist may look for precise properties of replicators that roughly capture what we mean by “life”. Just as there can be room for fuzziness about where exactly to draw the boundaries around “life”, different analytic functionalists may have different opinions about where to define the boundaries of “consciousness” and other mental states. This is why consciousness is “up to us to define”. There’s no hard problem of consciousness for the same reason there’s no hard problem of life: consciousness is just a high-level word that we use to refer to lots of detailed processes, and it doesn’t mean anything in addition to those processes.

Finally, Brian argues that the phenomenology of consciousness is identical with the phenomenology of computation:

I know that I’m conscious. I also know, from neuroscience combined with Occam’s razor, that my consciousness consists only of material operations in my brain — probably mostly patterns of neuronal firing that help process inputs, compute intermediate ideas, and produce behavioral outputs. Thus, I can see that consciousness is just the first-person view of certain kinds of computations — as Eliezer Yudkowsky puts it, “How An Algorithm Feels From Inside“. Consciousness is not something separate from or epiphenomenal to these computations. It is these computations, just from their own perspective of trying to think about themselves.

In other words, consciousness is what minds compute. Consciousness is the collection of input operations, intermediate processing, and output behaviors that an entity performs.

And if consciousness is all these things, so too is suffering. Which means suffering is computational, yet also inherently fuzzy, and at least a bit arbitrary; a leaky high-level reification impossible to speak about accurately, since there’s no formal, objective “ground truth”.

II. Why do I worry about FRI’s research framework?

In short, I think FRI has a worthy goal and good people, but its metaphysics actively prevent making progress toward that goal. The following describes why I think that, drawing heavily on Brian’s writings (of FRI’s researchers, Brian seems the most focused on metaphysics):

Note: FRI is not the only EA organization which holds functionalist views on consciousness; much of the following critique would also apply to e.g. MIRI, FHI, and OpenPhil. I focus on FRI because (1) Brian’s writings on consciousness & functionalism have been hugely influential in the community, and are clear enough *to* criticize; (2) the fact that FRI is particularly clear about what it cares about- suffering- allows a particularly clear critique about what problems it will run into with functionalism; (3) I believe FRI is at the forefront of an important cause area which has not crystallized yet, and I think it’s critically important to get these objections bouncing around this subcommunity.

Objection 1: Motte-and-bailey

Brian: “Consciousness is not a thing which exists ‘out there’ or even a separate property of matter; it’s a definitional category into which we classify minds. ‘Is this digital mind really conscious?’ is analogous to ‘Is a rock that people use to eat on really a table?’ [However,] That consciousness is a cluster in thingspace rather than a concrete property of the world does not make reducing suffering less important.”

The FRI model seems to imply that suffering is ineffable enough such that we can’t have an objective definition, yet sufficiently effable that we can coherently talk and care about it. This attempt to have it both ways seems contradictory, or at least in deep tension.

Indeed, I’d argue that the degree to which you can care about something is proportional to the degree to which you can define it objectively. E.g., If I say that “gnireffus” is literally the most terrible thing in the cosmos, that we should spread gnireffus-focused ethics, and that minimizing g-risks (far-future scenarios which involve large amounts of gnireffus) is a moral imperative, but also that what is and what isn’t gnireffus is rather subjective with no privileged definition, and it’s impossible to objectively tell if a physical system exhibits gnireffus, you might raise any number of objections. This is not an exact metaphor for FRI’s position, but I worry that FRI’s work leans on the intuition that suffering is real and we can speak coherently about it, to a degree greater than its metaphysics formally allow.

Max Daniel (personal communication) suggests that we’re comfortable with a degree of ineffability in other contexts; “Brian claims that the concept of suffering shares the allegedly problematic properties with the concept of a table. But it seems a stretch to say that the alleged tension is problematic when talking about tables. So why would it be problematic when talking about suffering?” However, if we take the anti-realist view that suffering is ‘merely’ a node in the network of language, we have to live with the consequences of this: that ‘suffering’ will lose meaning as we take it away from the network in which it’s embedded (Wittgenstein). But FRI wants to do exactly this, to speak about suffering in the context of AGIs, simulated brains, even video game characters.

We can be anti-realists about suffering (suffering-is-a-node-in-the-network-of-language), or we can argue that we can talk coherently about suffering in novel contexts (AGIs, mind crime, aliens, and so on), but it seems inherently troublesome to claim we can do both at the same time.

Objection 2: Intuition duels

Two people can agree on FRI’s position that there is no objective fact of the matter about what suffering is (no privileged definition), but this also means they have no way of coming to any consensus on the object-level question of whether something can suffer. This isn’t just an academic point: Brian has written extensively about how he believes non-human animals can and do suffer extensively, whereas Yudkowsky (who holds computationalist views, like Brian) has written about how he’s confident that animals are not conscious and cannot suffer, due to their lack of higher-order reasoning.

And if functionalism is having trouble adjudicating the easy cases of suffering–whether monkeys can suffer, or whether dogs can— it doesn’t have a sliver of a chance at dealing with the upcoming hard cases of suffering: whether a given AGI is suffering, or engaging in mind crime; whether a whole-brain emulation (WBE) or synthetic organism or emergent intelligence that doesn’t have the capacity to tell us how it feels (or that we don’t have the capacity to understand) is suffering; if any aliens that we meet in the future can suffer; whether changing the internal architecture of our qualia reports means we’re also changing our qualia; and so on.

In short, FRI’s theory of consciousness isn’t actually a theory of consciousness at all, since it doesn’t do the thing we need a theory of consciousness to do: adjudicate disagreements in a principled way. Instead, it gives up any claim on the sorts of objective facts which could in principle adjudicate disagreements.

This is a source of friction in EA today, but it’s mitigated by the sense that

(1) The EA pie is growing, so it’s better to ignore disagreements than pick fights;

(2) Disagreements over the definition of suffering don’t really matter yet, since we haven’t gotten into the business of making morally-relevant synthetic beings (that we know of) that might be unable to vocalize their suffering.

If the perception of one or both of these conditions change, the lack of some disagreement-adjudicating theory of suffering will matter quite a lot.

Objection 3: Convergence requires common truth

Mike: “[W]hat makes one definition of consciousness better than another? How should we evaluate them?”

Brian: “Consilience among our feelings of empathy, principles of non-discrimination, understandings of cognitive science, etc. It’s similar to the question of what makes one definition of justice or virtue better than another.”

Brian is hoping that affective neuroscience will slowly converge to accurate views on suffering as more and better data about sentience and pain accumulates. But convergence to truth implies something (objective) driving the convergence- in this way, Brian’s framework still seems to require an objective truth of the matter, even though he disclaims most of the benefits of assuming this.

Objection 4: Assuming that consciousness is a reification produces more confusion, not less

Brian: “Consciousness is not a reified thing; it’s not a physical property of the universe that just exists intrinsically. Rather, instances of consciousness are algorithms that are implemented in specific steps. … Consciousness involves specific things that brains do.”

Brian argues that we treat conscious/phenomenology as more ‘real’ than it is. Traditionally, whenever we’ve discovered something is a leaky reification and shouldn’t be treated as ‘too real’, we’ve been able to break it down into more coherent constituent pieces we can treat as real. Life, for instance, wasn’t due to élan vital but a bundle of self-organizing properties & dynamics which generally co-occur. But carrying out this “de-reification” process on consciousness– enumerating its coherent constituent pieces– has proven difficult, especially if we want to preserve some way to speak cogently about suffering.

Speaking for myself, the more I stared into the depths of functionalism, the less certain everything about moral value became– and arguably, I see the same trajectory in Brian’s work and Luke Muehlhauser’s report. Their model uncertainty has seemingly become larger as they’ve looked into techniques for how to “de-reify” consciousness while preserving some flavor of moral value, not smaller. Brian and Luke seem to interpret this as evidence that moral value is intractably complicated, but this is also consistent with consciousness not being a reification, and instead being a real thing. Trying to “de-reify” something that’s not a reification will produce deep confusion, just as surely trying to treat a reification as ‘more real’ than it actually is will.

Edsger W. Dijkstra famously noted that “The purpose of abstraction is not to be vague, but to create a new semantic level in which one can be absolutely precise.” And so if our ways of talking about moral value fail to ‘carve reality at the joints’- then by all means let’s build better ones, rather than giving up on precision.

Objection 5: The Hard Problem of Consciousness is a red herring

Brian spends a lot of time discussing Chalmers’ “Hard Problem of Consciousness”, i.e. the question of why we’re subjectively conscious, and seems to base at least part of his conclusion on not finding this question compelling— he suggests “There’s no hard problem of consciousness for the same reason there’s no hard problem of life: consciousness is just a high-level word that we use to refer to lots of detailed processes, and it doesn’t mean anything in addition to those processes.” I.e., no ‘why’ is necessary; when we take consciousness and subtract out the details of the brain, we’re left with an empty set.

But I think the “Hard Problem” isn’t helpful as a contrastive centerpiece, since it’s unclear what the problem is, and whether it’s analytic or empirical, a statement about cognition or about physics. At the Qualia Research Institute (QRI), we don’t talk much about the Hard Problem; instead, we talk about Qualia Formalism, or the idea that any phenomenological state can be crisply and precisely represented by some mathematical object. I suspect this would be a better foil for Brian’s work than the Hard Problem.

Objection 6: Mapping to reality

Brian argues that consciousness should be defined at the functional/computational level: given a Turing machine, or neural network, the right ‘code’ will produce consciousness. But the problem is that this doesn’t lead to a theory which can ‘compile’ to physics. Consider the following:

Imagine you have a bag of popcorn. Now shake it. There will exist a certain ad-hoc interpretation of bag-of-popcorn-as-computational-system where you just simulated someone getting tortured, and other interpretations that don’t imply that. Did you torture anyone? If you’re a computationalist, no clear answer exists- you both did, and did not, torture someone. This sounds like a ridiculous edge-case that would never come up in real life, but in reality it comes up all the time, since there is no principled way to *objectively derive* what computation(s) any physical system is performing.

I don’t think this is an outlandish view of functionalism; Brian suggests much the same in How to Interpret a Physical System as a Mind: “Physicalist views that directly map from physics to moral value are relatively simple to understand. Functionalism is more complex, because it maps from physics to computations to moral value. Moreover, while physics is real and objective, computations are fictional and ‘observer-relative’ (to use John Searle’s terminology). There’s no objective meaning to ‘the computation that this physical system is implementing’ (unless you’re referring to the specific equations of physics that the system is playing out).”

Gordon McCabe (McCabe 2004) provides a more formal argument to this effect— that precisely mapping between physical processes and (Turing-level) computational processes is inherently impossible— in the context of simulations. First, McCabe notes that:

[T]here is a one-[to-]many correspondence between the logical states [of a computer] and the exact electronic states of computer memory. Although there are bijective mappings between numbers and the logical states of computer memory, there are no bijective mappings between numbers and the exact electronic states of memory.

This lack of an exact bijective mapping means that subjective interpretation necessarily creeps in, and so a computational simulation of a physical system can’t be ‘about’ that system in any rigorous way:

In a computer simulation, the values of the physical quantities possessed by the simulated system are represented by the combined states of multiple bits in computer memory. However, the combined states of multiple bits in computer memory only represent numbers because they are deemed to do so under a numeric interpretation. There are many different interpretations of the combined states of multiple bits in computer memory. If the numbers represented by a digital computer are interpretation-dependent, they cannot be objective physical properties. Hence, there can be no objective relationship between the changing pattern of multiple bit-states in computer memory, and the changing pattern of quantity-values of a simulated physical system.

McCabe concludes that, metaphysically speaking,

A digital computer simulation of a physical system cannot exist as, (does not possess the properties and relationships of), anything else other than a physical process occurring upon the components of a computer. In the contemporary case of an electronic digital computer, a simulation cannot exist as anything else other than an electronic physical process occurring upon the components and circuitry of a computer.

Where does this leave ethics? In Flavors of Computation Are Flavors of Consciousness, Brian notes that “In some sense all I’ve proposed here is to think of different flavors of computation as being various flavors of consciousness. But this still leaves the question: Which flavors of computation matter most? Clearly whatever computations happen when a person is in pain are vastly more important than what’s happening in a brain on a lazy afternoon. How can we capture that difference?”

But if Brian grants the former point- that “There’s no objective meaning to ‘the computation that this physical system is implementing’”– then this latter task of figuring out “which flavors of computation matter most” is provably impossible. There will always be multiple computational (and thus ethical) interpretations of a physical system, with no way to figure out what’s “really” happening. No way to figure out if something is suffering or not. No consilience; not now, not ever.

Note: despite apparently granting the point above, Brian also remarks that:

I should add a note on terminology: All computations occur within physics, so any computation is a physical process. Conversely, any physical process proceeds from input conditions to output conditions in a regular manner and so is a computation. Hence, the set of computations equals the set of physical processes, and where I say “computations” in this piece, one could just as well substitute “physical processes” instead.

This seems to be (1) incorrect, for the reasons I give above, or (2) taking substantial poetic license with these terms, or (3) referring to hypercomputation (which might be able to salvage the metaphor, but would invalidate many of FRI’s conclusions dealing with the computability of suffering on conventional hardware).

This objection may seem esoteric or pedantic, but I think it’s important, and that it ripples through FRI’s theoretical framework with disastrous effects.

Objection 7: FRI doesn’t fully bite the bullet on computationalism

Brian suggests that “flavors of computation are flavors of consciousness” and that some computations ‘code’ for suffering. But if we do in fact bite the bullet on this metaphor and place suffering within the realm of computational theory, we need to think in “near mode” and accept all the paradoxes that brings. Scott Aaronson, a noted expert on quantum computing, raises the following objections to functionalism:

I’m guessing that many people in this room side with Dennett, and (not coincidentally, I’d say) also with Everett. I certainly have sympathies in that direction too. In fact, I spent seven or eight years of my life as a Dennett/Everett hardcore believer. But, while I don’t want to talk anyone out of the Dennett/Everett view, I’d like to take you on a tour of what I see as some of the extremely interesting questions that that view leaves unanswered. I’m not talking about “deep questions of meaning,” but about something much more straightforward: what exactly does a computational process have to do to qualify as “conscious”?

…

There’s this old chestnut, what if each person on earth simulated one neuron of your brain, by passing pieces of paper around. It took them several years just to simulate a single second of your thought processes. Would that bring your subjectivity into being? Would you accept it as a replacement for your current body? If so, then what if your brain were simulated, not neuron-by-neuron, but by a gigantic lookup table? That is, what if there were a huge database, much larger than the observable universe (but let’s not worry about that), that hardwired what your brain’s response was to every sequence of stimuli that your sense-organs could possibly receive. Would that bring about your consciousness? Let’s keep pushing: if it would, would it make a difference if anyone actually consulted the lookup table? Why can’t it bring about your consciousness just by sitting there doing nothing?

To these standard thought experiments, we can add more. Let’s suppose that, purely for error-correction purposes, the computer that’s simulating your brain runs the code three times, and takes the majority vote of the outcomes. Would that bring three “copies” of your consciousness into being? Does it make a difference if the three copies are widely separated in space or time—say, on different planets, or in different centuries? Is it possible that the massive redundancy taking place in your brain right now is bringing multiple copies of you into being?

…

Maybe my favorite thought experiment along these lines was invented by my former student Andy Drucker. In the past five years, there’s been a revolution in theoretical cryptography, around something called Fully Homomorphic Encryption (FHE), which was first discovered by Craig Gentry. What FHE lets you do is to perform arbitrary computations on encrypted data, without ever decrypting the data at any point. So, to someone with the decryption key, you could be proving theorems, simulating planetary motions, etc. But to someone without the key, it looks for all the world like you’re just shuffling random strings and producing other random strings as output.

You can probably see where this is going. What if we homomorphically encrypted a simulation of your brain? And what if we hid the only copy of the decryption key, let’s say in another galaxy? Would this computation—which looks to anyone in our galaxy like a reshuffling of gobbledygook—be silently producing your consciousness?

When we consider the possibility of a conscious quantum computer, in some sense we inherit all the previous puzzles about conscious classical computers, but then also add a few new ones. So, let’s say I run a quantum subroutine that simulates your brain, by applying some unitary transformation U. But then, of course, I want to “uncompute” to get rid of garbage (and thereby enable interference between different branches), so I apply U-1. Question: when I apply U-1, does your simulated brain experience the same thoughts and feelings a second time? Is the second experience “the same as” the first, or does it differ somehow, by virtue of being reversed in time? Or, since U-1U is just a convoluted implementation of the identity function, are there no experiences at all here?

Here’s a better one: many of you have heard of the Vaidman bomb. This is a famous thought experiment in quantum mechanics where there’s a package, and we’d like to “query” it to find out whether it contains a bomb—but if we query it and there is a bomb, it will explode, killing everyone in the room. What’s the solution? Well, suppose we could go into a superposition of querying the bomb and not querying it, with only ε amplitude on querying the bomb, and √(1-ε2) amplitude on not querying it. And suppose we repeat this over and over—each time, moving ε amplitude onto the “query the bomb” state if there’s no bomb there, but moving ε2 probability onto the “query the bomb” state if there is a bomb (since the explosion decoheres the superposition). Then after 1/ε repetitions, we’ll have order 1 probability of being in the “query the bomb” state if there’s no bomb. By contrast, if there is a bomb, then the total probability we’ve ever entered that state is (1/ε)×ε2 = ε. So, either way, we learn whether there’s a bomb, and the probability that we set the bomb off can be made arbitrarily small. (Incidentally, this is extremely closely related to how Grover’s algorithm works.)

OK, now how about the Vaidman brain? We’ve got a quantum subroutine simulating your brain, and we want to ask it a yes-or-no question. We do so by querying that subroutine with ε amplitude 1/ε times, in such a way that if your answer is “yes,” then we’ve only ever activated the subroutine with total probability ε. Yet you still manage to communicate your “yes” answer to the outside world. So, should we say that you were conscious only in the ε fraction of the wavefunction where the simulation happened, or that the entire system was conscious? (The answer could matter a lot for anthropic purposes.)

To sum up: Brian’s notion that consciousness is the same as computation raises more issues than it solves; in particular, the possibility that if suffering is computable, it may also be uncomputable/reversible, would suggest s-risks aren’t as serious as FRI treats them.

Objection 8: Dangerous combination

Three themes which seem to permeate FRI’s research are:

(1) Suffering is the thing that is bad.

(2) It’s critically important to eliminate badness from the universe.

(3) Suffering is impossible to define objectively, and so we each must define what suffering means for ourselves.

Taken individually, each of these seems reasonable. Pick two, and you’re still okay. Pick all three, though, and you get A Fully General Justification For Anything, based on what is ultimately a subjective/aesthetic call.

Much can be said in FRI’s defense here, and it’s unfair to single them out as risky: in my experience they’ve always brought a very thoughtful, measured, cooperative approach to the table. I would just note that ideas are powerful, and I think theme (3) is especially pernicious if incorrect.

III. QRI’s alternative

Analytic functionalism is essentially a negative hypothesis about consciousness: it’s the argument that there’s no order to be found, no rigor to be had. It obscures this with talk of “function”, which is a red herring it not only doesn’t define, but admits is undefinable. It doesn’t make any positive assertion. Functionalism is skepticism- nothing more, nothing less.

But is it right?

Ultimately, I think these a priori arguments are much like people in the middle ages arguing whether one could ever formalize a Proper System of Alchemy. Such arguments may in many cases hold water, but it’s often difficult to tell good arguments apart from arguments where we’re just cleverly fooling ourselves. In retrospect, the best way to *prove* systematized alchemy was possible was to just go out and *do* it, and invent Chemistry. That’s how I see what we’re doing at QRI with Qualia Formalism: we’re assuming it’s possible to build stuff, and we’re working on building the object-level stuff.

What we’ve built with QRI’s framework

Note: this is a brief, surface-level tour of our research; it will probably be confusing for readers who haven’t dug into our stuff before. Consider this a down-payment on a more substantial introduction.

My most notable work is Principia Qualia, in which I lay out my meta-framework for consciousness (a flavor of dual-aspect monism, with a focus on Qualia Formalism) and put forth the Symmetry Theory of Valence (STV). Essentially, the STV is an argument that much of the apparent complexity of emotional valence is evolutionarily contingent, and if we consider a mathematical object isomorphic to a phenomenological experience, the mathematical property which corresponds to how pleasant it is to be that experience is the object’s symmetry. This implies a bunch of testable predictions and reinterpretations of things like what ‘pleasure centers’ do (Section XI; Section XII). Building on this, I offer the Symmetry Theory of Homeostatic Regulation, which suggests understanding the structure of qualia will translate into knowledge about the structure of human intelligence, and I briefly touch on the idea of Neuroacoustics.

Likewise, my colleague Andrés Gomez Emilsson has written about the likely mathematics of phenomenology, including The Hyperbolic Geometry of DMT Experiences, Tyranny of the Intentional Object, and Algorithmic Reduction of Psychedelic States. If I had to suggest one thing to read in all of these links, though, it would be the transcript of his recent talk on Quantifying Bliss, which lays out the world’s first method to objectively measure valence from first principles (via fMRI) using Selen Atasoy’s Connectome Harmonics framework, the Symmetry Theory of Valence, and Andrés’s CDNS model of experience.

These are risky predictions and we don’t yet know if they’re right, but we’re confident that if there is some elegant structure intrinsic to consciousness, as there is in many other parts of the natural world, these are the right kind of risks to take.

I mention all this because I think analytic functionalism- which is to say radical skepticism/eliminativism, the metaphysics of last resort- only looks as good as it does because nobody’s been building out any alternatives.

IV. Closing thoughts

FRI is pursuing a certain research agenda, and QRI is pursuing another, and there’s lots of value in independent explorations of the nature of suffering. I’m glad FRI exists, everybody I’ve interacted with at FRI has been great, I’m happy they’re focusing on s-risks, and I look forward to seeing what they produce in the future.

On the other hand, I worry that nobody’s pushing back on FRI’s metaphysics, which seem to unavoidably lead to the intractable problems I describe above. FRI seems to believe these problems are part of the territory, unavoidable messes that we just have to make philosophical peace with. But I think that functionalism is a bad map, that the metaphysical messes it leads to are much worse than most people realize (fatal to FRI’s mission), and there are other options that avoid these problems (which, to be fair, is not to say they have no problems).

Ultimately, FRI doesn’t owe me a defense of their position. But if they’re open to suggestions on what it would take to convince a skeptic like me that their brand of functionalism is viable, or at least rescuable, I’d offer the following:

Re: Objection 1 (motte-and-bailey), I suggest FRI should be as clear and complete as possible in their basic definition of suffering. In which particular ways is it ineffable/fuzzy, and in which particular ways is it precise? What can we definitely say about suffering, and what can we definitely never determine? Preregistering ontological commitments and methodological possibilities would help guard against FRI’s definition of suffering changing based on context.

Re: Objection 2 (intuition duels), FRI may want to internally “war game” various future scenarios involving AGI, WBE, etc, with one side arguing that a given synthetic (or even extraterrestrial) organism is suffering, and the other side arguing that it isn’t. I’d expect this would help diagnose what sorts of disagreements future theories of suffering will need to adjudicate, and perhaps illuminate implicit ethical intuitions. Sharing the results of these simulated disagreements would also be helpful in making FRI’s reasoning less opaque to outsiders, although making everything transparent could lead to certain strategic disadvantages.

Re: Objection 3 (convergence requires common truth), I’d like FRI to explore what exactly might drive consilience/convergence in theories of suffering, and what precisely makes one theory of suffering better than another, and ideally to evaluate a range of example theories of suffering under these criteria.

Re: Objection 4 (assuming that consciousness is a reification produces more confusion, not less), I would love to see a historical treatment of reification: lists of reifications which were later dissolved (e.g., élan vital), vs scattered phenomena that were later unified (e.g., electromagnetism). What patterns do the former have, vs the latter, and why might consciousness fit one of these buckets better than the other?

Re: Objection 5 (the Hard Problem of Consciousness is a red herring), I’d like to see a more detailed treatment of what kinds of problem people have interpreted the Hard Problem as, and also more analysis on the prospects of Qualia Formalism (which I think is the maximally-empirical, maximally-charitable interpretation of the Hard Problem). It would be helpful for us, in particular, if FRI preregistered their expectations about QRI’s predictions, and their view of the relative evidence strength of each of our predictions.

Re: Objection 6 (mapping to reality), this is perhaps the heart of most of our disagreement. From Brian’s quotes, he seems split on this issue; I’d like clarification about whether he believes we can ever precisely/objectively map specific computations to specific physical systems, and vice-versa. And if so— how? If not, this seems to propagate through FRI’s ethical framework in a disastrous way, since anyone can argue that any physical system does, or does not, ‘code’ for massive suffering, and there’s no principled way derive any ‘ground truth’ or even pick between interpretations in a principled way (e.g. my popcorn example). If this isn’t the case— why not?

Brian has suggested that “certain high-level interpretations of physical systems are more ‘natural’ and useful than others” (personal communication); I agree, and would encourage FRI to explore systematizing this.

It would be non-trivial to port FRI’s theories and computational intuitions to the framework of “hypercomputation”– i.e., the understanding that there’s a formal hierarchy of computational systems, and that Turing machines are only one level of many– but it may have benefits too. Namely, it might be the only way they could avoid Objection 6 (which I think is a fatal objection) while still allowing them to speak about computation & consciousness in the same breath. I think FRI should look at this and see if it makes sense to them.

Re: Objection 7 (FRI doesn’t fully bite the bullet on computationalism), I’d like to see responses to Aaronson’s aforementioned thought experiments.

Re: Objection 8 (dangerous combination), I’d like to see a clarification about why my interpretation is unreasonable (as it very well may be!).

In conclusion- I think FRI has a critically important goal- reduction of suffering & s-risk. However, I also think FRI has painted itself into a corner by explicitly disallowing a clear, disagreement-mediating definition for what these things are. I look forward to further work in this field.

Mike Johnson

Qualia Research Institute

Acknowledgements: thanks to Andrés Gomez Emilsson, Brian Tomasik, and Max Daniel for reviewing earlier drafts of this.

Sources:

My sources for FRI’s views on consciousness:

Flavors of Computation are Flavors of Consciousness:

https://foundational-research.org/flavors-of-computation-are-flavors-of-consciousness/

Is There a Hard Problem of Consciousness?

http://reducing-suffering.org/hard-problem-consciousness/

Consciousness Is a Process, Not a Moment

http://reducing-suffering.org/consciousness-is-a-process-not-a-moment/

How to Interpret a Physical System as a Mind

http://reducing-suffering.org/interpret-physical-system-mind/

Dissolving Confusion about Consciousness

http://reducing-suffering.org/dissolving-confusion-about-consciousness/

Debate between Brian & Mike on consciousness:

https://www.facebook.com/groups/effective.altruists/permalink/1333798200009867/?comment_id=1333823816673972&comment_tracking=%7B%22tn%22%3A%22R9%22%7D

Max Daniel’s EA Global Boston 2017 talk on s-risks:

https://www.youtube.com/watch?v=jiZxEJcFExc

Multipolar debate between Eliezer Yudkowsky and various rationalists about animal suffering:

https://rationalconspiracy.com/2015/12/16/a-debate-on-animal-consciousness/

The Internet Encyclopedia of Philosophy on functionalism:

http://www.iep.utm.edu/functism/

Gordon McCabe on why computation doesn’t map to physics:

http://philsci-archive.pitt.edu/1891/1/UniverseCreationComputer.pdf

Toby Ord on hypercomputation, and how it differs from Turing’s work:

https://arxiv.org/abs/math/0209332

Luke Muehlhauser’s OpenPhil-funded report on consciousness and moral patienthood:

http://www.openphilanthropy.org/2017-report-consciousness-and-moral-patienthood

Scott Aaronson’s thought experiments on computationalism:

http://www.scottaaronson.com/blog/?p=1951

Selen Atasoy on Connectome Harmonics, a new way to understand brain activity:

https://www.nature.com/articles/ncomms10340

My work on formalizing phenomenology:

My meta-framework for consciousness, including the Symmetry Theory of Valence:

http://opentheory.net/PrincipiaQualia.pdf

My hypothesis of homeostatic regulation, which touches on why we seek out pleasure:

http://opentheory.net/2017/05/why-we-seek-out-pleasure-the-symmetry-theory-of-homeostatic-regulation/

My exploration & parametrization of the ‘neuroacoustics’ metaphor suggested by Atasoy’s work:

http://opentheory.net/2017/06/taking-brain-waves-seriously-neuroacoustics/

My colleague Andrés’s work on formalizing phenomenology:

A model of DMT-trip-as-hyperbolic-experience:

https://qualiacomputing.com/2017/05/28/eli5-the-hyperbolic-geometry-of-dmt-experiences/

June 2017 talk at Consciousness Hacking, describing a theory and experiment to predict people’s valence from fMRI data:

https://qualiacomputing.com/2017/06/18/quantifying-bliss-talk-summary/

A parametrization of various psychedelic states as operators in qualia space:

https://qualiacomputing.com/2016/06/20/algorithmic-reduction-of-psychedelic-states/

A brief post on valence and the fundamental attribution error:

https://qualiacomputing.com/2016/11/19/the-tyranny-of-the-intentional-object/

A summary of some of Selen Atasoy’s current work on Connectome Harmonics:

https://qualiacomputing.com/2017/06/18/connectome-specific-harmonic-waves-on-lsd/
Taking ‘brain waves’ seriously: Neuroacoustics

Posted:Wed, 14 Jun 2017 19:25:40 +0000

Our research collective has been doing a lot of work touching on brain dynamics, resonance, and symmetry: see here and here (video). Increasingly, a new implicit working ontology I’m calling ‘Neuroacoustics’ is taking shape. This is a quick outline of that new ontology.

What is Neuroacoustics?

A common frame in neuroscience is to talk about ‘brain waves’; alpha waves, gamma waves, and so on. Neuroacoustics essentially doubles down on this wave metaphor, but instead of focusing on specific wave frequencies, it focuses on their relative frequencies and the properties of the substrates in which these waves travel. In short, I propose that neural activity can be understood as waves in an adaptive medium, and adaptive behavior as arising from the interplay between the information encoded in the waves and the information encoded in the resonant and topological properties of the medium. Wave dynamics would arise from (e.g.):

(1) Inherent acoustics of the substrate. I.e., different brain regions propagate waves & resonate differently based on local connectivity, myelination, & gene expression.

(2) Topological permutations of multiple resonant substrates. Trumpets produce beautiful order by connecting many resonant chambers together in a way which leads to signal selection, purification, and amplification. The brain’s internal topology is likely organized around similar principles, just in a more complex, layered way that imparts semantic context.

(3) Adaptive interactions between wave & substrate. My intuition is that we could model short-term and long-term potentiation in terms of wave resonance wearing ‘grooves’ (wave guides) into the substrate. Contrawise, perhaps some substrates are primed for the opposite effect: patterns which are too resonant trigger the brain’s defense against monopolization of neural resources (“boredom”).

(4) Interaction between waves – i.e., constructive & destructive interference, especially when periodicities overlap or near-overlap.

(5) Neurotransmitters influencing acoustics. We can model neurotransmitters as ‘resonance-shifters’ which operate on a per-region basis to change their internal (Darwinian) pattern selection landscapes. Emotions, then, can be thought of as coordinated bundles of these resonance-shifters, with each bundle calibrated for different environmental challenges.

(6) Interactions with other periodic systems & patterns. Our brains are very good at getting ‘in synch with’ all manner of other systems (entrainment), especially other brains.

Truust Neuroimaging uses (1) and (2) to build 3d reconstructions of activity from EEG data; Selen Atasoy uses (1), (2), (5), and (6) to model the effects of LSD and music on brain harmonics; my colleague Andres recently gave a talk proposing how to use (1), (2), (4), (5), and (6) to build an objective measure for valence in humans. And we aren’t done yet- there’s a lot of cool stuff this framework makes easier, just waiting to be built.

Speculation: the Periodic Table of Brain Regions

We can approach the description above as a working metaphor and call it done, but there may also be room for something more formal. Specifically, Neuroacoustics seems to imply we could build a principled parametrization of neural substrate based on conditional resonance. In practice this would entail measuring various brain regions’ stochastic resonant properties, then classifying the effects of various neurotransmitters (serotonin, dopamine, opioids, etc) on these resonant properties, and organizing the result by periodic structure. I call this the “Periodic Table of Brain Regions”.

If this parametrization pans out, it could help comprehensively parametrize all mind-altering effects of all psychoactive substances, from Benadryl to chocolate to nicotine, and also clear a path for the next step: targeted, precision interventions.

See also: Atasoy’s work on Connectome Harmonics and Emilsson’s extension to valence; Adaptive Resonance Theory; Steve Lehar‘s work; Smolensky’s Computational Harmony; a broad range of work on neural entrainment.
Why we seek out pleasure: the Symmetry Theory of Homeostatic Regulation

Posted:Fri, 26 May 2017 20:18:46 +0000
Why do we seek out pleasure- what Freud called the “pleasure principle“?

More accurately: why do we seem to seek out pleasure most of the time, but occasionally seem indifferent to it or even averse to it, e.g. in conditions such as anhedonia & depression?

My answer in a nutshell:
1. Our brain networks are calibrated to the environment such that their symmetry gradients are tied to survival requirements. This is the core algorithm by which our brains regulate homeostasis. (Argued below)
2. Symmetry in the mathematical representation of phenomenology corresponds to pleasure. (Argued in Principia Qualia)
3. In combination, then, pleasure-seeking is what it feels like when our brain follows its core (ancestral/default) algorithm for maintaining homeostasis.
An algorithmic model of brain-dynamics-as-symmetry-gradient-climbing across multiple scales:

This is a simple model of a complex system, and as such it can only be approximately correct. But I think it’s also substantially correct, and that any future gears-level model will need to be compatible with this model. It centers around coalitions & symmetry & attractor basins.

The basics: the brain is a complex chaotic system centered around self-organized criticality. This criticality is based around coalitions of neurons, which “vote” (and thus produce activity cascades) with their firing patterns.

The brain has many attractor basins, but the central & most important one is harmony (and harmony ultimately boils down to symmetry). Thus the brain is always searching for actions, plans, and environments which lead to activity cascades toward more harmonious/symmetrical states. But never too harmonious! We can think of boredom as a very sophisticated ‘anti-wireheading’ technology that gradually pushes our networks’ symmetry (and thus stability / disposition toward inaction) down when our environment becomes too predictable. Our brains are calibrated to dynamically circle around the symmetry basin without ever falling in and getting stuck, since the computationally-useful part of this is the high-dimensional symmetry gradient which is finely calibrated to our environment and homeostatic requirements… not the actual symmetry itself. (You’re a cruel one, Azathoth…)

Basically, brains try to climb their internal symmetry gradients through both internal & external means: when we move our bodies, change our environment, make plans, listen to music, eat food, respond to threats, and so on, we’re increasing our internal symmetry gradient (or at least preventing it from dropping). Symmetry gradients are how our brain internally represents value, and much of the brain’s complexity has to do with calibrating these gradients to the environment in order to make them computationally useful. I.e., to make the contextually correct things increase & decrease symmetry in core regulatory networks.

Getting a little more technical:

A good algorithmic theory of cognition will collect, unify, and simplify a lot of things that look like odd psychological quirks, and recast them as deeply intertwined with, and naturally arising out of, how our brains process information. I’m optimistic that Symmetry Theory will be able to do just this- e.g.,
- Cognitive dissonance happens when two (or more) patterns in your head are battling for your neural real estate, and they’re incompatible– i.e., they’re collectively dissonant/asymmetrical.
- Denial is what happens when your brain attempts to isolate/quarantine such patterns, and is actively working to prevent this tug-of-war for neurons.
This model implies that your brain can evaluate the “internal consistency/harmony” of a neural pattern, and reject it if there’s a negative result- and also the “simulated relative compatibility/harmony” of two neural patterns, and try to keep them isolated if there’s a negative result. I’d suggest the best way to understand this is in terms of projective geometry, resonance, and symmetry: i.e., to evaluate a pattern’s “internal harmony” and whether it ‘runs well (is stable) on existing hardware’, the brain uses principles of resonance to apply certain geometric projections (high-dimensional-to-lower-dimensional transformations) to the pattern to see if the result is stable (unchanged, or predictably oscillating, or still strongly resonant) under these transforms. Stable patterns are allocated territory; unstable ones (=dangerous neural code) are not. The internal mechanics of this will vary across brain areas (based on the specific resonance profile of each area) and emotional states, which might contribute to how certain types of information tend to end up in certain brain regions. Likewise, this could explain how moods coordinate information processing– by changing the resonance landscape in the brain, thus preferentially selecting for certain classes of patterns over others. A core implication of this model is that different kinds of dissonance will drive different kinds of behavior (feel like different kinds of imperatives), and based on what action is needed, a mood may create (or be the creation of) a certain kind of dissonance.

Stepping back:

Ultimately, this is the first move in an attempt to subsume all of cognitive processing under the mantle of symmetry (similar to Smolensky and Legendre’s Harmonic Mind hypothesis, but not limited to phonology), just like QRI is subsuming all valence under the mantle of symmetry. I think this is not merely ‘a useful way to model the brain’- I think it’s literally how the brain works. We could call this the Symmetry Theory of Affective Cognition – with the added implication that most things the brain does, from high-level cognition to low-level autonomic regulation, are all variations on this core theme. More technically, I’d suggest calling this the Symmetry Theory of Homeostatic Regulation.

This package of mechanisms doesn’t seem like the only way to build intelligent systems, but it does seem like a particularly good one that can start simple yet climb the ladder of abstraction, allows implicit social coordination from the bottom-up (“‘vibes’ are like, a real thing, man”), and can evolve in incrementally-useful forms. Symmetry is just an insanely powerful principle to build or evolve a computational system around.

Waxing philosophical, I think this is my favorite high-compression description of what humans fundamentally are— the essence of this “human condition” thing we seem to only have indirect access to. I.e., to be human is to be a complex dynamic physical system which maintains homeostasis via a strong symmetry attractor, on one hand, yet on the other hand also has sophisticated anti-wireheading technologies that ensure we never stay deep in this attractor for very long. If we want to build a computer with human-like qualia, human-like drives, and human-like cognition, we’d need to build it around these principles. (Necessary, but not sufficient, etc)

Implications & extensions:

This is only a down payment on an actual gears-level account of cognitive processing. A few brief suggestions: it’ll probably be fruitful to:
1. try to match natural failure modes of this system to various psychological & psychiatric conditions- e.g., let’s page through the DSM-V and see if we can spot dysfunctions which correspond to breakdowns in the processes I talk about above, and if that offers a new way to carve the problem space of psychiatry;
2. dig into the “brain regions as neuroacoustic chambers” angle;
3. explore how the brain’s symmetry gradients are initially (developmentally) correlated with the environment, and how the brain actively & passively maintains this correlation (Adam Safron is doing interesting work here);
4. see if this sort of system could emerge from a handful of simple, known neural rules (e.g., “neurons that fire together, wire together”);
5. try to rebuild various top-down psychological theories (like CT, or CFAR stuff) in terms of Symmetry Theory;
6. try to reclassify primary neurotransmitters such as opioids/dopamine/serotonin in terms of their local & global effects on neuroacoustics. (Andres Gomez Emilsson is doing interesting work here, and will be speaking about it on June 7th.)
In particular, I think it’d be useful to brainstorm about Symmetry Theory, affective disorders, and effective treatments. E.g., listening to music might be unexpectedly effective at helping regulate mood.

Looking more abstractly, we could also view social dynamics, moods, and memetic desire through the lens of transmissible neuroacoustics (Girard’s mimesis).

Symmetry Theory’s relationship to similar theories:

How does this framework compare to e.g.,
- Schmidhuber’s Compression Drive, which posits that the brain-as-learning-organ is constantly seeking efficient compression schemes for input and storage, and actively seeks out elegance to improve its compression library;
- Friston’s Free Energy Principle, which argues that “[A]ny self-organizing system that is at equilibrium with its environment must minimize its free energy. The principle is essentially a mathematical formulation of how adaptive systems (that is, biological agents, like animals or brains) resist a natural tendency to disorder. … In short, the long-term (distal) imperative — of maintaining states within physiological bounds — translates into a short-term (proximal) avoidance of surprise.”
- Seth et al.’s predictive coding / Bayesian Brain framework, which suggests that cognition revolves around sophisticated predictive models of sensory input, and will reduce bandwidth requirements by only passing surprises up the cognitive hierarchy.
In short, all these theories seem to be like different blind men examining different parts of the same elephant: they’re essentially doing the same thing, just in different ways. Here’s how I see the Symmetry Theory of Homeostatic Regulation in this context:
- Mechanism: Schmidhuber et al.’s work operate on Marr’s computational level, and describe the high-level logic of the brain, whereas Symmetry Theory is an algorithmic description and supplies a mechanism by which theories like Compression Drive could be implemented. This focus on mechanism lets us see more details, e.g. how patterns can compete, how patterns can interact with substrate & selection landscapes, how boredom/denial/etc work, how pattern compression occurs, etc. (There are probably multiple algorithms you could build CD out of, but ST seems by far the best one.)
- Generality/scope: symmetry gradients seem usable for regulation of homeostasis at all levels, from simple autonomic functions to complex cognition, whereas compression drive deals with information, and as such seems narrower in scope & requires more complexity/context. Predictive coding theories may fall somewhere in the middle, and essentially describe efficient information flows in a complex ‘symmetry attractor type’ system.
- Frame-invariance (ease of application): One of the biggest problems with the class of theories Schmidhuber’s work belongs to (Free Energy, Bayesian Brain, etc) is that it’s ambiguous how they actually apply to any given system. Symmetry Theory should be a lot easier to apply since symmetry is literally the thing that enables frame-invariance. (This is not meant as a critique of Schmidhuber’s/Friston’s/Seth’s work, of which I am a huge fan.)
In short, I think symmetry, and the Symmetry Theory of Homeostatic Regulation, can play a part in unifying & contextualizing theories in this class, since the core of each of these theories is some different feature or flavor of symmetry/regularity. However, we haven’t done the actual work of unification yet, and need to do some deep thinking about what kinds of problems each approach is best at, and why.

Question for my readers:

What would you want out of a theory like this? If you’re the customer, what’s the product you’d like to see? How can I translate this from a bunch of words into something that makes your life better?

Inspirations & surrounding literature:

Principia Qualia; many conversations with Andres Gomez Emilsson; Smolensky‘s Harmony theory of neuro-linguistic processing; Predictive coding / Bayesian Brain work by Friston, Seth, Schmidhuber, and others; conversations with Adam Safron, and his entrainment model of orgasm; Perceptual Control Theory; Marblestone et al.‘s three hypotheses; David Pearce’s notion of the pleasure-pain axis as computationally relevant; Lin & Tegmark‘s findings of symmetry in deep learning; Trivers‘ arguments about symmetry & developmental biology; various conversations with PG & Romeo & Randal & Radhika, & probably others I’m forgetting.
Symmetry Theory of Valence “Explain Like I’m 5” edition

Posted:Sat, 15 Apr 2017 08:07:40 +0000

When someone on Reddit says “ELI5”, it means “I’m having a hard time understanding this, could you explain it to me like I’m 5 years old?”

Here’s my attempt at an “ELI5” for the Symmetry Theory of Valence (Part II of Principia Qualia).

—
We can think of conscious experiences as represented by a special kind of mathematical shape. The feeling of snowboarding down a familiar mountain early in the morning with the air smelling of pine trees is one shape; the feeling of waking up to your new kitten jumping on your chest and digging her claws into your blankets is another shape. There are as many shapes as there are possible experiences.

Now, the interesting part: if we try to sort experiences by how good they feel, is there a pattern to which shapes represent more pleasant experiences? I think there is, and I think this depends on the symmetry of the shape.

There’s a lot of evidence for this, and if this is true, it’s super-important! It could lead to way better painkillers, actual cures for things like Depression, and it would also give us a starting point for turning consciousness research into a real science (just like how alchemy turned into chemistry). Basically, it would totally change the world.

But first thing’s first: we need to figure out if it’s true or not.
—
Principia Qualia: the executive summary

Posted:Wed, 07 Dec 2016 07:27:02 +0000
Put simply, Principia Qualia (click for full version) is a blueprint for building a new Science of Qualia.

PQ begins by considering a rather modest question: what is emotional valence? What makes some things feel better than others?

This sounds like the sort of clear-cut puzzle affective neuroscience should be able to solve, yet all existing answers to this question are incoherent or circular. Giulio Tononi’s Integrated Information Theory (IIT) is an example of the kind of quantitative theory which could in theory address valence in a principled way, but unfortunately the current version of IIT is both flawed and incomplete. I offer a framework to resolve generalize IIT by distilling the problem of consciousness into eight discrete & modular sub-problems (of which IIT directly addresses five).

Finally, I return to valence, and offer a crisp, falsifiable hypothesis as to what it is in terms of something like IIT’s output, and discuss novel implications for neuroscience.

The most important takeaways are:
1. My “Eight Problems” framework is a blueprint for a “full-stack” science of consciousness & qualia. Addressing all eight problems is both necessary & sufficient for ‘solving’ consciousness.
2. One of IIT’s core shortcomings is that it doesn’t give any guidance for what its output means. I offer some useful heuristics.
3. Emotional valence could be the easiest quale to reverse-engineer– the “c. elegans of qualia.”
Section-by-section highlights:

I. Why some things feel better than others: the view from neuroscience (2600 words)

Affective neuroscience knows a lot about valence… but its knowledge is very haphazard, disorganized, and often circular. The techniques it uses are good at assembling data, but not so good at finding clear patterns in the data, or knowing what data to gather in the first place.

II. Clarifying the Problem of Valence (900 words)

It could be that there are no clean answers to be found, and this is the best we can do. But I don’t buy that- I think we’re just looking at it from the wrong level of abstraction.

To really get traction on the problem of what makes some things feel better than others, we need to look for universal principles true in all conscious systems, not just things that are often true in the human brain.

This also implies that any attempt to solve valence which tries to avoid addressing the larger mystery of consciousness simply won’t work.

III. The Integrated Information Theory of consciousness (1900 words)

IIT is an attempt at a fully quantitative theory of consciousness. Think of it like a mathematical translation function: give it a system (like the brain), and it gives you a mathematical representation of what it feels like to be that system. It’s currently the best (and only) attempt at a full theory of consciousness we have.

IV. Critiques of IIT (2600 words)

Unfortunately, the current iteration of IIT is probably wrong, and definitely incomplete. There are some problems with its math, it’s a little vague on what its input should be, and it says almost nothing about what its output means.

V. Alternative versions of IIT: Perceptronium and FIIH (1500 words)

Other people are trying to translate IIT’s math into the language of physics in order to fix these problems. Unfortunately, these attempts are mostly stuck at the ‘idea’ stage.

VI. Summary and synthesis: eight problems for a new science of consciousness (900 words)

(Sections I-V are a literature review; Sections VI onward are original work.)

Even if IIT is wrong, we can use it as a template for understanding what tasks a theory of consciousness should be able to do:

Secondly, these categories break down into well-defined sub-problems:

Solve all eight problems, and you’ve solved consciousness. (Easy, right??)

VII. Three principles for a mathematical derivation of valence (1000 words)

IIT’s general approach implies that the problem of valence is a problem of mathematical interpretation: i.e.,

“Given a mathematical representation of my qualia (e.g., IIT’s output), what mathematical property of this representation corresponds to how pleasant it feels to be me?”

Here’s a Venn Diagram of assumptions:

(I explain each of these further in the paper.)

VIII. Distinctions in qualia: charting the explanation space for valence (1000 words)

If valence is some mathematical property of IIT’s output (or some future version of IIT which addresses its flaws), then what kind of property should we look for?

Here’s the clever heuristic: if we assume consciousness has a mathematical representation, then for any distinction you can make about qualia, you get a corresponding distinction in the domain of mathematics ‘for free’. And vice-versa! We can use this to explore what kind of mathematical property valence could correspond with:
- Valence seems global rather than local: when you eat a candy bar, everything feels good, not just your mouth;
- Valence seems simple rather than complex: i.e., in terms of Kolmogorov complexity;
- Valence seems atomic rather than composite: it’s a building block for more complex emotions;
- Valence seems intuitively important rather than intuitively trivial: it’s hard to miss.
Why this matters: if valence is [global,simple,atomic,intuitively important], then its mathematical representation is too. This significantly narrows the search space.

IX. Summary of heuristics for reverse-engineering the pattern for valence (2000 words)

This is a “throw everything but the kitchen sink at the problem” section: let’s list facts that we know, and clever heuristics, and hope some patterns emerge. Friston’s Free Energy Principle, Seth’s Predictive Error Minimization, Smolensky’s Computational Harmony metric, and my own “Non-adaptedness Principle” all make an appearance here. Also, I have this neat triangle graphic describing the state space of valence (it’s not just a one-dimensional pain-pleasure axis):

X. A simple hypothesis about valence (2000 words)

There’s a lot of context in the paper that I can’t give justice to in this summary- but given a mathematical object isomorphic to a system’s qualia, I think I’ve identified the mathematical property which corresponds to its valence. This property is the object’s symmetry.

XI. Testing this hypothesis today (1700 words)

Obviously, a hypothesis is only as good as the predictions it makes. I propose three specific tests:

(1) More pleasant brain states should be more compressible (as measured by zipping EEG data, controlling for degree of consciousness);

(2) Low-power rhythmic TMS of consciousness centers such as the thalamus at consonant frequencies should feel dramatically better than such stimulation at subtly dissonant frequencies;

(3) Similarly, consonant stimulation of the vagus nerve should feel better than stimulation with dissonant patterns (upshot: possibly testable with consumer-grade gear).

XII. Taking stock (1800 words)

If I’m right that symmetry in the mathematical object isomorphic to a conscious experience maps to valence, it allows us to recontextualize certain things in neuroscience in interesting ways. For instance:

On the anatomy & network topology of valence:

My hypothesis strongly implies that ‘hedonic’ brain regions influence mood by virtue of acting as ‘tuning knobs’ for symmetry/harmony in the brain’s consciousness centers. Likewise, nociceptors, and the brain regions which gate & interpret their signals, will be located at critical points in brain networks, able to cause large amounts of salience-inducing antisymmetry very efficiently. We should also expect rhythm to be a powerful tool for modeling brain dynamics involving valence- for instance, we should be able to extend (Safron 2016)’s model of rhythmic entrainment in orgasm to other sorts of pleasure.

More speculatively…

On valence & neuropharmacology:

Non-opioid painkillers and anti-depressants are complex, but it may turn out that a core mechanism by which they act is by introducing noise into neural activity and connectivity, respectively. This would explain the odd findings that acetaminophen blunts acute pleasure (Durso, Luttrell, and Way 2015), and that anti-depressants can induce long-term affective flattening.

This would also predict that psychedelic substances, although often pleasurable, actually increase emotional variance by biasing the brain toward symmetrical structure, and could result in enhanced pain if this structure is then broken– i.e., they are in this sense the opposite of painkillers. Additionally, we may find that some uncomfortable sensations are caused by ‘competing symmetries’- patterns that are internally symmetrical but not symmetrical to each other, which would predict complex and sometimes destructive interactions between different normally-pleasurable activities and psychoactives.

Furthermore, I would anticipate that severe tinnitus could lead to affective flattening for similar interference-based reasons: insofar as the brain’s subconscious preprocessing can’t tune it out, the presence of a constant pattern in consciousness would likely make it more difficult to generate symmetries (valence) on-the-fly. This would also imply that the specific frequency pattern of the perceived tinnitus sensation may matter more than is commonly assumed.

On self-organization & deep learning:

My hypothesis implies that symmetry/harmony is a core component of the brain’s organizational & computational syntax: specifically, we should think of symmetry as one (of many) dynamic attractors in the brain.

This suggests that mammals got a bit lucky that we evolved to seek out pleasure! But not that lucky, since symmetry is a very functionally-relevant anduseful property for systems to self-organize around, for at least two reasons:

First, self-organizing systems such as the brain must develop some way to perform error-correction, measure & maintain homeostasis, and guide & constrain morphological development. Symmetry-as-a-dynamic-attractor is a profoundly powerful solution to all of these which could evolve in incrementally-useful forms, and so symmetry-seeking seems like a common, perhaps nigh-universal evolutionary path to take. Indeed, it might be exceedingly difficult to develop a system with complex adaptive traits without heavy reliance upon principles of symmetry.

Second, the brain embodies principles of symmetry because it’s an efficient structure for modeling our world. (Lin and Tegmark 2016) note that physics and deep learning neural networks display cross-domain parallels such as “symmetry, locality, compositionality and polynomial log-probability”, and that deep learning can often avoid combinatorial explosion due to the fact that the physical world has lots of predictable symmetries, which enable unusually efficient neural network encoding schemes.

On Boredom:

Why do we find pure order & symmetry boring, and not particularly beautiful? I posit boredom is a very sophisticated “anti-wireheading” technology which prevents the symmetry/pleasure attractor basin from being too ‘sticky’, and may be activated by an especially low rate of Reward Prediction Errors (RPEs). Musical features which add mathematical variations or imperfections to the structure of music– e.g., syncopated rhythms (Witek et al. 2014), vocal burrs, etc– seem to make music more addictive and allows us to find long-term pleasure in listening to it, by hacking the mechanic(s) by which the brain implements boredom.

XIII. Closing thoughts (200 words)

I’ve used valence as a ‘pilot project’, but ultimately, the goal is to build a full Science of Qualia — something that can turn qualia research from alchemy into chemistry, and unify our different modes of knowing in neuroscience.

—

If this intrigues you, I suggest checking out the full paper.