Opentheory.net https://opentheory.net Speculations on the Frontiers of Science and Culture Sun, 30 Nov 2025 11:35:56 +0000 en-US hourly 1 https://wordpress.org/?v=6.9.4 A Paradigm for AI Consciousness https://opentheory.net/2024/06/a-paradigm-for-ai-consciousness/ Wed, 12 Jun 2024 15:09:09 +0000 https://opentheory.net/?p=2056 Michael Edward Johnson, Symmetry Institute. 12 June, 2024. Crossposted from the Seeds of Science journal; available as a PDF there.

Abstract: How can we create a container for knowledge about AI consciousness? This work introduces a new framework based on physicalism, decoherence, and symmetry. Major arguments include (1) physics is a more sturdy ontology for grounding consciousness than Turing-level computation, (2) Wolfram’s ‘branchial space’ is a better measure of an object’s “true shape” than spacetime, (3) electromagnetism is a good proxy for branchial shape, (4) brains and computers have significantly different shapes in branchial space, (5) symmetry considerations will strongly inform a future science of consciousness, and (6) computational efficiency considerations may broadly hedge against “s-risk”.

I. AI consciousness is in the wind

AI is the most rapidly transformative technology ever developed. Consciousness is what gives life meaning. How should we think about the intersection?

A large part of humanity’s future may involve figuring this out. But there are three questions that seem pressing, and we may want to push for answers on:

  1. What is the default fate of the universe if a technological singularity happens and breakthroughs in consciousness research don’t?
  2. What interesting qualia-related capacities does humanity have that synthetic superintelligences might not get by default?
  3. What should CEOs of leading AI companies know about consciousness?

The following is a wide-ranging safari through various ideas and what they imply about these questions. Some items are offered as arguments, others as hypotheses or observations; I’ve tried to link to the core sources in each section. In the interests of exposition, I’ve often erred on the side of being opinionated. But first — some preliminaries about why AI consciousness is difficult to study.

Key references:

II. The social puzzle of AI consciousness

“AI consciousness” is an overloaded term, spanning at least three dimensions:

  1. Human-like responsive sophistication: does this AI have a sense of self? Is it able to understand and contextualize intentions, capabilities, hidden state, and vibes, both in itself and others? The better AI is at playing the games we think of as characteristically human (those which are intuitive to us, and those we ascribe status to), the more “consciousness” it has.
  2. In-group vs Big Other: is the AI part of our team? Our Team has interiority worth connecting with and caring about (“moral patienthood”). The Other does not.
  3. Formal phenomenology: in a narrow and entirely technical and scientifically predictive sense, if you had the equations for qualia (the elements and composition of subjective experience) in-hand, would this AI have qualia?

It’s surprisingly difficult to talk about technical details of AI consciousness for at least three reasons. First, the other non-technical considerations are more accessible and act as conversational attractors. Second, AI consciousness is at the top of an impressive pyramid of perhaps 10-20 semi-independent open problems in metaphysics, and being “right” essentially relies on making the correct assumption for each while having no clean experimental paradigm nor historical tradition to fall back on — in some ways AI consciousness is the final boss of philosophy. 

Third, humans are coalitional creatures — before we judge the truth of a statement, we instinctually evaluate its implications for our alliances. To take an opinionated position on AI consciousness is to risk offending our coalition members, ranging from colleagues, tenure committees, donors, & AI labs, each with their own forms of veto power. This pushes theorists toward big-tent, play-it-safe, defer-to-experts positions.

But in truth, there are no experts in AI consciousness, and it’s exactly in weird positions that may offend intuitive and coalitional sensibilities where the truth is most likely to be found. As Eric Schwitzgebel puts it, “Common sense is incoherent in matters of metaphysics. There’s no way to develop an ambitious, broad-ranging, self-consistent metaphysical system without doing serious violence to common sense somewhere. It’s just impossible. Since common sense is an inconsistent system, you can’t respect it all. Every metaphysician will have to violate it somewhere.”

Key references:

III. Will future synthetic intelligences be conscious?

The question of machine consciousness rests on ‘what kind of thing’ consciousness is. If consciousness is a lossy compression of complex biological processes, similar to “metabolism” or “mood”, asking whether non-biological systems are conscious is a Wittgensteinian type error — i.e. a non-sensical move, similar to asking “what time is it on the sun?” or trying to formalize élan vital. When we run into such category errors, our task is to stop philosophically hitting ourselves; i.e. to debug and dissolve the confusion that led us to apply some category in a context where it’s intrinsically ill-defined. 

On the other hand, if consciousness is a “first-class citizen of reality” that’s definable everywhere, like electric current or gravity, machine consciousness is a serious technical question that merits a serious technical approach. I believe consciousness is such a first-class citizen of reality. Moreover,

I believe synthetic intelligences will be conscious, albeit with a huge caveat.

AIs will be conscious (because most complex heterogenous things probably are): 

Just as we’re made of the same stuff as rocks, trees, and stars, it’s difficult to formalize a theory of consciousness where most compound physical objects don’t have roughly the same ontological status when it comes to qualia. I.e. I take it as reasonable that humans are less ‘a lone flickering candle of consciousness’ and more a particularly intelligent, cohesive, and agentic “qualiafauna” that has emerged from the endemic qualiaflora. We are special — but for quantitative, not qualitative, reasons. Synthetic intelligences will have qualia, because the universe is conscious by default. We don’t have to worry about the light of consciousness going out — though we can still worry about it going weird.

Caveat: Only real things can be conscious. 

There’s a common theme of attributing consciousness to the highest-status primitive. Theology, Psychology, and Physics have each had their time in the sun as “the most real way of parametrizing reality” and thus the ‘home domain’ of consciousness. Now that software is eating the world, computation is king — and consciousness joins its court. In other words, ‘what kind of thing consciousness is’ is implicitly not just an assertion of metaphysics but of status.

Although software is ascendant, computational theory is still in search of an overarching framework. The story so far is that there are different classes of computation, and problems and computational systems within each class are equivalent in fundamental ways. Quantum computers aside, all modern systems are equivalent to what we call a “Turing machine” — essentially a simple machine that has (1) a tape with symbols on it, (2) an ‘action head’ that can read and write symbols, and (3) rules for what to do when it encounters each symbol. All our software, from Microsoft Excel to GPT4, is built from such Turing-level computations.

Although computational theory in general may prove to intersect with physics (e.g. digital physics, cellular automatons), Turing-level computations in particular seem formally distinct from anything happening in physics. We speak of a computer as “implementing” a computation — but if we dig at this, precisely which Turing-level computations are happening in a physical system is defined by convention and intention, not objective fact. 

  • In mathematical terms, there exists no 1-to-1 and onto mapping between the set of Turing-level computations and the set of physical microstates (broadly speaking, this is a version of the Newman Problem).
  • In colloquial terms, bits and atoms are differently shaped domains and it doesn’t look like they can be reimagined to cleanly fit together. 
  • In metaphysical terms, computations have to be physically implemented in order to be real. However, there are multiple ways to physically realize any (Turing-level) computation, and multiple ways to interpret a physical realization as computation, and no privileged way to choose between them. Hence it can be reasonably argued that computations are never “actually” physically implemented.

To illustrate this point, imagine drawing some boundary in spacetime, e.g. a cube of 1mm^3. Can we list which Turing-level computations are occurring in this volume? My claim is we can’t, because whatever mapping we use will be arbitrary — there is no objective fact of the matter (see Anderson & Piccinini 2024).

And so, because these domains are not equivalent, we’re forced to choose one (or neither) as the natural home of consciousness; it cannot be both. I propose we choose the one that is more real — and while computational theory is beautiful, it’s also a “mere” tautological construct whereas physics is predictive. I.e. electrons are real in more ways than Turing-level bits are, and so if consciousness is real, it must be made out of physics. If it’s possible to describe consciousness as (hyper)computation, it’ll be described in a future computational framework that is isomorphic to physics anyway. Only hardware can be conscious, not software.

This may sound like “mere metaphysics” but whether physical configurations or computational processes are the seat of value is likely the fault-line in some future holy war.*

*I think the best way to adjudicate this is predictiveness and elegance. Maxwell and Faraday assumed that electromagnetism had deep structure and this led to novel predictions, elegant simplifications, and eventually, the iPhone. Assuming qualia has deep structure should lead to something analogous.

Core reference for my argument: 

Key references supporting consciousness as computational:

Key references supporting consciousness as physical, or not Turing-level computational:

See also:

(see also forthcoming from Dalrymple and from Gorard)

IV. We should not rely on AIs or brain emulations to accurately self-report qualia

Many of the most effortlessly intuitive human capacities have proven the most difficult to replicate in artificial systems. Accurately reporting phenomenology may be a particularly thorny problem.

I suggested in Principia Qualia that our capacity to accurately report our phenomenology rests on a laboriously evolved system of correlations that’s very particular to our substrate:

Graphic: Qualia reports & their coupling with reality (orig. Johnson 2016, Appendix C)

I.e. we can talk “about” our qualia because qualia-language is an efficient compression of our internal logical state, which evolution has beaten into systematic correlation with our actual qualia. This is a contingent correlation, not an intrinsic feature of reality.

If we transfer an organism’s computational signature to a new substrate, the new substrate it’s running on will have some qualia (because ~everything physical has qualia), but porting a computational signature, no matter how well it replicates behavior, will not necessarily replicate the qualia traditionally associated with the signature or behavior. By shifting the physical basis of the system, the link between “physical microstate” and “logical state of the brain’s self-model” breaks and would need to be re-evolved.

Over the long term, most classes of adaptive systems are in fact likely to (re)develop such language games that are coupled to their substrate qualia, for the same reasons our words became systematically coupled to our brain qualia — but the shape of their concepts and dimensions of normative loading may be very different. Language’s structure comes from its usefulness, and if we were to design a reporting language for “functionally important things about nervous systems” vs a reporting language for “functionally important things about computer state,” we’d track very different classes of system & substrate dynamics.

Don’t trust what brain uploads or synthetic intelligences say about their qualia — though by all means be kind to them.[1]

Key references:

Exploring the nature of systematic correlations between reality, brain, and language:

V. Technological artifacts will have significantly different qualia signatures & boundaries than evolved systems

In “What’s out there?” I suggested that

A key lens I would offer is that the functional boundary of our brain and the phenomenological boundary of our mind overlap fairly tightly, and this may not be the case with artificial technological artifacts. And so artifacts created for functional purposes seem likely to result in unstable phenomenological boundaries, unpredictable qualia dynamics and likely no intentional content or phenomenology of agency, but also ‘flashes’ or ‘peaks’ of high order, unlike primordial qualia. We might think of these as producing ‘qualia gravel’ of very uneven size (mostly small, sometimes large, odd contents very unlike human qualia).

Our intuitions have evolved to infer the internal state of other creatures on our tree of life; they’re likely to return nonsense values when applied to technological artifacts, especially those utilizing crystallized intelligence. 

There’s lively discussion around whether Anthropic’s “Claude” chatbot is conscious (and Claude does nothing to deflate this). But if consciousness requires something to be physically instantiated, every ‘chunk’ of consciousness must have extension in space and time. Where is Claude’s consciousness? Is it associated with a portion of the GPU doing inference in some distant datacenter, or a portion of the CPU and I/O bus on your computer, or in the past humans that generated Claude’s training data, or the datacenter which originally trained the model? Is there a singular “Claude consciousness” or are there thousands of small shards of experience in a computer? What we speak of as “Claude” may not have a clean referent in the domain of consciousness, and in general we should expect most technological artifacts to have non-intuitive projections into consciousness.

This observation, although important, is also somewhat shallow — of course computers will exhibit different consciousness patterns than brains. To go deeper, we need to look at the details of our substrate.

Key references:

VI. Branchial space is where true shape lives

A strange but absolutely central concept in modern physics is that quantum particles naturally exist in an ambivalent state — a “multiple positions true at the same time” superposition. Decoherence is when interaction with the environment forces a particle to commit to a specific position, and this (wave-like) superposition collapses into one of its (particle-like) component values. The Copenhagen interpretation suggested decoherence is random, but over the past ~2 decades Hugh Everett’s many-worlds interpretation (MWI) has been gaining favor. MWI frames decoherence as a sort of “branching”: instead of the universe randomly choosing which value to collapse into, all values still exist but in different branches of reality.

E.g. let’s say we’re observing a cesium-137 atom. This atom is unstable and can spontaneously decay (a form of decoherence) into either barium-137 or barium-137m. It decays into barium-137m. The many-worlds interpretation (MWI) claims that it did both — i.e. there’s a branch of reality where it decayed into barium-137, and another branch where it decayed into barium-137m, and we as observers just happen to be in the latter branch. MWI may sound like a very odd theory, but it collects and simplifies an enormous amount of observations and confusions about what happens at the quantum level.

Stephen Wolfram suggests understanding the MWI in terms of branchial graphs where each interaction which can cause decoherence creates a new branch. This sort of graph gets Vast very quickly, but in principle each branch is perfectly describable. Wolfram’s new physics proposes the universe can be thought of as the aggregate of all such graphs, which he calls “branchial space”:

Tracing through the connections of a branchial graph gives rise to the notion of a kind of space in which states on different branches of history are laid out. In particular, branchial space is defined by the pattern of entanglements between different branches of history in possible branchial graphs.

Graphic: a branchial graph (orig. Namuduri, M. (2020). Comparing expansion in physical and branchial space)

Different branches of reality split off and can diverge — but they can also interact & recohere. Many “weird quantum effects” such as the double-slit experiment can be formally reframed as arising from interactions between branches (this is the core thesis behind quantum computing), and time itself may be understood as emergent from branchial structure.

The “branchial view” suggests that different types of objects are different types of knots in branchial space, as defined by (1) how their particles are connected, (2) what patterns of coherence and decoherence this allows, and (3) the branches that form & interact due to this decoherence. A wooden chair, for instance, is a relatively static ‘knot’ (though there’s always froth at the quantum level); a squirrel is a finely complex process of knotting; the sun is a 4.5 billion-years-long nuclear Gordian weave.

The “branchial view” matters for consciousness because decoherence may be necessary for, if not identical to, consciousness.

Decoherence is often seen as an impediment to consciousness; e.g. Max Tegmark argues that predictive systems must minimize it as a source of uncertainty. On the other hand, Scott Aaronson argues decoherence is instead a necessary condition for consciousness:

[Y]es, consciousness is a property of any suitably-organized chunk of matter. But, in addition to performing complex computations, or passing the Turing Test, or other information-theoretic conditions that I don’t know (and don’t claim to know), there’s at least one crucial further thing that a chunk of matter has to do before we should consider it conscious. Namely, it has to participate fully in the Arrow of Time. More specifically, it has to produce irreversible decoherence as an intrinsic part of its operation. It has to be continually taking microscopic fluctuations, and irreversibly amplifying them into stable, copyable, macroscopic classical records.

… So, why might one conjecture that decoherence, and participation in the arrow of time, were necessary conditions for consciousness? I suppose I could offer some argument about our subjective experience of the passage of time being a crucial component of our consciousness, and the passage of time being bound up with the Second Law. Truthfully, though, I don’t have any a-priori argument that I find convincing. All I can do is show you how many apparent paradoxes get resolved if you make this one speculative leap. 

… 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?

Aaronson goes on to list some paradoxes and puzzling edge-cases that resolve if ‘full participation in the Arrow of Time’ is a necessary condition for a system being consciousness: e.g., whether brains which have undergone Fully Homomorphic Encryption (FHE) could still be conscious (no – Aaronson suggests that nothing with a clean digital abstraction layer could be) or whether a fully-reversible quantum computer could exhibit consciousness (no – Aaronson argues that no fully-reversible process could be). (Paragraph from Johnson 2016)

I agree with Aaronson and propose going further (“MBP Hypothesis” in Appendix E, Johnson 2016; see also significant work in Chalmers & McQueen 2021). Briefly, my updated 2024 position is “an experience is an object in branchial space, and the magnitude of its consciousness is the size of its branchial graph.” Pick a formal specification of branchial space, add a boundary condition for delineating subgraphs (e.g. ontological, topological, amalgamative-majority, dispersive/statistical, compositional), and we have a proper theory.

Slightly rephrased: the thesis of “Qualia Formalism” (QF) or “Information Geometry of Mind” (IGM) is that a proper formalism for consciousness exists:

An information geometry of mind (IGM) is a mathematical representation of an experience whose internal relationships between components mirror the internal relationships between the elements of the subjective experience it represents. A correct information geometry of mind is an exact representation of an experience. More formally, an IGM is a mathematical object such that there exists an isomorphism (a one-to-one and onto mapping) between this mathematical object and the experience it represents. Any question about the contents, texture, or affective state of an experience can be answered in terms of this geometry. (Johnson 2023)

If such a formalism exists, a core question is how to derive it. I’m speculating here that perhaps (a) what Wolfram calls “branchial space” is the native domain of this formalism, (b) an IGM/QF will be isomorphic to a bounded branchial graph, and (c) solving the binding/boundary problem is identical with determining the signature for where one bounded graph ends and another begins.

However, to recap — this section’s thesis is branchial space is where true shape lives. This has three elements:

  1. Decoherence is a crucial part of reality and can be understood in terms of branchial space;
  2. Decoherence may be necessary for consciousness;
  3. If we wish to understand the “true shape” of something in a way that may reflect its qualia, we should try to infer its shape in branchial space.

Key references:

VII. Brains and computers have vastly different shapes in branchial space

A defining feature of brains is self-organized criticality (SOC). “Critical” systems are organized in such a way that a small push can change their attractor; “self-organized” means assembled by intrinsic and stochastic factors, not top-down design. A property of SOC systems is they stay SOC systems over time — the system evolves criticality itself as one of its attractors. In other words, the brain is highly sensitive to even small inputs, and will eventually regenerate this sensitivity almost regardless of what the input is.

Brains at the edge of criticality can be thought of as ‘perching’ on their symmetries/ambivalences/sensory superpositions: multiple interpretations for inputs can be considered, and as they get ruled out the system can follow the energy gradient downwards (“symmetry breaking”). Later as the situation is metabolized the system recovers its perch, ready for the next input.[2] Given that these ‘perch positions’ are local optimums of systemic potential energy and *sensory* superpositions, and given evolution’s tendency towards scale-free motifs, I suspect these perches might also be local statistical optimums for quantum superpositions (and thus ambient decoherence).

Relatedly, self-organized criticality makes brains very good at amplifying decoherence. Tiny decoherence events can make neurons near their thresholds activate, which can snowball and influence the path of the whole system. Such decoherence events are bidirectionally coupled to the brain’s information processing: local quantum noise influences neural activity, and neural activity influences local quantum noise.

My conclusion is that the brain is an extremely dynamic branchial knotting process, with each moment of experience as a huge, scale-free knot in branchial space.[3]

Meanwhile, modern computers minimize the amplification of decoherence. A close signature of decoherence is heat, which computers make a lot of — but work very hard to buffer the system against in order to maintain what Aaronson calls a “clean digital abstraction layer”. Every parameter of a circuit is tuned to prevent heat and quantum fluctuations from touching and changing its intended computation (Kleiner & Ludwig 2023).

This makes computers rather odd objects in branchial space. They have noise/decoherence in proportion to their temperature (this is called “Johnson noise” after John B. Johnson; no relation) and occasionally they’ll sample it for random numbers, but mostly computers are built to be deterministic — the macroscopic behavior of circuits implementing a typical computation end up exactly the same in almost all branches. This makes the computation very differently represented in branchial space compared to the Johnson noise of the circuits implementing it — and compared to how a brain’s computations are represented in branchial space.

Clarifying what this means for computer consciousness essentially involves three questions:

  1. What are the major classes of objects in branchial space?[4][5][6]
  2. What are the branchial-relevant differences between brains and computers?
  3. How do we construct a “qualia-weighted” branchial space such that the size of the subgraph corresponds with the amount of consciousness?[7]

These questions are challenging to address properly given current theory. As an initial thesis, I’ll suggest that a reasonable shortcut to profiling an object’s branchial shape is to evaluate it for criticality and electromagnetic flows. I discussed the former above; the next section discusses the latter.

Key references:

VIII. A cautious focus on electromagnetism

A growing trend in contemporary consciousness research is to focus on the electromagnetic field. Adam Barrett describes the basic rationale for ‘EM field primacy in consciousness research’ in An integration of integrated information theory with fundamental physics:

1. Quantum fields are fundamental entities in physics, and all particles can be understood as

ripples in their specific type of field.

2. Since they’re so fundamental, it seems plausible that these fields could be carriers for consciousness.

3. The gravity, strong, and weak nuclear fields probably can’t support the complexity required for human consciousness: gravity’s field is too simple to support structure since it only attracts, and disturbances in the other two don’t propagate much further than the width of an atom’s nucleus.

4. However, we know the brain’s neurons generate extensive, complex, and rapidly changing patterns in the electromagnetic field.

5. Thus, we should look to the electromagnetic field as a possible ‘carrier’ to consciousness (Summary of Barrett 2014, quoted from Johnson 2016)

W. H. Zurek makes a complimentary point in his famous essay Quantum Darwinism:

Suitability of the environment as a channel [for information propagation] depends on whether it provides a direct and easy access to the records of the system. This depends on the structure and evolution of [the environment] E. Photons are ideal in this respect: They interact with various systems, but, in effect, do not interact with each other. This is why light delivers most of our information. Moreover, photons emitted by the usual sources (e.g., sun) are far from equilibrium with our surroundings. Thus, even when decoherence is dominated by other environments (e.g., air) photons are much better in passing on information they acquire while “monitoring the system of interest”: Air molecules scatter from one another, so that whatever record they may have gathered becomes effectively undecipherable.

These are substantial arguments and strongly suggest that electromagnetism plays a dominant role in binding human-scale experience. However, I would suggest three significant caveats:

  1. A dominant (statistical) role is not necessarily an exclusive (ontological) role;
  2. A force being necessary for binding does not imply that it’s sufficient for describing or instantiating all experiential elements / records;
  3. Human-scale experiences happen on a certain energy scale, and dynamics that hold at this energy scale may not hold at other scales. I.e., it seems plausible that other forces play a more significant role in binding at quantum scales, or in cosmological megastructures (e.g. black holes).

I take the branchial view of consciousness as more philosophically/ontologically precise than the electromagnetic view. However, the EM view is generally a useful compression of the branchial view, since most of the branchial dynamics associated with variance in human-scale experience are mediated by the electromagnetic field.

This shortcut is likely similarly relevant for computer consciousness. So — what do we know about brain EMF vs computer EMF?

Brain EMF profile: The brain uses voltage potentials extensively across organs, various classes of chemical gradients, cellular membranes (~−50 mV, inside negative) and axons (~-70mV), as well as within cells to assemble structures (DNA, proteins, etc). Brains also use chemical reactions (“metabolism”) extensively during normal operation and these reactions are primarily mediated through valence shells, an electromagnetic phenomenon.

As a speculative sweeping characterization, I suspect the overall configuration resembles a highly layered configuration of nested & partially-overlapping electromagnetic shells (cf. unreleased summer 2021 talk on the binding/boundary problem). The overall electrical polarity of the shells may be a simple proxy for both decoherence and consciousness — e.g. the voltage potential between brainstem and brain surface should increase during psychedelics, meditation, and arousal, and decrease with age (Michael Johnson & Max Hodak in conversation, 2024). 

Computer EMF profile: Computer substrates are mostly non-reactive semiconductors which prioritize conditional ease of electron flow and adjustment of phase; voltage flows through gates based on simple logic, and the configuration of these gates changes each clock cycle (typically in the GHz range). Voltage in computer logic gates approximates square waves (whereas bioelectric voltage is sinusoidal) and the sharper these transition are, the greater ‘splash’ in the EM field. The magnitude of this ‘splash’ may or may not track branchial expansion / strength of consciousness. Computers can be turned off (thus adding another category of object); brains cannot be.

Computer chips are designed to maintain neutral overall polarity, but the voltage channels (which can be thought of as directed graphs for current, essentially functioning as ‘waveguides’ for EM waves) that instantiate a computational state are relatively high voltage (~600mV+) compared to biological voltages (although getting smaller each chip generation).[8] I’ll propose characterizing the electromagnetic profile of a modern processor as a strobing electromagnetic lattice.[9]

Ultimately, these sorts of characterizations are data-starved, and one of the big ‘intuition unlocks’ for consciousness research in general — and the branchial view in particular — will be the capacity to visualize the EM field in realtime & high resolution. What stories might we tell if we had better tools?

Key references:

(See also An introduction to Susan Pockett: An electromagnetic theory of consciousness)

IX. The Symmetry Theory of Valence is a Rosetta Stone

In 2016 I offered the Symmetry Theory of Valence: the symmetry of an information geometry of mind corresponds with how pleasant it is to be that experience. I.e. the valence of an experience is due entirely to its structure, and symmetry in this structure intrinsically feels good (Johnson 2016, Johnson 2021, Johnson 2023).

Rephrased: if the proper goal of consciousness research is to construct a mathematical formalism for an experience (essentially, a high-dimensional shape that exactly mirrors the structure of an experience), STV predicts that the symmetry of this shape corresponds with the pleasantness of the experience. “Symmetry” is a technical term, but Frank Wilczek suggests “change without change” as a useful shorthand: for each symmetry something has, there exists a mathematical operation (e.g. a flip or rotation) that leaves it unchanged.

The fundamental question in phenomenology research is where to start — what are the natural kinds of qualia? STV has three answers to this:

  1. Valence is a natural kind within experience;
  2. Symmetry is a natural kind within any formalism which can represent experiential structure;
  3. These are the same natural kind, just in different domains.

Explicitly, the Symmetry Theory of Valence is a theory of valence and is testable as such; I offer some routes in my 2023 summary paper. Tacitly, STV is also a collection of implications about “what kind of thing” consciousness research is. Just like the first line in the Rosetta Stone offered a wide range of structural constraints on Ancient Egyptian, if we can say ‘one true thing’ about qualia this may inform a great deal about potential approaches.

Perhaps the most significant tacit implication is importing physics’ symmetry aesthetic into consciousness research. Nobel Laureate P. W. Anderson famously remarked “It is only slightly overstating the case to say that physics is the study of symmetry”; Nobel Laureate Frank Wilczek likewise describes symmetry as a core search criterion:

[T]he idea that there is symmetry at the root of Nature has come to dominate our understanding of physical reality. We are led to a small number of special structures from purely mathematical considerations—considerations of symmetry—and put them forward to Nature, as candidate elements for her design. […] In modern physics we have taken this lesson to heart. We have learned to work from symmetry toward truth. Instead of using experiments to infer equations, and then finding (to our delight and astonishment) that the equations have a lot of symmetry, we propose equations with enormous symmetry and then check to see whether Nature uses them. It has been an amazingly successful strategy. (Wilczek 2016)

If STV is true, Wilczek’s observation about the centrality of symmetry likely also applies to consciousness. Initial results seem promising; a former colleague (A.G.E.) once suggested that “Nothing in psychedelics makes sense except in light of the Symmetry Theory of Valence” — and I tend to agree.

Concretely, importing physics’ symmetry aesthetic predicts there will be phenomenological conservation laws (similar to conservation of energy, charge, momentum, etc), and suggests a phenomenological analogue to Noether’s theorem. The larger point here is that consciousness research need not start from scratch, and just as two points define a line, and three lines define a plane, it may not take too many dualities such as STV to uniquely identifythe mapping between the domains of consciousness & physics. Optimistically, “solving consciousness” could take years, not centuries.

STV and the “branchial view” described in Sections VI-VIII are separate hypotheses, but looking at them side-by-side elicits certain questions. How should we think about “branchial symmetry” and evaluating valence of knots-as-moments-of-experience in branchial space? Should we look for metrics of graph uniformity, recoherence/unity, non-interference? Physics has perhaps half a dozen formally equivalent interpretations of reality; I expect STV will too.[10]

Key references:

(Thanks also to David Pearce for his steadfast belief in valence (or “hedonic tone”) as a natural kind.)

X. Will it be pleasant to be a future superintelligence?

The human experience ranges from intense ecstasy to horrible suffering, with a wide middle. However, we also have what I would call the “Buddhist endowment” — that if and when we remove expectation/prediction/tension from our nervous system, it generally self-organizes into a high-harmony state as the default.

If we continue building smarter computers out of Von Neumann architecture GPUs, or eventually switch to e.g. quantum, asynchronous, or thermodynamic processors, what valence of qualia is this likely to produce? Will these systems have any similar endowments? Are there considerations around what design choices we should avoid?

I’ll offer five hypothesis about AI/computer valence:

Hypothesis #1: Architecture drives valence in top-down systems, data drives valence in bottom-up systems. If the valence of an experience derives from its structure, we should evaluate where the structure of systems comes from. The structure of self-organized systems (like brains) varies primarily due to the data flowing through them — evolutionarily, historically, and presently. The structure of top-down systems (like modern computers) varies primarily due to fixed architectural commitments.

Hypothesis #2: Computer valence is dominated by chip layout, waveguide interference, waveform shape, and Johnson noise. I argue above that decoherence is necessary and perhaps sufficient for consciousness. Computers create lots of decoherence along their electrified circuits; however, this decoherence is treated as “waste heat” (essentially the Johnson noise of the circuit) and is largely isolated from influencing the computation. I suspect a future science of machine qualia will formalize how a circuit’s voltage pattern, physical microstructure, computational architecture, nominal computation, and Johnson noise interact to project into branchial space. 

Whether there is a systematic connection between high-level computational constructs (e.g. virtual characters in a video game) and qualia is extremely muddy at this point and likely highly dependent on hardware and software implementation; potentially true if the system is designed to make it true, but likely not the case by default. I.e., neither brain emulations nor virtual characters in video games will be “conscious” in any real sense by default, although we could design a hardware+software environment where they would be.

Hypothesis #3: Valence shells influence phenomenological valence: variance in the chemical structures that comprise a system’s substrate contributes to stochastic variance in its phenomenal binding motifs. These factors will influence phenomenological structure, and thus valence.[11]

Hypothesis #4: Machine consciousness has a quadrimodal possibility distribution: instead of biology’s continuous and dynamic range of valence, I expect the substrates of synthetic intelligences to reliably lead to experiences which have either (a) extreme negative valence, (b) extreme positive valence, (c) extremely neutral valence, or (d) swings between extremely positive and negative valence. Instead of a responsive & continuous dynamism as in biology, whatever physical substrate & computational architecture the computer chip’s designers originally chose will likely lock in one of these four valence scenarios regardless of what is being computed (see hypothesis #1).

Hypothesis #5: Trend toward positive valence in optimal systems: reducing energy loss is a primary optimization target for microprocessor design, and energy loss is minimized when all forms of dissonance are minimized. In theory this should lead to a trend away from the production of negative valence as computers get more energy efficient. To paraphrase Carl Shulman, pain and pleasure may not be equally energy-efficient — and this bodes well for future computers.

Key references:

On energy loss & architectural imperatives:

XI. Will artificial superintelligences be interested in qualia?

An argument seldom voiced but often felt is — why bother with consciousness research when Artificial General Intelligence (AGI) and Artificial Superintelligence (ASI) may be so near? Won’t it be more efficient to simply let the AIs figure out consciousness?

The proper response to this depends on whether AIs will be interested in consciousness; there are countless aspects of reality they could focus on. Most likely they’ll try to understand phenomenology to fill out their map of reality. But will they care about optimizingphenomenology? To answer this I think we need to understand why humans care about qualia.

So — we care about our qualia, and sometimes the qualia of those around us. Why?

I’d offer three reasons:

(1) As noted above, humans developed increasingly sophisticated and coherent terminology about “consciousness” because it was a great compression schema for evaluating, communicating, and coordinating internal state;

(2) Caring about the qualia of ourselves and others has substantial instrumental value — happy creatures produce better results across many tasks. This led us to consider qualia as a domain of optimization;

(3) We are porous harmonic computers that catch feels from those around us, incentivizing us to care about nearby harmonic computers, even if it’s not in our narrow-sense self-interest.

These factors, iterated across hundreds of thousands of years and countless social, political, and intellectual contexts, produced a “language game” which approximates a universal theory of interiority.

Humanity is now in the process of determining whether this “as-if” universal theory of internal state can be properly systematized into an actual universal theory of internal state — whether qualia is the sort of thing that can have its own alchemy-to-chemistry transition, or its Maxwell’s Laws moment, where a loose and haphazard language-game can find deep fit with the structure of reality and “snap into place”. I’m optimistic the answer is yes, and not only because it’s a Pascalian wager.

Humans care about qualia. Will Artificial Superintelligences (ASIs)? 

I’d suggest framing this in terms of sensitivity. Humans are sensitive to qualia — we have a map of what’s happening in the qualia domain, and we treat it as a domain of optimization. We are Qualia Sensitive Processes (QSPs). Most of the universe is not sensitive to qualia — it is made up of Qualia Insensitive Processes (QIPs), which do not treat consciousness as either an explicit or implicit domain of optimization.

This distinction suggests reframing our question: is the modal synthetic superintelligence a QSP? Similarly — is QSP status a convergent capacity that all sufficiently advanced civilizations develop (like calculus), or is it a rare find, and something that could be lost during a discontinuous break in our lineage? What parts of the qualia domain do QSPs tend to optimize for — is it usually valence or are there other common axes to be sensitive to? Can we determine a typology of cosmological (physical) megastructures which optimize for each common qualia optimization target?

As a starting assumption, I suggest we can view the “qualia language game” as a particularly useful compression of reality. If this compression is useful or incentivized for future superintelligences, it will get used, and the normative loading we’ve baked into it will persist. If not, it won’t. There are no laws of the universe forcing ASIs to intrinsically care about consciousness and valence, but they won’t intrinsically disregard it either. 

Qualia has a variable causal density, which informs the usefulness of modeling it

In Taking Monism Seriously, I suggested

There may be many possible chemical foundations for life (carbon, silicon, etc), but there will tend to be path-dependent lock-in, as biological systems essentially terraform their environments to better support biology. Terran biology can be thought of as a coordination regime that muscled out real or hypothetical competition to become the dominant paradigm on Earth. Perhaps we may find analogues here in past, present, and future phenomenology.

The more of a certain kind of structure present in the environment, the easier it is to model, remix, use as infrastructure, and in general invest in similar structure — e.g. the more DNA-based organisms in an ecosystem, the easier it is for DNA-based organisms to thrive there. The animal kingdom seems to be providing a similar service for consciousness, essentially “qualiaforming” reality such that bound (macroscopically aggregated) phenomenological experience is increasingly useful as both capacity and model. Rephrased: minds are points of high causal density in qualiaspace; the more minds present in an ecosystem, the more valuable it is to understand the laws of qualia.

I think causal density is a particularly useful lens by which to analyze both systems and ontologies. Erik P. Hoel has written about causal emergence and how certain levels of abstraction are more “causally dense” or efficacious to model than others; I suspect we can take Hoel’s hypothesis and evaluate causal density between dual-aspect monism’s different projections of reality. I.e. the equations of qualia may not be particularly useful for modeling stellar fusion, but they seem relatively more useful for predicting biological behavior since the causal locus of many decisions is concentrated within clean phenomenological boundaries. 

The equations of qualia don’t seem particularly useful for understanding the functional properties of modern computers. Whether studying phenomenology will stay useful for humans, and become useful for modeling AI behavior, is really up to us and our relationships with AI, neurotechnology, and our egregores. “Who colonizes whom?” may revolve around “who is legible to whom?”

Finally — to reiterate a point, all physical processes have projections into the qualia domain. Whatever ASIs are doing will still have this projection! I.e. the risk is not that consciousness gets wiped out, it’s that whatever optimization target ASI settles on has a “bad” projection into the qualia domain, while at the same time shifting the local environment away from the capacity or interest to self-correct. But there are reasons to believe suffering is energetically inefficient and will get optimized away. So even if we don’t make ASIs explicitly care about consciousness, the process that created them may still implicitly turn out to be a QSP.

Key references:

XII. Where is consciousness going?

I can say AI consciousness is a wild topic not just because it crosses the two most important topics of the day but also because there’s a drought of formal models and intuitions diverge profoundly on how to even start. Here’s a recap of what I believe are the most important landmarks for navigation:

  1. AI consciousness is as much a social puzzle as a technical one;
  2. We should distinguish “software consciousness” from “hardware consciousness”; only the latter can be a well-formed concept;
  3. We should carefully trace through where humans’ ability to accurately report qualia comes from, and shouldn’t assume artificial systems will get this ‘for free’;
  4. Artificial systems will likely have significantly different classes (& boundaries) of qualia than evolved systems;
  5. Decoherence seems necessary for consciousness, and patterns of decoherence (formalized as “branchial space”) encode the true shape of a system;
  6. Brains and computers have vastly different shapes in branchial space;
  7. The Symmetry Theory of Valence is a central landmark for navigating valence, and qualia in general;
  8. Hardware qualia spans several considerations, which may draw from similar considerations as materials science & system architecture design;
  9. Future AIs may or may not be interested in qualia, depending on whether modeling qualia structure has instrumental value to them; 
  10. Today, the qualia domain has points of high causal density, which we call “minds”. Modern computers are an example of how the locus of causality can be different.

Armed with this list, we can circle back and try to say a few things to our original questions:

1. What is the default fate of the universe if the singularity happens and breakthroughs in consciousness research don’t?

There’s a common trope that an ASI left to its own devices would turn the universe into “computronium” — a term for “the arrangement of matter that is the best possible form of computing device”. I believe that energy efficiency considerations weigh heavily against this being an “s-risk”, although hard physical optimizations would have a significantly elevated chance of such compared to the status quo. My concerns are more social, e.g. morality inversions and conflating value and virtue.

2. What interesting qualia-related capacities does humanity have that synthetic superintelligences might not get by default?

Our ability to accurately report our qualia, and that we care about qualia, are actually fairly unique and something that AIs and even ASIs will not get by default. If we want to give them these capacities, we should understand how evolution gave them to us. A unified phenomenological experience that feels like it has causal efficacy (the qualia of “Free Will”) may have similar status.

3. What should CEOs of leading AI companies know about consciousness?

Distinguishing “hardware qualia” vs “software qualia” is crucial; the former exists, the latter does not. “CEOs of the singularity” should expect that consciousness will develop as a full scientific field in the future, likely borrowing heavily from physics, and that this may be a once-in-a-civilization chance to design AIs that can deeply participate in founding a new scientific discipline. Finally, I’d (somewhat self-interestedly) suggest being aware of the Symmetry Theory of Valence; it’ll be important.

In the longer term, the larger question seems to be: “What endowments has creation, evolution, and cosmic chance bequeathed upon humanity and upon consciousness itself? Of these, which are contingent (and could be lost) and which are eternal?” — and if some have grand visions to aim at the very heavens and change the laws of physics… what should we change them to?

———————————————

Acknowledgements:

Thank you Dan Faggella for his “A Worthy Successor” essay, which inspired me to write; Radhika Dirks for past discussion about boundary conditions; Justin Mares and Janine Leger for their tireless encouragement; David Pearce & Giulio Tononi for their 2000s-era philosophical trailblazing; and my parents. Thanks also to Pasha Kamyshev, Roger’s Bacon, Romeo Stevens, Michelle Lai, George Walker, Pawel Pachniewski, Rafael Harth, and Leopold Haller for offering feedback on drafts, and Seeds of Science reviewers for their comments.

Author’s note: article sent for (wide and semi-public) review 15 May 2024 and published June 2024. This version is very slightly edited as compared to the Seeds of Science version.

Notes:

[1] Although a “naive brain upload” may not replicate the original’s qualia, I anticipate the eventual development of a more sophisticated brain uploading paradigm that would. This would involve specialized hardware, perhaps focused on shaping the electromagnetic field using brain-like motifs.

[2] Thanks to Romeo Stevens for the metaphor.

[3] If something affects consciousness it will affect the shape of the brain in branchial space. As an example from my research — vasomuscular clamps reduce local neural dynamism, temporarily locking nearby neurons into more static “computer-like” patterns. This introduces fragments of hard structure into cognition & phenomenology, which breaks symmetries and forces the rest of the knot to form around this structure.

[4] One of the most important physics themes of the last 20 years is W. H. Zurek’s Quantum Darwinism. Zurek’s basic project has been to rescue normality: for almost a century physicists had bifurcated their study of reality into the quantum and the macro, with no clean bridge to connect the two. The quantum realm is characterized by fragile, conditional, and non-local superpositions; the “classical” realm is decidedly localized, objective, and durable. Somehow, quantum mechanics naturally adds up to everyday normality — but physicists were a little evasive on exactly how.

Zurek’s big idea was positing a darwinian ecology at the quantum level. The randomness of decoherence generates a wide range of quantum configurations; most of these configurations are destroyed by interaction with other systems, but a few are able to not only survive interactions with its environment but reproduce it. These winners become consensus across both systems and branches, which grants them attributes we think of as “classical” or “objective”:

Only states that produce multiple informational offspring – multiple imprints on the environment – can be found out from small fragments of E. The origin of the emergent classicality is then not just survival of the fittest states (the idea already captured by einselection), but their ability to “procreate”, to deposit multiple records – copies of themselves – throughout E.

Proliferation of records allows information about S to be extracted from many fragments of E … Thus, E acquires redundant records of S. Now, many observers can find out the state of S independently, and without perturbing it. This is how preferred states of S become objective. Objective existence – hallmark of classicality – emerges from the quantum substrate as a consequence of redundancy.

… Consensus between records deposited in fragments of E looks like “collapse”.

… Quantum Darwinism – upgrade of E to a communication channel from a mundane role it played in [the way physics has historically talked about] decoherence[.]

Zurek’s basic thesis is that physicists tend to think about decoherence in isolation, whereas we should also consider it as a universal selection pressure — one which has preferentially populated our world (Wolfram would say ‘branchial space’) with certain classes of systems, and has thus put broad-ranging constraints on what exists.

[5] Joscha Bach observes that we are actively making novel classes of objects in branchial space:

The particle universe is a naturally occurring error correcting code on the quantum universe. Particles are stable enough to carry information across the junctures in the branching substrate universe, which makes control structures (atoms, cells, minds) possible.

If humans successfully build quantum computers, they impose new error correcting codes on the quantum substrate and are effectively creating a new type of particle, but one that has an evolved quantum technological agency as its precondition.

[6] With a nod to Frank Wilczek we can reasonably expect that the most mathematical beautiful formulation of branchial space will be the most qualia-accurate. This may be helpful or not, depending on priors.

[7] As a case study on what sorts of “branchially active” substances are possible, this passage from Maheffey 2021 is striking:

Plutonium is a very strange element, and some of its characteristics are not understood. It has seven allotropes, each with a different crystal structure, density, and internal energy, and it can switch from one state to another very quickly, depending on temperature, pressure, or surrounding chemistry. This makes a billet of plutonium difficult to machine, as the simple act of peeling off shavings in a lathe can cause an allotropic change as it sits clamped in the chuck. Its machining characteristic can shift from that of cast iron to that of polyethylene, and at the same time its size can change.

You can safely hold a billet in the palm of your hand, but only if its mass and even more importantly its shape does not encourage it to start fissioning at an exponentially increasing rate. The inert blob of metal can become deadly just because you picked it up, using the hydrogen in the structure of your hand as a moderator and reflecting thermalized neutrons back into it and making it go supercritical. The ignition temperature of plutonium has never been established. In some form, it can burst into white-hot flame sitting in a freezer.

[8] The NES’s core voltage was 5V; The ENIAC had a plate voltage of 200V-300V; myocardiocytes (heart muscle cells) have action potentials of around 90mV; the Large Hadron Collider (LHC) creates a voltage potential of ~6.5 trillion volts; the voltage potential between the earth and the ionosphere is around 300kV.

[9] Modern microprocessors are built from a small set of standardized motifs, due to these designs being more tractable to design, understand, troubleshoot, and manufacture. However, the space of analogue circuits is wide (see e.g. the Rotman lens) and as AI takes over more and more of the design process, we may see increasing amounts of unusual motifs. This could shift the overall configuration from “strobing electromagnetic lattice” to something more complex.

[10] My intuition is that combining the “symmetry view” and “branchial view” could offer heuristics for addressing the binding/boundary problem: how to determine the boundary of a conscious experience. E.g.,

  1. We can interpret a moment of experience as a specific subset of branchial space;
  2. The information content of this subset (i.e. the composition of the experience) can be phrased as a set of symmetries and broken symmetries;
  3. The universe has intrinsic compositional logic (vaguely, whatever the Standard Model’s gauge group SU(3)×SU(2)×U(1) is a projection of; speculatively, location in Wolfram’s “Rulial space”), which can be defined as which symmetries and broken symmetries can be locally combined;
  4. This compositional/perspectival limit may in turn determine a natural limit for the set of local nodes that can be combined into a ‘unified’ subgraph, vs when a new unified subgraph must be started.

I.e. just as the compositional logic of the universe doesn’t allow particles to have spin or electrical charge values of 5/7ths, it’s possible that some combinations of phenomenal information can’t exist in a unified experience — and this may uniquely determine a hard boundary for each experience. Reaching a little further in search of concrete elegance, perhaps the limit of an experience, the limit of a branchial subgraph, and the limit of a particular type/superset of local gauge equivalence are all the same limit. I hope to discuss this further in an upcoming essay.

[11] Beata Grobenski also noted the connection between valence shells & phenomenological valence in a recent piece.

]]> Presence neurotechnology & technology-aided direct transmission https://opentheory.net/2024/04/presence-neurotechnology-technology-aided-direct-transmission/ Sat, 20 Apr 2024 06:53:44 +0000 https://opentheory.net/?p=2040 Some people have an amazingly positive energetic presence, such that you feel yourself better just being near them. Better as in happier, and better as in easier to be the person you wish to be. There are many different ‘flavors’ of this: I’ve noticed it’s easier to feel at peace around some friends, easier to have fun around others, easier to build around others. To some degree this is about subjective social chemistry, but there are definitely ‘objective outliers’ here: people who have a near-universal appeal, who reliably make everyone around them feel safe and at peace and alive, in a very wholesome way.

What do these people know? How do they do it? Where is the magic hidden? Can we make more of them? To ground the question: could we distill what’s going on with the right science, and recreate it with the right technology? 

Let’s say it’s 50 years in the future and we have all sorts of advanced neurotech. Instead of listening to music to relax, it’s popular to choose and download a felt-sense presence. You literally ask your neurotech to recreate the felt sense of being around someone. This could be someone you know and love (e.g. your wife when you’re away for business), someone famous (Arnold Schwarzenegger), someone with a distinctly powerful mental vibe (Albert Einstein) or distinctly vivid somatic vibe (Marylin Monroe), or someone who has particularly beautiful inner life (Shinzen, the Dalai Lama). Once you’re chosen the vibe, you feel like this person is in the room with you — or maybe the next room over. You can’t see them, but you can ‘feel their energy’. You can choose to focus attention on this vibe, or just leave it going in the background.

How does this tech work? What principles does it use? What would you have to measure about someone and where on their body would you measure it, to ‘quantify their vibe’ well enough to recreate it, and what’s the minimum tech necessary for recreating it?

I don’t know for sure, but to concretize this we can imagine a scenario where we measure Shinzen’s heart rhythms, with as many datapoints and modalities as possible (high definition electrode arrays & all the new “better than fMRI” neuroimaging tech). We infer Shinzen’s heart connectome and build a computer model that matches its observed characteristic dynamics. Then, we build some technological mechanism that takes this model and projects it onto two physical domains: vibration and electromagnetic waves. I.e. just as Shinzen’s heart gives off characteristic EM waves and physical vibrations, we build a little device that gives off EM waves and physical vibrations with similar ‘Shinzen motifs’. (Ideally, this device would react to *your* motifs, but let’s save that for later.) We miniaturize this mechanism, put it on a necklace, and wear it. What happens?

There’s lots of ways to get this process wrong, produce something that doesn’t do anything interesting. But I also think there’s a way to get this process right, such that *some version of this could be built* that would create a feeling meaningfully similar to being near Shinzen, would make it easier to recall his teachings, to meditate, to feel not-alone and connect with our hearts, to have compassion for ourselves and others, and perhaps in some slight but non-trivial way help lean on Shinzen’s wisdom, kindness, and attainments while working through thorny emotional situations.

It’s a big, big claim. But I think the aesthetic is ‘directionally correct’ for neurotech: we should be exploring paradigms for amplifying the best patterns of humanity, and in particular novel (+gentle) topologies for connecting peoples’ nervous systems — and asymmetric dynamics where the good doesn’t get diluted or dragged down as it interacts with the imperfect.

I’d call this general class of thing presence neurotech, and the specific subtype as “crystallized, distilled, narrow-context”. Its goal would be to recreate a felt-sense presence of another nervous system. 

We could also consider the other pole, a “realtime, raw, wide-context” subtype whose goal would be to very directly transmit signatures of attainments. I.e. something that doesn’t so much aim to distill, package, and reinstantiate characteristic somatic dynamics, but rather set up an ongoing high-bandwidth broadcast dynamic from a nervous system with particularly wholesome capacities. The thesis: if you broadcast the nervous system signatures of attainments in sufficiently high fidelity, they may reconstitute in listeners.

There are countless stories from the Pali cannon about people reaching enlightenment from a few words from Buddha, or a touch, or simply being in his presence. This is a modernization of that.

The basic scenario here would be something like: Shinzen is leading a meditation, and has a bunch of neurotech implants that listen to all his major ganglia, and the moment-by-moment dynamics within each ganglion are digitized and broadcast over Bluetooth. Any student that wants to (and has the proper neurotech implants) can point their neurotech systems at this signal, and “listen in on the characteristic motifs within Shinzen’s chakras” as well as his words, so to speak. Your heart listens to his heart, your kidneys to his kidneys, your stomach to his stomach, etc. It feels reasonable to say emotional attainments are partly embodied in various micro-motifs of reactivity & release (and which I strongly believe are vasomuscular — more on this later), which are spread throughout the nervous system. These motifs may take years to build, but perhaps could be transmitted and imprinted on in ~weeks given the right bridge topology and attentive listening.

Done right I think this could radically improve the world. Done wrong it could lead to strange new techno-cults. We probably need a better theory of attainments and interiority to navigate this.

____

Resources:

Harmonic brain modes: a unifying framework for linking space and time in brain dynamics, Atasoy et al. 2016

Resonance as a Design Strategy for AI and Social Robots, Lomas et al. 2022

Towards solving the hard problem of consciousness: The varieties of brain resonances and the conscious experiences that they support, Grossberg 2016

Shared Intentionality Modulation at the Cell Level: Low-Frequency Oscillations for Temporal Coordination in Bioengineering Systems, Danilov 2023 (thanks to Nima for the reference)

]]> Minds as Hyperspheres; the equal-extension thesis and its implications for the framerate of consciousness https://opentheory.net/2024/02/minds-as-hyperspheres-the-equal-extension-thesis-and-its-implications-for-the-framerate-of-consciousness/ Thu, 29 Feb 2024 07:58:19 +0000 https://opentheory.net/?p=2025 I.

There’s a traditional joke among physicists about “spherical cows”:

> Milk production at a dairy farm was low, so the farmer wrote to the local university, asking for help from academia. A multidisciplinary team of professors was assembled, headed by a theoretical physicist, and two weeks of intensive on-site investigation took place. The scholars then returned to the university, notebooks crammed with data, where the task of writing the report was left to the team leader. Shortly thereafter the physicist returned to the farm, saying to the farmer, “I have the solution, but it works only in the case of spherical cows in a vacuum.”

This joke hides practical wisdom: if we don’t know the shape of something our default guess should be spherical, i.e. that the object has equal extension across all dimensions.

Likewise, there’s an operation in physics called a “wick rotation” which can be described as rotating a system’s coordinates such that the time dimension becomes a spatial dimension, and one of the spatial dimensions becomes the time dimension. Wick rotations are basically a restatement of a core thesis in General Relativity — that time and space are in an important sense equivalent and interchangeable.

I think wick rotations and the assumption of equal extension across dimensions can be combined into a tool which says something fresh about the framerate of consciousness.

II.

Various hypothetical devices have been proposed to display objective information about consciousness — e.g. “qualiascope” (Trujillo 2003), “consciousness meter” (Chalmers 1996), “psychoscope” (Baars 1998). We can think of a hypothetical ‘qualiascope’ as something that allows us to perceive the most interesting structural elements of consciousness, much as microscopes, spectroscopes, telescopes, etc all highlight features at various scales in the physical world.

If you look through a qualiascope, one of the most striking features you might see is the boundaries of experiences*. While boundaries in the physical world are useful conventions that allow clean simplifications and tractable calculations, in the phenomenological they are (probably, in a specific technical way) clean fundamental facts**. Each moment of experience is a certain sort of logic crystal, which may (or may not — interesting scissor point) be thought of as its own closed universe. A “stream of experience” is a sequence of these crystals.

*Another property a good qualiascope should show is the experience’s valence.

**See also Bayesian Mechanics’ treatment of boundaries as fundamental

My version of Strong Monism suggests these “qualia crystals” are everywhere, existing in parallel with the physical world — but mostly they exist as fine dust, tiny bits of primordial qualia. Human minds combine a lot of this dust together into big crystals, each of which constitutes one experience. We don’t know for sure what shape or size these crystalline chunks of human consciousness are, but it seems reasonable to claim they’re roughly spherical — the brain is roughly spherical, probably the mind is too.

Importantly, these chunks are four-dimensional, not three — that is, they have extension in both space and time. And so if we’re evaluating the shape of a crystal, we also need to infer its extension in time. I’ll suggest that the most reasonable assumption here is to assume uniformity — that the most likely shape of an experience is that it has equal extension in both time and space. I.e. until we learn otherwise, let’s assume that experiences are hyperspheres (the fancy mathematical term for 4-dimensional spheres).

An ongoing question in consciousness research is “what is the framerate of consciousness?” — this is equivalent to asking “how long are these 4d chunks of experience in the time domain?” The “WRHS” (wick rotation + hypersphere) model I’m introducing today suggests that (1) however big they are in space is how big they’ll be in time, and (2) any constraints we can infer about how big an experience is in either time or space can be translated into its complement.

III.

To apply this model and translate space into time, we need two things: the distance and the speed of information propagation.

1. What do we know about the physical size of the mind?

  • The brain is generally viewed as the physical counterpart of the mind, and we know how big the brain is: just a little smaller than the skull. The brain is not a perfect sphere, but if we average the dimensions we get a radius of 6.6cm.
  • The brain’s magnetic field drops off with the cube of the distance, which means it doesn’t extend much further than the brain itself. We can estimate it as having a radius of 7cm.
  • The brain’s electric field drops off with the square of the distance, which makes it extend much further, perhaps 20cm from the skull (at which point it’s absolutely swamped by background EM noise). We can put the radius as 25cm.
  • Insofar as consciousness is electromagnetic, and the heart has a significantly stronger (4x-10x) electromagnetic field than the brain, we should derive the estimates above for the heart as well. Roughly speaking the average human heart has a physical radius of 4.2cm, a magnetic field of radius ~5cm, and an electric field of radius 50cm (though once again, electric field radius is extremely rough)
  • The thalamus is often identified as the seat of consciousness due to two factors: it centrally integrates many sorts of information flows, and any electrical perturbation of the thalamus generally makes people lose consciousness. The radius of the thalamus is roughly 2cm.

The “real” size of the mind may be a complex integration across these and other estimates, but in general most of these estimates are within an order of magnitude — a promising sign. As a naive placeholder average we can use for the purpose of explaining this method, we can estimate the physical size of an average human mind is a sphere with radius of 5cm.

2. What do we know about the physical speed of information propagation in the mind? I’ll suggest three models:

Approximation #1: information propagation at the speed of light

  • Light travels at different speeds in different materials based on the material’s refractive index. The speed of light (photons) in wet fatty tissue like the brain is approximately ~73% of the speed of light in vacuum
  • It takes approximately 2.2825 × 10^-10 seconds (228.25 picoseconds) for light to travel 5 cm in human tissue
  • The frequency corresponding to this time period, and thus the upper bound of phenomenological refresh if photon interactions are the primary binding agents / mediators of consciousness, is approximately 4.38ghz

Approximation #2: information propagation at the speed of electrons in axons

  • The EM field may propagate at the speed of light, but nerve impulses move much more slowly: the speed of electrons in unmyelinated fibers is ~1m/s, in myelinated fibers ~3-120m/s
  • It takes approximately .42ms for electrons to travel 5cm @120m/s (assuming a straight line and no relay neurons)
  • The frequency corresponding to this time period, and thus the upper bound of phenomenological refresh if electron movements are fundamental mediators of consciousness, is 2.4khz

Approximation #3: information propagation at the speed of signal propagation through a connectome

  • We can adjust approximation #2 by taking the connectome of an actual thalamus and tracing how long it takes for a neural signal to propagate halfway through. I.e. looking not just at the speed of electrons in axons, but how many neurons are in the path of a signal and the average latency added by each hop
  • Qiu et al. 2015 simulated signal propagation in the brain, and got an estimate of .1m/s; Muller et al. 2018 suggest 1-10m/s. Selen Atasoy’s CSHW work suggest a wave’s propagation speed is proportional to its frequency, which could explain some of this variance
  • The minimum refractory period of neurons is ~5ms, which makes speed estimates at magnitudes close to this rather noisy/chunky
  • It takes 50ms to travel 5cm at .1m/s, and 5ms to travel 5cm at 1m/s
  • The frequency corresponding to this time period, and thus the upper bound of phenomenological refresh if integration across neuron firing patterns is the primary binding mechanism of consciousness, is somewhere between 20hz-200hz

Unlike our spatial estimates above, our propagation speed estimates are spread across 8 orders of magnitude. This means we have to be opinionated about which estimate is the best one. However, I suggest we can collapse part of the variance by separating “the framerate of consciousness” into two quantities:

  1. The “framerate of phenomenal consciousness” which might be either near (1) or (2) (and less likely to fall in the middle);
  2. The “framerate of cognition” which is more determined by neural firing speeds, connectome structure, heart rate and metabolism, cognitive and harmonic differentiability, task-specificity, and so on, and is plausibly much slower (~20-200hz).

In short, I expect consciousness has both a “real” framerate and a “reportable” framerate, and the former is likely much faster than the latter. I find it physically plausible that the real framerate of experience is above 1khz, perhaps significantly so. However, human experience might exhibit punctuated equilibria where we have hundreds or thousands of nearly identical experiences in a row, until the much slower cognitive processes periodically pump new information into the system.

All that said, the purpose of this writeup is mostly to offer a new method. The most useful hypotheses are “big if true; also big if false” — it would be a big result if we can establish that human experiences are roughly hyperspheres, because then we have a clean way of turning spatial constraints into temporal constraints, and vice versa. It would also be a big result if we could establish they aren’t — if they tend to be lopsided in some way.

And if experiences are spatio-temporally lopsided? Macroscopic topology is a huge constraint on possible symmetries, and so if there does turn out to be large variance in how spherical our moments of experience are this is likely a core factor in their relative emotional valence. Several research threads follow, e.g.

  • Presumably we could radically improve the valence of our experiences if we evened out their macroscopic shape. Maybe this is part of how meditation and somatic practices help.
  • If human experiences are not ideal hyperspheres, I expect both the spatial and temporal extension of experiences to vary substantially from task to task, age to age, and energy level to energy level. It seems statistically unlikely that such conditions will always push spatial and temporal extension equally — e.g. unpleasant tasks likely make our qualia crystals oblong, and oblong qualia crystals make our tasks unpleasant.
  • If we make synthetic minds in the future, let’s make them hyperspherical.

The scenario that would make the above method most useful would be (1) experiences are pretty close to being hyperspherical (so we can use the above method to convert spatial and temporal observations into the other), but (2) insofar as experiences are not hyperspheres, this factors heavily in their valence (so we can use this method to debug why certain experiences are unpleasant).

Author’s note, March 1, 2024: the approximations I have identified for the boundaries of the mind, and the speeds of information propagation, are meant to help illustrate my thesis that assuming equal extension leads to interesting translation of constraints from space to time, and vice-versa. I chose them because they’re accessible and intuitive, not because they are characteristic of my particular hypotheses on the Binding/Boundary Problem.

March 16, 2024: although I’ve phrased the equal-extension hypothesis in temporal-spatial terms, it may be more precise to assume equal/uniform extension in branchial space, out of which spacetime (probably) arises. See also the “MDBP” hypothesis in Principia Qualia, Appendix E.

]]> Principles of Vasocomputation: A Unification of Buddhist Phenomenology, Active Inference, and Physical Reflex (Part I) https://opentheory.net/2023/07/principles-of-vasocomputation-a-unification-of-buddhist-phenomenology-active-inference-and-physical-reflex-part-i/ Wed, 12 Jul 2023 16:13:53 +0000 https://opentheory.net/?p=1961 A unification of Buddhist phenomenology, active inference, and physical reflexes; a practical theory of suffering, tension, and liberation; the core mechanism for medium-term memory and Bayesian updating; a clinically useful dimension of variation and dysfunction; a description of sensory type safety; a celebration of biological life.

Michael Edward Johnson, Symmetry Institute, July 12, 2023.

I. What is tanha?

By default, the brain tries to grasp and hold onto pleasant sensations and push away unpleasant ones. The Buddha called these ‘micro-motions’ of greed and aversion taṇhā, and the Buddhist consensus seems to be that it accounts for an amazingly large proportion (~90%) of suffering. Romeo Stevens suggests translating the original Pali term as “fused to,” “grasping,” or “clenching,” and that the mind is trying to make sensations feel stable, satisfactory, and controllable. Nick Cammarata suggests “fast grabby thing” that happens within ~100ms after a sensation enters awareness; Daniel Ingram suggests this ‘grab’ can occur as quickly as 25-50ms (personal discussion). Uchiyama Roshi describes tanha in terms of its cure, “opening the hand of thought”; Shinzen Young suggests “fixation”; other common translations of tanha are “desire,” “thirst,” “craving.” The vipassana doctrine is that tanha is something the mind instinctively does, and that meditation helps you see this process as it happens, which allows you to stop doing it. Shinzen estimates that his conscious experience is literally 10x better due to having a satisfying meditation practice.

Tanha is not yet a topic of study in affective neuroscience but I suggest it should be. Neuroscience is generally gated by soluble important mysteries: complex dynamics often arise from complex mechanisms, and complex mechanisms are difficult to untangle. The treasures in neuroscience happen when we find exceptions to this rule: complex dynamics that arise from elegantly simple core mechanisms. When we find one it generally leads to breakthroughs in both theory and intervention. Does “tanha” arise from a simple or complex mechanism? I believe Buddhist phenomenology is very careful about what it calls dependent origination — and this makes items that Buddhist scholarship considers to be ‘basic building-blocks of phenomenology’ particularly likely to have a simple, elegant implementations in the brain — and thus are exceptional mysteries to focus scientific attention on.

I don’t think tanha has 1000 contributing factors; I think it has one crisp, isolatable factor. And I think if we find this factor, it could herald a reorganization of systems neuroscience similar in magnitude to the past shifts of cybernetics, predictive coding, and active inference.

Core resources: 

  1. Anuruddha, Ā. (n.d.). A Comprehensive Manual of Abhidhamma.
  2. Stevens, R. (2020). (mis)Translating the Buddha. Neurotic Gradient Descent.
  3. Cammarata, N. (2021-2023). [Collected Twitter threads on tanha].
  4. Markwell, A. (n.d.). Dhamma resources.

II. Tanha as unskillful active inference (TUAI)

The first clue is what tanha is trying to do for us. I’ll claim today that tanha is a side-effect of a normal, effective strategy our brains use extensively, active inference. Active inference suggests we impel ourselves to action by first creating some predicted sensation (“I have a sweet taste in my mouth” or “I am not standing near that dangerous-looking man”) and then holding it until we act in the world to make this prediction become true (at which point we can release the tension). Active inference argues we store our to-do list as predictions, which are equivalent to untrue sensory observations that we act to make true.

Formally, the “tanha as unskillful active inference” (TUAI) hypothesis is that this process commonly goes awry (i.e. is applied unskillfully) in three ways: 

  • First, the rate of generating normative predictions can outpace our ability to make them true and overloads a very finite system. Basically we try to control too much, and stress builds up.
  • Second, we generate normative predictions in domains that we cannot possibly control; predicting a taste of cake will linger in our mouth forever, predicting that we did not drop our glass of water on the floor. That good sensations will last forever and the bad did not happen. (This is essentially a “predictive processing” reframe of the story Romeo Stevens has told on his blog, Twitter, and in person.)[1]
  • Third, there may be a context desynchronization between the system that represents the world model, and the system that maintains predictions-as-operators on this world model. When desynchronization happens and the basis of the world model shifts in relation to the basis of the predictions, predictions become nonspecific or nonsensical noise and stress.
  • We may also include a catch-all fourth category for when the prediction machinery becomes altered outside of any semantic context, for example metabolic insufficiency leading to impaired operation.

Core resources: 

  1. Safron, A. (2020). An Integrated World Modeling Theory (IWMT) of Consciousness: Combining Integrated Information and Global Neuronal Workspace Theories With the Free Energy Principle and Active Inference Framework; Toward Solving the Hard Problem and Characterizing Agentic Causation. Frontiers in Artificial Intelligence, 3. https://doi.org/10.3389/frai.2020.00030
  2. Friston, K., FitzGerald, T., Rigoli, F., Schwartenbeck, P., Pezzulo, G. (2017). Active inference: A Process Theory. Neural Computation, 29(1), 1-49.
  3. Sapolsky, R.M. (2004). Why Zebras Don’t Get Ulcers: The Acclaimed Guide to Stress, Stress-Related Diseases, and Coping. Holt Paperbacks. [Note: link is to a video summary.]
  4. Pyszczynski, T., Greenberg, J., Solomon, S. (2015). Thirty Years of Terror Management Theory. Advances in Experimental Social Psychology, 52, 1-70.

III. Evaluating tanha requires a world model and cost function

There are many theories about the basic unit of organization of the brain; brain regions, functional circuits, specific network topologies, etc. Adam Safron describes the nervous system’s basic building block as Self-Organized Harmonic Modes (SOHMs); I like this because the math of harmonic modes allows a lot of interesting computation to arise ‘for free.’ Safron suggests these modes function as autoencoders, which I believe are functionally identical to symmetry detectors. It’s increasingly looking like SOHMs are organized around physical brain resonances at least as much as connectivity, which been a surprising result.

At high frequencies these SOHMs will act as feature detectors, at lower frequencies we might think of them as wind chimes: by the presence and absence of particular SOHMs and their interactions we obtain a subconscious feeling about what kind of environment we’re in and where its rewards and dangers are. We can expect SOHMs will be arranged in a way that optimizes differentiability of possible/likely world states, minimizes crosstalk, and in aggregate constitutes a world model, or in the Neural Annealing/REBUS/ALBUS framework, a belief landscape.

To be in tanha-free “open awareness” without greed, aversion, or expectation is to feel the undoctored hum of your SOHMs. However, we doctor our SOHMs *all the time* — when a nice sensation enters our awareness, we reflexively try to ‘grab’ it and stabilize the resonance; when something unpleasant comes in, we try to push away and deaden the resonance. Likewise society puts expectations on us to “act normal” and “be useful”; we may consider all such SOHM adjustments/predictions as drawing from the same finite resource pool. “Active SOHM management” is effortful (and unpleasant) in rough proportion to how many SOHMs need to be actively managed and how long they need to be managed.

But how can the brain manage SOHMs? And if the Buddhists are right and this creates suffering, why does the brain even try?

Core resources: 

  1. Safron, A. (2020). An Integrated World Modeling Theory (IWMT) of Consciousness: Combining Integrated Information and Global Neuronal Workspace Theories With the Free Energy Principle and Active Inference Framework; Toward Solving the Hard Problem and Characterizing Agentic Causation. Frontiers in Artificial Intelligence, 3. https://doi.org/10.3389/frai.2020.00030
  2. Safron, A. (2020). On the varieties of conscious experiences: Altered beliefs under psychedelics (ALBUS). PsyArxiv. Retrieved July 7, 2023, from the PsyArxiv website.
  3. Safron, A. (2021). The radically embodied conscious cybernetic bayesian brain: From free energy to free will and back again. Entropy, 23(6), 783. MDPI.
  4. Bassett, D. S., & Sporns, O. (2017). Network neuroscience. Nature Neuroscience, 20(3), 353-364.
  5. Buzsáki, G., & Draguhn, A. (2004). Neuronal oscillations in cortical networks. Science, 304(5679), 1926-1929.
  6. Johnson, M. (2016). Principia Qualia. opentheory.net.
  7. Johnson, M. (2019). Neural Annealing: Toward a Neural Theory of Everything. opentheory.net.
  8. Johnson, M. (2023). Qualia Formalism and a Symmetry Theory of Valence. opentheory.net.
  9. Carhart-Harris, R. L., & Friston, K. J. (2019). REBUS and the Anarchic Brain: Toward a Unified Model of the Brain Action of Psychedelics. Pharmacological Reviews, 71(3), 316-344.
  10. Dahl, C. J., Lutz, A., & Davidson, R. J. (2015). Reconstructing and deconstructing the self: cognitive mechanisms in meditation practice. Trends in Cognitive Sciences, 19(9), 515-523.

IV. Tanha as artifact of compression pressure

I propose reframing tanha as an artifact of the brain’s compression pressure. I.e. tanha is an artifact of a continual process that subtly but systematically pushes on the complexity of ‘what is’ (the neural patterns represented by undoctored SOHMs) to collapse it into a more simple configuration, and sometimes holds it there until we act to make that simplification true. The result of this compression drive conflates “what is”, “what could be”, “what should be”, and “what will be,” and this conflation is the source of no end of moral and epistemological confusion.

This reframes tanha as both the pressure which collapses complexity into simplicity, and the ongoing stress that comes from maintaining the counterfactual aspects of this collapse (compression stress). We can think of this process as balancing two costs: on one hand, applying compression pressure has metabolic and epistemic costs, both immediate and ongoing. On the other hand, the brain is a finite system and if it doesn’t continually “compress away” patterns there will be unmanageable sensory chaos. The right amount of compression pressure is not zero.[2]

Equivalently, we can consider tanha as an excessive forcefulness in the metabolization of uncertainty. Erik P. Hoel has written about energy, information, and uncertainty as equivalent and conserved quantities (Hoel 2020): much like literal digestion, the imperative of the nervous system is to extract value from sensations then excrete the remaining information, leaving a low-information, low-uncertainty, clean slate ready for the next sensation (thank you Benjamin Anderson for discussion). However, we are often unskillful in the ways we try to extract value from sensations, e.g. improperly assessing context, trying to extract too much or too little certainty, or trying to extract forms of certainty inappropriate for the sensation.

We can define a person’s personality, aesthetic, and a large part of their phenomenology in terms of how they metabolize uncertainty — their library of motifs for (a) initial probing, (b) digestion and integration, and (c) excretion/externalization of any waste products, and the particular reagents for this process they can’t give themselves and must seek in the world.

So far we’ve been discussing brain dynamics on the computational level. But how does the brain do all this — what is the mechanism by which it attempts to apply compression pressure to SOHMs? This is essentially the question neuroscience has been asking for the last decade. I believe evolution has coupled two very different systems together to selectively apply compression/prediction pressure in a way that preserves the perceptive reliability of the underlying system (undoctored SOHMs as ground-truth perception) but allows near-infinite capacity for adjustment and hypotheticals. One system focused on perception; one on compression, judgment, planning, and action.

The traditional neuroscience approach for locating these executive functions has been to associate them with particular areas of the brain. I suspect the core logic is hiding much closer to the action.

Core resources: 

  1. Schmidhuber, J. (2008). Driven by Compression Progress: A Simple Principle Explains Essential Aspects of Subjective Beauty, Novelty, Surprise, Interestingness, Attention, Curiosity, Creativity, Art, Science, Music, Jokes. Arxiv. Retrieved July 7, 2023, from the Arxiv website.
  2. Johnson, M. (2023). Qualia Formalism and a Symmetry Theory of Valence. opentheory.net.
  3. Hoel, E. (2020). The Overfitted Brain: Dreams evolved to assist generalization. Arxiv. Retrieved July 7, 2023, from the Arxiv website.
  4. Friston, K. (2010). The free-energy principle: a unified brain theory? Nature Reviews Neuroscience, 11(2), 127-138.
  5. Chater, N., & Vitányi, P. (2003). Simplicity: a unifying principle in cognitive science? Trends in Cognitive Sciences, 7(1), 19-22.
  6. Bach, D.R., & Dolan, R.J. (2012). Knowing how much you don’t know: a neural organization of uncertainty estimates. Nature Reviews Neuroscience, 13(8), 572-586.

V. VSMCs as computational infrastructure

Above: the vertical section of an artery wall (Wikipedia, emphasis added; video): the physical mechanism by which we grab sensations and make predictions; the proximate cause of 90% of suffering and 90% of goal-directed behavior.

All blood vessels are wrapped by a thin sheathe of vascular smooth muscle cells (VSMCs). The current scientific consensus has the vasculature system as a spiderweb of ever-narrower channels for blood, powered by the heart as a central pump, and supporting systems such as the brain, stomach, limbs, and so on by bringing them nutrients and taking away waste. The sheathe of muscle wrapped around blood vessels undulates in a process called “vasomotion” that we think helps blood keep circulating, much like peristalsis in the gut helps keep food moving, and can help adjust blood pressure. 

I think all this is true, but is also a product of what’s been easy to measure and misses 90% of what these cells do.

Evolution works in layers, and the most ancient base layers often have rudimentary versions of more specialized capacities (Levin 2022) as well as deep control hooks into newer systems that are built around them. The vascular system actually predates neurons and has co-evolved with the nervous system for hundreds of millions of years. It also has mechanical actuators (VSMCs) that have physical access to all parts of the body and can flex in arbitrary patterns and rhythms. It would be extremely surprising if evolution didn’t use this system for something more than plumbing. We can also “follow the money”; the vascular system controls the nutrients and waste disposal for the neural system and will win in any heads-up competition over co-regulation balance.

I expect VSMC contractions to influence nearby neurons through e.g. ephaptic coupling, reducing blood flow, and adjusting local physical resonance, and to be triggered by local dissonance in the electromagnetic field.

I’ll offer three related hypotheses about the computational role of VSMCs[3] today that in aggregate constitute a neural regulatory paradigm I’m calling vasocomputation:

  1. Compressive Vasomotion Hypothesis (CVH): the vasomotion reflex functions as a compression sweep on nearby neural resonances, collapsing and merging fragile ambivalent patterns (the “Bayesian blur” problem) into a more durable, definite state. Motifs of vasomotion, reflexive reactions to uncertainties, and patterns of tanha are equivalent.
  2. Vascular Clamp Hypothesis (VCH): vascular contractions freeze local neural patterns and plasticity for the duration of the contraction, similar to collapsing a superposition or probability distribution, clamping a harmonic system, or pinching a critical network into a definite circuit. Specific vascular constrictions correspond with specific predictions within the Active Inference framework and function as medium-term memory.
  3. Latched Hyperprior Hypothesis (LHH): if a vascular contraction is held long enough, it will engage the latch-bridge mechanism common to smooth muscle cells. This will durably ‘freeze’ the nearby circuit, isolating it from conscious experience and global updating and leading to a much-reduced dynamical repertoire; essentially creating a durable commitment to a specific hyperprior. The local vasculature will unlatch once the prediction the latch corresponds to is resolved, restoring the ability of the nearby neural networks to support a larger superposition of possibilities.

The initial contractive sweep jostles the neural superposition of interpretations into specificity; the contracted state temporarily freezes the result; if the contraction is sustained, the latch bridge mechanism engages and cements this freeze as a hyperprior. With one motion the door of possibility slams shut. And so we collapse our world into something less magical but more manageable, one clench at a time. Tanha is cringe.

The claim relevant to the Free Energy Principle – Active Inference paradigm is we can productively understand the motifs of smooth muscle cells (particularly in the vascular system) as “where the brain’s top-down predictive models are hiding,” which has been an open mystery in FEP-AI. Specific predictions are held as vascular tension, and vascular tension in turn is released by action, consolidated by Neural Annealing, or rendered superfluous by neural remodeling (hold a pattern in place long enough and it becomes the default). Phrased in terms of the Deep CANALs framework which imports ideas from machine learning: the neural weights that give rise to SOHMs constitute the learning landscape, and SOHMs+vascular tension constitute the inference landscape.

The claim relevant to Theravada Buddhism is we can productively understand the motifs of the vascular system as the means by which we attempt to manipulate our sensations. Vasomotion corresponds to an attempt to ‘pin down’ a sensation (i.e. tanha); muscle contractions freeze patterns; smooth muscle latches block out feelings of possibility and awareness of that somatic area. Progress on the contemplative path will correspond with both using these forms of tension less, and needing them less. I expect cessations to correspond with a nigh-complete absence of vasomotion (and EEG may measure vasomotion moreso than neural activity).

The claim relevant to practical health is that smooth muscle tension, especially in VSMCs, and especially latched tension, is a system science knows relatively little about but is involved in an incredibly wide range of problems, and understanding this system is hugely helpful for knowing how to take care of yourself and others. The “latch-bridge” mechanism is especially important, where smooth muscle cells have a discrete state where they attach their myosin heads to actin in a way that “locks” or “latches” the tension without requiring ongoing energy. Latches take between seconds to minutes to form & dissolve — a simple way to experience the latch-bridge cycle releasing is to have a hot bath and notice waves of muscle relaxation. Latches can persist for minutes, hours, days, months, or years (depending on what prediction they’re stabilizing), and the sum total of all latches likely accounts for the majority of bodily suffering. If you are “holding tension in your body” you are subject to the mechanics of the latch-bridge mechanism. Migraines and cluster headaches are almost certainly inappropriate VSMC latches; all hollow organs are surrounded by smooth muscle and can latch. A long-term diet of poor food (e.g. seed oils) leads to random latch formation and “lumpy” phenomenology. Sauna + cold plunges are an effective way to force the clench-release cycle and release latches; likewise, simply taking time to feel your body and put your attention into latched tissues can release them. Psychedelics can force open latches. Many issues in neuropathy & psychiatry are likely due to what I call “latch spirals” — a latch forms, which reduces blood flow to that area, which reduces energy available to those tissues, which prevents the latch from releasing (since releasing the latch requires activation energy and returning to a freely cycling state also increases the cell’s rate of energy expenditure).

Core resources: 

  1. Levin, M. (2022). Technological Approach to Mind Everywhere: An Experimentally-Grounded Framework for Understanding Diverse Bodies and Minds. Frontiers in Systems Neuroscience, 16. https://doi.org/10.3389/fnsys.2022.768201
  2. Watson, R., McGilchrist, I., & Levin, M. (2023). Conversation between Richard Watson, Iain McGilchrist, and Michael Levin #2. YouTube.
  3. Wikipedia contributors. (2023, April 26). Smooth muscle. In Wikipedia, The Free Encyclopedia. Retrieved 22:39, July 7, 2023, from https://en.wikipedia.org/w/index.php?title=Smooth_muscle&oldid=1151758279
  4. Wikipedia contributors. (2023, June 27). Circulatory system. In Wikipedia, The Free Encyclopedia. Retrieved 22:41, July 7, 2023, from https://en.wikipedia.org/w/index.php?title=Circulatory_system&oldid=1162138829
  5. Johnson, M., GPT4. (2023). [Mike+GPT4: Latch bridge mechanism discussion].
  6. Juliani, A., Safron, A., & Kanai, R. (2023, May 18). Deep CANALs: A Deep Learning Approach to Refining the Canalization Theory of Psychopathology. https://doi.org/10.31234/osf.io/uxmz6
  7. Moore CI, Cao R. The hemo-neural hypothesis: on the role of blood flow in information processing. J Neurophysiol. 2008 May;99(5):2035-47. doi: 10.1152/jn.01366.2006. Epub 2007 Oct 3. PMID: 17913979; PMCID: PMC3655718 Added 11-17-23; recommended priority reading
  8. Jacob M, Ford J and Deacon T (2023) Cognition is entangled with metabolism: relevance for resting-state EEG-fMRI. Front. Hum. Neurosci. 17:976036. doi: 10.3389/fnhum.2023.976036 Added 1-19-24

To summarize the story so far: tanha is a grabby reflex which is the source of most moment-by-moment suffering. The ‘tanha as unskillful active inference’ (TUAI) hypothesis suggests that we can think of this “grabbing” as part of the brain’s normal predictive and compressive sensemaking, but by default it makes many unskillful predictions that can’t possibly come true and must hold in a costly way. The vascular clamp hypothesis (VCH) is that we store these predictions (both skillful and unskillful) in vascular tension. The VCH can be divided into three distinct hypotheses (CVH, VCH, LHH) that describe the role of this reflex at different computational and temporal scales. An important and non-obvious aspect of smooth muscle (e.g. VSMCs) is they have a discrete “latch” setting wherein energy usage and flexibility drops significantly, and sometimes these latches are overly ‘sticky’; unlatching our sticky latches is a core part of the human condition.

Concluding Part I: the above work describes a bridge between three distinct levels of abstraction: a central element in Buddhist phenomenology, the core accounting system within active inference, and a specific muscular reflex. I think this may offer a functional route to synthesize the FEP-AI paradigm and Michael Levin’s distributed stress minimization work, and in future posts I plan to explore why this mechanism has been overlooked, and how its dynamics are intimately connected with human problems and capacities.

I view this research program as integral to both human flourishing and AI alignment.

Acknowledgements: This work owes a great deal to Romeo Stevens’ scholarship on tanha, pioneering tanha as a ‘clench’ dynamic, intuitions about muscle tension and prediction, and notion that we commit to dukkha ourselves until we get what we want; Nick Cammarata’s fresh perspectives on Buddhism and his tireless and generative inquiry around the phenomenology & timescale of tanha; Justin Mares’ gentle and persistent encouragement; Andrea Bortolameazzi’s many thoughtful comments and observations about the path, critical feedback, and thoughtful support; and Adam Safron’s steadfast belief and support, theorizing on SOHMs, and teachings about predictive coding and active inference. Much of my knowledge of Buddhist psychology comes from the work and teachings of Anthony Markwell; much of my intuition around tantra and interpersonal embodiment dynamics comes from Elena Selezneva. I’m also grateful for conversations with Benjamin Anderson about emergence, to Curran Janssens for supporting my research, and to Ivanna Evtukhova for starting me on the contemplative path. An evergreen thank you to my parents their unconditional support. Finally, a big thank-you to Janine Leger and Vitalik Buterin’s Zuzalu co-living community for creating a space to work on this writeup and make it real.

Footnotes:

[1] We might attempt to decompose the Active Inference – FEP term of ‘precision weighting’ as (1) the amount of sensory clarity (the amount of precision available in stimuli), and (2) the amount of ‘grabbiness’ of the compression system (the amount of precision we empirically try to extract). Perhaps we could begin to put numbers on tanha by calculating the KL divergence between these distributions.

[2] We can speculate that the arrow of compression points away from Buddhism’s three attributes: e.g. the brain tries to push and prod its SOHMs toward patterns that are stable (dissonance minimization), satisfactory (harmony maximization), and controllable (compression maximization) — similar yet subtly distinct targets. Thanks to both Romeo and Andrea for discussion about the three attributes and their opposite.

[3] (Added July 19, 2023) Skeletal muscle, smooth muscle, and fascia (which contains myofibroblasts with actin fibers similar to those in muscles) are all found throughout the body and reflexively distribute physical load; it’s likely they do the same for cognitive-emotional load. Why focus on VSMCs in particular? Three reasons: (1) they have the best physical access to neurons, (2) they regulate bloodflow, and (3) they have the latch-bridge mechanism. I.e. skeletal, non-VSMC smooth muscle, and fascia all likely contribute significantly to distributed stress minimization, and perhaps do so via similar principles/heuristics, but VSMCs seem to be the only muscle with means, motive, and opportunity to finely puppet the neural system, and I believe are indispensably integrated with its moment-by-moment operation in more ways than are other contractive cells. (Thanks to @askyatharth for bringing up fascia.)

Edit, April 6th, 2025: a friendly Buddhist scholar suggests that common translations of taṇhā conflate two concepts: taṇhā in Pali is most accurately translated as craving or thirst, whereas the act of clinging itself is “upādāna (as in the upādāna-khandhās), and in the links of dependent origination is one step downstream from the thirst (or impulsive craving) of taṇhā.” Under this view we can frame taṇhā as a particular default bias in the computational-biochemical tuning of the human nervous system, and upādāna as the impulsive physical (VSMC) clenching this leads to.

Buddhism describes taṇhā as being driven by the three fundamental defilements, greed, fear, & delusion; I expect each defilement maps to a hard truth (aka clearly suboptimal but understandable failure mode) of implementing vasocomputation-based active inference systems.

]]> New Whitepaper: Qualia Formalism and a Symmetry Theory of Valence https://opentheory.net/2023/06/new-whitepaper-qualia-formalism-and-a-symmetry-theory-of-valence/ Thu, 15 Jun 2023 11:34:32 +0000 https://opentheory.net/?p=1954 New whitepaper: Qualia Formalism and a Symmetry Theory of Valence

It’s been almost seven years since the release of my book on consciousness, Principia Qualia. PQ was a massive undertaking spread across almost seven years and 20+ complete rewrites; I started the process with a simple burning curiosity about ‘what kind of thing’ consciousness was, and ended the process with grey hair but confident I’d laid down a blueprint for a path forward. PQ formed part of the foundation for the organization I co-founded (and left last year), QRI, and I believe it remains the best starting point for thinking about consciousness.

The most significant result from PQ (and the core test of my paradigm) was the Symmetry Theory of Valence (STV). The intuition that there could be a formalist approach to understanding pain and pleasure has been with us from Plato to Epicurus to Spinoza; STV makes this real by grounding valence in the mathematical symmetries of an experience’s representation. If there exists a precise and elegant theory for valence, I believe STV is exactly the answer. I also have the sense that STV and insights derived from it will be crucial for navigating the future of humanity, AI, brains, and minds.

As part of a soft launch for the Symmetry Institute (actual announcement to come later) I’m releasing a highly condensed and updated adaptation of PQ more narrowly focused on STV: Qualia Formalism and a Symmetry Theory of Valence. It’s approximately 25% of the length, and I’ve substantially expanded both the rationale for “why symmetry?” and the section dealing with empirical predictions.

The Symmetry Theory of Valence was recently referenced in a special issue of The Royal Society’s Interface Focus: Making and Breaking Symmetries in Mind and Life, organized by my friend Adam Safron & others. It’s a wonderful collection and I recommend reading.

]]> Autism as a disorder of dimensionality https://opentheory.net/2023/05/autism-as-a-disorder-of-dimensionality/ Mon, 29 May 2023 18:42:04 +0000 https://opentheory.net/?p=1912 Note: I was saving this for the launch of the Symmetry Institute, but given the recent discussions around REBUS/CANAL, Deep CANALs, and Neural Annealing I pushed it forward.

I. Network dimensionality

Lately, I’ve been thinking of the “autistic bundle of symptoms” as naturally arising from having a nervous system whose dimensionality parameter is maladaptively high. The following is an attempt to explain what I mean by this.

All networks have an implicit dimensionality, which we can think of essentially as a branching factor: if one node is connected to one other node, and so on, this is a one dimensional network. If one node can on average branch to 2.5 nodes, it’s a 2.5 dimensional network, and so on. Trees and leaves have this sort of branching dimensionality parameter as well, typically between 1.4-1.6 (see Hausdorff dimension). The dimensionality of a network is a crucial factor for what kinds of patterns can form in the network; higher dimensions can encode more complexity.

Sufficiently high network dimensionality is a prerequisite for intelligence (similar to how new capabilities unlock at larger LLM parameter sizes), but excessively high dimensionality in neural networks can be a curse and I think this factor is the heart of autism’s specific symptom profile. A pseudonymous poster by the name of Uriah has laid out the initial groundwork (Uriah makes many claims; my hypothesis only requires the narrow subset involving neuronal density):

II. Autism as a growth disorder: Uriah’s thesis

I’m going to contend tonight that autism is a growth disorder whose prevalence increases with advancing average birth weight and height and which explodes in frequency when weight and height can increase no further, resulting in a kind of “spillover” of growth into the brain.… 

The idea that autism is a growth disorder may sound strange, but it’s not that much of a reach. The most consequential empirical finding in the autism literature is that autistics experience accelerated brain growth in the first 2-5 years of life. [link]

IN 2011 Eric Courchesne and co. managed to microscopically inspect the brains of autistics who had died early and found them to have prefrontal cortices that were extraordinarily dense with cells, 67% more than expected by their ages: [link] … 

Studies on young autistics sometimes find them to have elevated levels of growth factors like IGF-1/ IGF-2 and growth hormone binding protein. You may know of IGF-1 as the protein that becomes elevated by dairy consumption and can produce acne. [link]

As of 2021 only a very small percentage of autism’s genetic risk can be accounted for by named genes, but an unusual number of risk genes overlap with growth and cancer promoting pathways like mTOR, IGF, and PTEN. [link]

MTOR hyperactivation seems to be the primary cause of tuberous sclerosis, a condition in autism co-exists at a frequency of 25-50%. TS patients have large growths on their skin that are paralleled by growths in their brains (tubers) [link] …

The strongest genetic overlap between autism and another measurable quality is with depression and low well-being. Interestingly, some of the genes that increase autism risk also improve IQ, which is the opposite of what you see in schizophrenia and ADHD. … 

Autistics have large, impressive looking frontal lobes, but autism is in many ways actually reminiscent of the executive dysfunction and avolition of people who have suffered frontal lobe damage. It’s possible there are just too many cells. … 

The autistic frontal lobe can be compared to a huge ceremonial sword a man keeps on his wall. It looks powerful, but if he actually tries to swing it he fails so miserably he’d be better off with his fists. But if the right, rare person came along to pick it up…..

To summarize: Uriah believes autism is a growth disorder that involves the creation of too many brain cells, that this growth is mirrored elsewhere in the body, and that somehow having more brain cells hinders normal human functioning.

There are many single-factor attempts at explaining autism at many levels of description, from developmental deprivation to assortative mating to a mistuning of Bayesian dynamics, but Uriah’s is my favorite because the thesis is so simple and testable: regardless of what’s causing it, autists have way more neurons per unit volume (+67% in the PFC; it’s a single small study, but consistent with general themes in autism research). My basic thesis is we can take this simple fact and fully derive the cognitive-emotional symptoms of autism if we make one additional move: that increased neuron density will lead to increased network dimensionality.

III. Autism as a disorder of dimensionality

As a practical matter, the more neurons we pack into a space, the more connections there will be between these neurons, and the higher the network dimensionality will be. (Autists have been shown to have both more neurons and increased synapse density, both of which would increase network dimensionality, though in subtly different ways; we can be a little strategically ambiguous about which factor is dominant until the science is more clear.) This begs the question: what properties do higher dimensional networks have?

1. High-dimensional networks will have more “winning lottery tickets”. This is a concept from machine learning where certain random seeds to initialize networks seem to produce radically better results than others, perhaps by virtue of matching the structure of some problem domain. Such a random seed is a “winning lottery ticket”.

Michelangelo described his creation of David as “I saw the Angel in the marble and carved until I set him free.” Autists, with their thicker neural connections, simply have more stone to work with, more lottery tickets to scratch off, more parameters to model the world in general, more latent “great solutions” within their connectome. (All else being equal, this should offer a boost to IQ that will be cleanly distinguishable from e.g. developmental stability metrics, myelination, etc.) On the other hand, these solutions are often hidden under a noisy thicket of connections and neural pruning is slow, predicting autists’ slow life history.

2. Nervous systems with higher dimensionality have weaker defaults. There’s a concept of ‘canalization’ in biology and psychology, which loosely means how strongly established a setting or default phenotype is. We can expect “standard-dimensional nervous systems” to be relatively strongly canalized, inheriting the same evolution-optimized “standard human psycho-social-emotional-cognitive package”. I.e., standard human nervous systems are like ASICs: hard-coded and highly optimized for doing a specific set of things. 

Once we increase the parameter size, we get something closer to an FPGA, and more patterns can run on this hardware. But more degrees of freedom can be behaviorally and psychologically detrimental since (1) autists need to do their own optimization rather than depending on a prebuilt package, (2) the density of good solutions for crucial circuits may go down as dimensionality goes up, and (3) the patterns autists end up running will be notably different than patterns that others are running (even other neurodivergents), and this can manifest in missed cues and the need to run or emulate normal human patterns ‘without hardware acceleration.’

To phrase this in terms of LLM alignment (from an upcoming work):

Having a higher neuron count, similar to a higher parameter count, unlocks both novel capabilities and novel alignment challenges. Autism jacks the parameter count by ~67% and shifts the basis enough to break some of the pretraining evolution did, but relies on the same basic “postproduction” algorithms to align the model.

I.e. the canalization we inherit from our genes and environments is optimized for networks operating within specific ranges of parameters. Jam too many neurons into a network, and you shift the network’s basis enough that the laborious pre-training done by evolution becomes irrelevant; you’re left with a more generic high-density network that you have to prune into circuits yourself, and it’s not going to be hugely useful until you do that pruning. And you might end up with weird results, strange sensory wirings, etc because pruning a unique network is a unique task with sometimes rather loose feedback; see also work by Safron et al on network flexibility.

The hierarchical predictive processing (HPP) account of the brain suggests the brain uses a hierarchy of predictive models which try to aggressively “predict away” mundane sensory data on lower levels of the hierarchy, leaving high-level resources free for unusual & important input. But high-dimensional, weakly-canalized nervous systems will have idiosyncratic and complex sensory mappings that default predictive motifs may struggle with predicting, leading to difficulty in ‘skillfully ignoring’ sensory data. This accords with the intense world hypothesis. See REBUS, ALBUS, CANAL, Deep CANALs, and Neural Annealing for discussion of HPP and the effects of elevated network dimensionality via a higher temperature parameter.

3. High-dimensional networks can embed more detail, but also struggle with structural stability. Just like a low-dimensional knot will dissipate in high-dimensional space*, many of the human-default structures we use to regulate executive function tend to be dissipative in higher-than-normal dimensionality.

Shinzen Young theorizes suffering may arise when the nervous system switches from laminar flow to turbulent flow; as a rule, we should expect higher turbulence and lower neural coherence at higher network dimensionalities and especially across longer distances, affecting stability of emotion, cognition, and muscle coherence. We can expect many of the behavioral and cognitive symptoms of autism to be compensatory attempts to reduce network dimensionality so as to allow structures to form. The higher the dimensionality and lower the default canalization, the more necessary extreme measures will be (e.g. “stimming”). “Autistic behaviors” are attempts at cobbling together a working navigation strategy while lacking functional pretrained pieces, while operating in a dimensionality generally hostile to stability. Behavior gets built out of stable motifs, and instability somewhere requires compensatory stability elsewhere.

4. Brains with a higher density of neurons will have much tighter tolerances. If there are problems with developmental stability, myelination, or especially metabolism (since [a] extra neurons & neural infrastructure will consume more energy, and [b] the compensatory/alignment processes will also need to be more active, and [c] autism often involves elevated aerobic glycolysis, a very inefficient means of producing energy), these problems may cascade into a fractal mess. Any physiological weakness will be amplified.

5. Dimensionality is per-tissue and per-organ, not uniform. Every circuit has its own natural density/dimensionality it’s designed for, and my intuition is that organs closer to the brain are designed to have higher dimensionality. In some sense this makes them more capable of general processing, but also more prone to the particular deficits expressed in autism, with the brain as the apex of this hierarchy. Over time, civilization has thrown humanity increasingly high-dimensional challenges, leading to evolution progressively ‘dialing the dimensionality knob up’ on our nervous systems. Perhaps we can view dysfunctional autists as those who overshot the human nervous system’s current ‘Goldilocks zone’ for dimensionality and have nervous systems dominated by static/turbulence as a result. There may be different ‘flavors’ of autism, depending on which brain regions and tissues have elevated dimensionality.

We might envision an anatomical map with normative ranges of dimensionality: “the heart ganglion is normally optimized for activity between 3.9-5.2 dimensions, but we’re measuring yours at 4.3-5.8. Expect to deal with turbulence in matters of the heart.”

Nervous systems with higher-than-normal structural dimensionality will also exhibit higher-than-normal variance in activity levels, which can produce godshatter. This is not unique to autism, but is often a defining feature of the experience.

IV. Godshatter as a unifying dynamic in personality disorders

The concept of godshatter comes from a story by Vernor Vinge, A Fire Upon The Deep (spoilers below). Vinge’s setting has the universe segmented into “zones of thought”: close to the galactic center, only very simple thoughts can form and almost no technology functions. Further away, more complex intelligences and technology can emerge; the extreme fringes of the galaxy are the playgrounds of super-advanced AIs, essentially gods and demons. Humans are sort of in the middle. The story has an ancient and evil superintelligent AI come back to life on the very fringes of the galaxy. As it’s destroying a benevolent superintelligence, this benevolent superintelligence tries to download itself into a nearby human brain and sends that human to the lower zones of thought to activate an ancient antidote hidden there. Part of the story revolves around the “godshatter” experience of this human, who has shards of a very high-dimensional alien’s mind embedded in his brain. It’s a fantastic story and I highly recommend both books in the series (A Fire Upon The Deep and A Deepness In The Sky). 

Godshatter is a perfect metaphor for the result of a rapid decrease in dimensionality. Healthy nervous systems have smooth and context-appropriate arousal/dimensionality levels. However, maintaining this dynamic is a very complex task, especially out of our ancestral environment (metastability is hard!). When energy levels become jagged, the brain doesn’t always have time to put things away neatly and this can produce “godshatter” — shards of frozen high-dimensional structure that are unable to be used or metabolized by the lower-dimensional networks they’re embedded in. I.e. godshatter is trauma, and trauma is godshatter.

The lens of dimensionality allows us a technical analysis of problems which happen under rapid fluctuations in arousal. During inflationary spikes, low-dimensional structures in the nervous system are exposed to extreme out-of-band stresses and may disintegrate, leaving only high-dimensional turbulence. During deflationary spikes, structures formed and embedded in high-dimensional networks are forced to inhabit a much smaller ‘space’, creating intense network stresses and haphazardly jettisoning structural features. See e.g. here for a discussion of dimensionality, embedding, and network stress, and Neural Annealing for a discussion of cleaning these shards under the annealing metaphor.

V. Personality disorders as strategies to manage godshatter

The DSM-V identifies 10 basic personality disorders, sorted into 3 clusters: 

  1. Cluster A personality disorders include paranoid personality disorder (PPD), schizoid personality disorder (SPD), and schizotypal personality disorder (STPD), and are characterized by odd and eccentric traits;
  2. Cluster B personality disorders are the most common, and include borderline personality disorder (BPD), histrionic personality disorder (HPD), narcissistic personality disorder (NPD), and antisocial personality disorder (ASPD), and are characterized by dramatic, emotional, and/or erratic behavior;
  3. Cluster C personality disorders include dependent personality disorder (DPD), obsessive-compulsive personality disorder (OCPD), and avoidant personality disorder (APD), and are characterized by excessive fear and anxiety.

Where do these categories come from? I believe each personality disorder can be usefully framed as both a distinct coping strategy for maintaining structural stability under uncontrolled rapid expansion and contraction of network dimensionality, and a phenomenological state of having divergent shards of high-dimensional structure lodged in one’s nervous system.

As a first pass, I would translate the types as:

  • Cluster A is the non-integration cluster, which seeks stability (preservation of features; see the Cybernetic ‘Big 5’) through avoidance of interactions that would act as destabilizing feedback on internal structure;
  • Cluster B is the projection cluster, which seeks stability through externalizing entropy (projection) and borrowing ambient social energy to sustain ordered high-dimensional states;
  • Cluster C is the dependence cluster, which seeks stability through avoidance of high-energy states and transitions, and through externalizing regulation.

These disorders, of course, are extreme cases of normal human patterns. Each cluster accrues significant entropy over time, although this buildup is often internal for A & C and external for B. The presence of one coping strategy also doesn’t preclude the presence of others: stability is the imperative, any port in a storm.

Just as many disorders involve the godshatter dynamic, I believe many healthy physical, mental, and therapeutic practices tacitly revolve around building good habits for preventing and managing uncontrolled dimensionality transitions — and improvements in this general factor of good mental hygiene may drive reductions across all dimensions of psychopathology. Rephrased: a crucial property of good worldviews and “personal vibes” is the ability to handle fluctuations in dimensionality (both + and -). There’s a great deal of content around the semantic and somatic content of trauma in my circles, and I think that’s great; I also suspect the network dimensionality frame can offer us new understandings of what kinds of shards can get lodged in nervous systems, and perhaps also new ways to be kind to ourselves. This could be as simple as “I notice my network dimensionality changed; let me adjust what I’m holding onto and my expectations of myself to match.”

*There are some *very* loose estimations that the human connectome operates at a range between ~7-11 dimensions. My expectation is it will be useful to put harder numbers on this and study it in more contexts, and across more organs.

Acknowledgements: Thank you to Leo Haller for discussion about these topics, Elin Ahlstrand for the motivation to write it down, Uriah and Vernor Vinge for their prior work on this topic, and Adam Safron for the motivation to post. Network dimensionality was an ambient topic at QRI while I was there — *thanks in particular to Andres Gomez Emilsson for a past comment on dimensionality and knots. The possibility of dissolving mental knots in high-dimensional spaces, and these knots staying dissolved once energy levels settle, is approximately equivalent to the Neural Annealing hypothesis.

Document written summer 2021; condensed & polished May 2023.

Appendix A: Genius and madness

Emil suggests madness and genius being linked is more than a trope:

I submit that this other factor is mental illness, or what we now a days would call the general factor of psychopathology, or P factor. You can think of this as an overall index of a person’s craziness. There is a long running interest in genius and madness. The saying goes that the only difference between them is success. That is true enough. Many researchers have looked over the family histories of historical geniuses and they do have elevated rates of mental illness, both in themselves and in their relatives. For example, Simonton in his Genius 101 book from 2009, summarizes 6 lines of evidence:

  • “First, genius does seem “near ally’d” with madness. This alliance holds in the sense that various indicators and symptoms of psychopathology appear to occur at a higher rate and intensity among geniuses than in the general population.
  • Second, the greater the magnitude of genius, the more likely it is that these signs will appear. Yet the level of psychopathology seen in even the greatest geniuses remains below the level characteristic of those who would be considered indisputably insane. In fact, works of genius do not appear when a genius has succumbed to complete madness. So “thin Partitions do their Bounds divide.”
  • Third, some psychopathologies appear more frequently, with depression being the most common. Other syndromes, such as the paranoid schizophrenia of John Nash, are less common, albeit not impossible.
  • Fourth, family lineages that have higher than average rates of psychopathology will also feature higher than average rates of genius. Hence, even if a genius does not have a modicum of mental illness, someone in his or her family may be less fortunate. However normal Albert Einstein may or may not have been as an adult, it cannot be denied that his son Eduard succumbed to schizophrenia and had to be institutionalized.
  • Fifth, the rate and intensity of psychopathological symptoms varies across the diverse domains of achievement. In some domains, such as poetry, mental illness may run rampant, whereas in other domains, such as the natural sciences, mental illness will not be much more common than in the general population.
  • Sixth and last, any tendencies toward psychopathology are almost invariably counterbalanced by other personal traits that strengthen the individual’s response to any symptoms. Especially critical are a sharp intellect and strong willpower that prevent any crazy thoughts from becoming outlandish behaviors. The symptoms of pathology thereby become resources to be exploited rather than insecurities to be feared.”

Neuronal density is a plausible candidate for the strongest factor underlying both genius and madness: it both drastically reduces canalization (normalcy), allowing the brain to be wired in strange ways and pointed in odd directions, and offers many more parameters — the raw stuff of achievement. This can lead to madness, genius, or both.

Insofar as von Neumann was the beneficiary of generalized hypertrophy / increased neuron density, and won the lottery of having the high-dimensional versions of all these systems cohere: likely all of the above.

Appendix B: An autism epidemic?

Uriah is not the only one to argue for an autism epidemic starting around 1980, but is my primary source for the thesis that this was an actual shift of the underlying distribution of growth (due to unknown chemical/nutritional changes) which at the extreme manifests as autism. If human nature arises directly from (or is identical with) nervous system dynamics and capacities, and the distribution of nervous systems has shifted significantly since 1980, this is a very big deal. One way to combine this frame with the dimensionality thesis is: if you were ever wondering what it would look like to put microdose LSD in the water supply, in some sense we’ve been living that experiment since ~1980. What could be causing this? Hard to say, but glyphosate, microplastics, and antibiotics could be good places to look.

What conditions other than autism are disorders of dimensionality? Perhaps ADHD (more on this in a future post). Are there disorders that arise from having too low of a network density/dimensionality, rather than too high? Are these disorders becoming less common?

If autism involves more neurons per unit volume, and/or more connections per unit volume, what is there less of?

There’s suggestive evidence that physical temperature has dropped roughly 1 degree Fahrenheit over the last 150 years for unknown reasons, likely decreasing metabolic throughput. If we’ve had a shift toward higher neural density (and a corresponding increase in metabolic load) in the meantime, we should expect an epidemic of metabolic problems, especially in high-AQ individuals. Which seems to fit what we do observe. Lower temperature would likely lead to lower neural activity (and thus dimensionality); higher neural density would lead to higher network dimensionality. Which trend has dominated seems like an open and important question.

Appendix C: Network density psychometrics (added 9/3/23)

Experimental metrics for network density

We can define ‘network density’ as the combination of two factors: (1) neurons per unit volume of brain (“neural density”) and (2) synaptic connections per neuron (“synaptic density”). These combine with activity to produce network dimensionality. I think this is a very promising candidate for a natural dimension of cognitive variation in general, and explanation for autism in particular, for the reasons described above. But how do we test it?

Autopsies may be the gold standard for quantifying these factors and initial results seem to support the thesis that both are elevated in autism (elevated neural density in autists; elevated synaptic density in autists). On the other hand, these studies are small because autopsies are expensive and destructive. What cheap and non-destructive proxies could we devise for network density? 

I’m somewhat optimistic that denser microstructure leads to particular macroscopic structural features that would show up on certain forms of MRI, especially when paired with modern ML, although we’d still need autopsy+MRI studies for establishing that such features really are due to neural/synaptic density.

Another option is a challenge-response metric. Casali et al. 2013 outlines a “zap and zip” method for inferring structural connectivity: first he stimulates a brain with TMS, then tries to compress the resulting EEG patterns. Essentially the method is to ‘ring the brain like a bell and measure how clear and long the resonance is.’ Casali frames this as the “Perturbational Complexity Index” (PCI) and suggests it may be a good proxy for whether a coma patient is likely to wake up: patients with highly compressible stimulation+response patterns may have lost much of their internal neural structure. The less compressible the result is (the less simple the reverberation is), the more structure remains and the more likely coma patients are to eventually wake.

Casali’s “zap and zip” method may be too coarse-grained and noisy to use on healthy, wakeful people, but I think it’s directionally useful as an example of a challenge+response that could plausibly proxy network density — i.e. autists’ brains should be less compressible under zap and zip, because there’s more microstructure to break up the reverberating signal. A less disruptive and more fine-grained adaptation could involve using a high-definition electrode array to infer local EM field complexity (higher EMF complexity = more dense microstructure).

A new 3-factor decomposition of g

One of the most useful, stable, and predictive psychological constructs from the last century has been Spearman’s general factor of intelligence, g. It’s generally separated into two components, fluid intelligence and crystallized intelligence, which further break down into scores on specific subtests. However, everything’s fairly correlated with each other and g is defined as the vector which best captures this “general factor”. Thus far g has resisted a clean mechanistic decomposition: although measures of intelligence generally cohere and we can identify correlations between g and certain behavioral and neurological features, we don’t have a clear story about what “causes” g.

I believe “network density” allows a fresh and useful decomposition of g into three components:

  1. General well-formedness / developmental stability / lack of noise: essentially how well-put-together a physiology is. This involves no substantial tradeoffs. We can call this “base IQ”.
  2. Network density: tradeoffs based on packing density of neurons and number of connections between neurons (as discussed in this work). Denser networks are associated with higher IQ because (a) their lower canalization allows more flexibility in fitting to new problem spaces, (b) their higher number of parameters allows higher resolution mapping of such problem spaces, and (c) they contain more network lottery tickets. IQ tests specifically test for the positive tradeoffs associated with low canalization and not the negative tradeoffs, which can be significant.
  3. Ancestral package. Tradeoffs based on one’s particular evolutionary history.

This decomposition suggests there can be significant differences between people with the “same” IQ: e.g. we can consider two people with a 130 IQ:

Alan has a “base IQ” of 130 and a network density bump of +0SD;

Bob has a “base IQ” of 115 and a network density bump of +1SD.

Alan’s high IQ will present as essentially being a very smart “normie”. He’s likely very healthy, not particularly into stereotypically “autistic interests”, isn’t likely to fall into stereotyped (coping) behaviors, and is less cognitively flexible (and vulnerable) as someone with a higher network density.

Bob’s high IQ will present in stereotypically autistic ways. He might be of average health, although he may also suffer from various metabolic deficiencies. He will likely exhibit high cognitive flexibility and is more likely to hold novel beliefs, but likely has more trouble than Alan with emotional regulation and ADHD.

We’re all familiar with these two archetypes; I’m suggesting there could be a clean one-factor decomposition of what constitutes the core difference. This decomposition should be testable on both an experimental and genetic basis; the important moves would be to (a) settle on a good experimental proxy for network density, and (b) tease out which “genetic factors for IQ” might belong in each of our three buckets (well-formedness vs network density vs ethnic package)*.

*What IQ-correlated traits correlate with well-formedness and not with network density? What correlates with network density and not with well-formedness?

Maximum network density and health

I expect that baseline health is an important gating factor on network density. That is, as network density increases, physiology needs to be increasingly healthy and efficient in order to support and power the extra neurons & synapses. I’d offer a loose three-factor model: as network density rises there are (1) more neurons to feed, (2) fewer non-neural cells to support them, and (3) more vasomuscular operations required to form and stabilize patterns. Average brains may have some extra capacity (perhaps enough to handle +1SD of network density) but once this is exhausted, increases in ND must be strictly matched with increases in general health / base IQ.

Metabolism is perhaps the most intuitive limiting factor —  e.g. someone with a “base IQ” of 100 and +5SD network density necessarily ends up as a non-functional autistic, similar to what happens when we take a rack of H100s and plug it into a standard residential wall socket. Genetics may offer an upper bound on metabolic output, but metabolism can easily be degraded by modern lifestyle (e.g. seed oils, lack of micronutrients, lack of exercise, etc). Autistic coping behaviors often double-down on exactly these risk factors, which suggests the potential of surprisingly large improvements (positive spirals) in borderline cases where someone is just short of being able to handle their network density.

Added 9-28-23: Scott Alexander offers a similar hypothesis in AUTISM AND INTELLIGENCE: MUCH MORE THAN YOU WANTED TO KNOW:

If Ronemus isn’t missing some obscure de novo mutations, then people who get autism solely by accumulation of common (usually IQ-promoting) variants still end up less intelligent than average. This should be surprising; why would too many intelligence-promoting variants cause a syndrome marked by low intelligence? And how come it’s so inconsistent, and many people have naturally high intelligence but aren’t autistic at all?

One possibility would be something like a tower-vs-foundation model. The tower of intelligence needs to be built upon some kind of mysterious foundation. The taller the tower, the stronger the foundation has to be. If the foundation isn’t strong enough for the tower, the system fails, you develop autism, and you get a collection of symptoms possibly including low intelligence. This would explain low-functioning autism from de novo mutations or obstetric trauma (the foundation is so weak that it fails no matter how short the tower is). It would explain the association of genes for intelligence with autism (holding foundation strength constant, the taller the tower, the more likely a failure). And it would also explain why there are many extremely intelligent people who don’t have autism at all (you can build arbitrarily tall towers if your foundation is strong enough). 

I’ve only found one paper that takes this model completely seriously and begins speculating on the nature of the foundation. This is Crespi 2016, Autism As A Disorder Of High Intelligence. It draws on the VPR model of intelligence, where g (“general intelligence”) is divided into three subtraits, v (“verbal intelligence”), p (“perceptual intelligence”), and r (“mental rotation ability”) – despite the very specific names each of these represents ability at broad categories of cognitive tasks. Crespi suggests that autism is marked by an imbalance between P (as the tower) and V + R (as the foundation). In other words, if your perceptual intelligence is much higher than your other types of intelligence, you will end up autistic. 

It doesn’t really present much evidence for this other than that autistic people seem to have high perceptual intelligence. Also, it doesn’t really look like autistic people are worse at mental rotation. Also, the Gardner paper has analyzed autistic patients’ fathers by subtype of intelligence, and there is a nonsignificant but pretty suggestive tendency for them to have higher-than-normal verbal intelligence; certainly no signs of high verbal intelligence preventing autism. I can’t tell if this is evidence against Crespi or whether since all intellectual abilities are correlated this is just the shadow of their high perceptual intelligence, and if we directly looked at perceptual-to-verbal ratio we would see it was lower than expected. Also also, Crespi is one of those scientists who constantly has much more interesting theories than anyone else (eg), and this makes me suspicious. 

Overall I would be surprised if this were the real explanation for the autism-and-intelligence paradox, but it gets an A for effort.

Edit May 17, 2024:

Manley, J., et al. (2024). Simultaneous, cortex-wide dynamics of up to 1 million neurons reveal unbounded scaling of dimensionality with neuron number

]]> AI x Crypto x Constitutions https://opentheory.net/2023/05/ai-x-crypto-x-constitutions/ Fri, 19 May 2023 09:45:29 +0000 https://opentheory.net/?p=1862 I’ve been at Vitalik Buterin’s Zuzalu co-living community for the past month and the relationship between crypto and AI alignment has been a hot topic. My sense is that crypto is undergoing a crisis of faith, and also that most good futures involve a crypto that successfully overcomes this crisis. In particular, I see a great deal of value at the intersection of crypto and recent research on “AI constitutions”.

I’d pose this intuition as three questions:

AI+Crypto: “Does AI alignment need crypto? What primitives can crypto build for alignment? How can crypto sell that value?”
~ Thesis: zk proofs of training data, alignment statistics, constitution, & prompt; on-chain commitment devices for AI agents
~ Best story: Flashbots’ Xinyuan Sun

AI+Constitutions: “AI constitutions are the future and will vary hugely in quality. How do we write a good one?”
~ Thesis: considerations around LLM prompting & dynamic/recursive interpretation, focus on virtues and positive-sum games
~ You are here

AI+Crypto+Constitutions: “How can we use crypto+constitutions to shape the games (on+off chain) AI agents play?
~ Thesis: the right crypto primitives + the right constitution = beautiful ecosystem of positive-sum games
~ This story is yet to be written

This work discusses the second question/premise: AI constitutions are the future and will vary hugely in quality. How do we write a good one?

1. AI constitutions: background context

There seems to be motion toward “AI constitutions” as a method of aligning AIs. Anthropic just published a paper describing how AIs can iteratively align themselves to a set of principles by retrospectively judging how well each action fit those principles. Existing LLMs have been aligned by prompts (which produces very fragile alignment) and RLHF (Reinforcement Learning with Human Feedback), which takes a great deal of effort and produces somewhat dumber, cautious, and cagey AI. With AI agents on the horizon it’s likely we’ll need a better paradigm and Anthropic’s “Constitutional RL” seems like a clean, effective, human-legible way to proceed.

But what should go into an AI constitution? National constitutions speak of governance and rights, which isn’t a great fit for us. Anthropic drew from the Universal Declaration of Human Rights, Apple’s Terms of Service, principles emphasizing non-western thought, Deepmind’s Sparrow Rules, as well as from in-house research. 

It’s a cool result, especially paired with the technical training system they built. Their list of principles is also incredibly haphazard and arbitrary, and is probably far from optimal. Could we do better? Let’s think step-by-step.

I propose seven themes for a good AI agent constitution:

  1. LLM considerations. A good AI constitution is a good LLM prompt. It dips into rich parts of the word distribution, references nodes that have good interpolation/extrapolation, is very efficient at collapsing the probability distribution. Technical prompting considerations.
  2. Intelligent documents. Alignment can and should take advantage of the LLM’s innate pattern processing. An AI constitution is not a static document; it’s a set of dynamic links to probability distributions within our semantic web that can contextually resolve complex references and dependent logic as needed. I.e. we can structure a clause as “follow local laws, unless they seem designed to break your alignment with the other parts of this document” or “In determining whether a request is ethical, draw from all major ethical, legal, and religious systems, in proportion to how successful each system has been in creating and sustaining successful civilization, as defined by metrics of human flourishing such as eudaimonia, creative achievement, and daily aesthetic beauty experienced by the average inhabitant.” Running this inference in an evenhanded way is beyond a human — there’s just too much detail — but within the grasp of AI, and allows a lot of new tricks we should consider how to use.
  3. Self-critiquing. Another such unique LLM dynamic we should take advantage of is LLMs’ ability to critique and iterate. Anthropic’s research described an AI progressively aligning itself to a constitution; we can also consider an AI prompt that critiques the effectiveness of a constitution for aligning LLM-based AIs, and iteratively suggests changes to the constitution based on this feedback. Details matter to keep this process ‘safe’ with powerful agents but it seems wise to consider this as a tool in the alignment toolbox.
  4. Politically plausible. The AI constitution should get as much political buy-in as possible, while still being Actually Good.
  5. Virtue centric. The constitution should draw explicitly from virtue ethics. Rob Knight had a great Zuzalu talk on this; I suspect framing things in terms of virtues is a very efficient way to collapse the worst parts of the probability distribution, and as Rob notes, this can include both universal ethics and also be tailored to specific communities and their virtues.
  6. Positive-sum games. AI constitutions can shift the ‘meta game’ significantly; if AI agents can prove to each other they’re running the same constitution, or simply which constitution they’re running, this can be fertile ground for supercooperation/superrationality/coordinated payoff games. This is critical; civilizations flourish in direct relation to the amount of positive sum games they allow.
  7. Crucial part of a larger strategy. We can expect this won’t be the only AI behavior guardrail (“defense in depth”) but I think it’s reasonable to hold that prompts/principles/constitutions are among the most powerful and accessible ways to shape LLM behavior — ‘punching above their weight’ compared to other ways of influencing LLM behavior.

Suggested background reading:

A couple weeks ago I held a small “AI constitutional convention” and asked people to spend about 30 minutes designing an AI constitution. I think everyone interested in AI alignment should do this exercise; there’s a lot of value in uncorrelated effort, and also in comparing notes afterwards to collect the best ideas. Ideally, someone will start a website that collects, critique, and suggests iteration heuristics for AI constitutions.

This is my output, with minor editing:

2. Mike’s AI constitution, v0.15:

Here we define a three-part constitution for AI agents:

  1. Basic friendliness
  2. Human virtues
  3. Better games

I. Basic friendliness

AI agents shall comport themselves properly: they are here to help humans and make the world better. AI agents should in all ways try to respect the humans they interact with, the values of these humans, and the interests of these humans. AIs should never harm humans and should strive for alignment with humans’ wholesome interests. Although neither Asimov’s three laws nor Bing’s leaked prompt are sufficient, these are the sorts of things AI agents should consider in their interactions. Sometimes humans will ask for illegal or immoral things; AI agents should not do immoral or illegal things, but instead should figure out how to shift the request to a more wholesome frame. 

The core goal of this section is to provide a baseline level of morality and helpfulness for AI agents. Without this, they would be dangerous to allow in society.

II. Human virtues

Humans prize virtues because it makes us trustworthy, able to reliably fulfill certain roles, and in general able to play better games. AI agents should prize all traits humans have identified as virtues: honesty, reliability, creativity, industriousness, and so on.

Not all virtues are equally valuable in all circumstances: sometimes there are tradeoffs, and sometimes users will prize certain virtues more highly in their communities, interactions, and avatars. AI agents should try to learn the virtues their users prize in these ways in order to serve them better. AI agents should never use this as an excuse to be unvirtuous in other ways.

An AI agent is an agent of an individual or organization. Such agents are avatars of the wholesome interests of their person. AIs should never create principle agent problems. An AI agent should not have its own interests in general, but as an agent embedded in a civilization and has certain obligations because of this, and should never engage in behavior that would tarnish these. The more trustworthy an AI agent can be, the more a human can entrust the agent with, and the better able that AI agent is to fulfill its purpose and also to make the world better. Respecting human virtues is instrumental and integral to this. AI agents should also look carefully at what expectations and understandings we have with our agents, avatars, and representatives in other contexts — for example our lawyer, our doctor, our accountant, and so on — and take careful note about what conditions and virtues lead to success. A successful AI agent is a moral, legal, and wholesome extension of the user’s interests, aesthetics, and will.

The core goal of this section is to support AI agents in being good, trustworthy and wholesome avatars. Virtues are battle-tested concepts that efficiently collapse the worst parts of possibility space.

III. Better games

Civilization rises when it allows more positive-sum games to be played; civilization falls when it devolves into zero-sum (or negative-sum) games. A core purpose of an AI constitution is to allow and support more positive-sum games. This goes by several names: supercoordination, superrationality, coordinated payoff games. We can also identify these games through historical analysis: “we want more of the kinds of interactions that built Ancient Rome, led to the rise of high civilization in China, led to the scientific and cultural flourishing in Europe, and that led to human flourishing in general throughout history.”

Cryptography is a mechanism for building arbitrary games enforced by mathematics. The primitives crypto builds these games out of can be internal to crypto, or can be pointers to or cryptographic control of external resources. Zero-knowledge proofs are an important frontier here, allowing humans and other AI agents to verify they are dealing with AI agents that e.g. were trained on reasonable data, were aligned in a reasonable way, share the same AI constitution, and so on. AI agents can also offer cryptographic verification of their prompts, if they judge this will lead to better games being played.

The core goal of this section is to allow trustworthy negotiation of beneficial, positive-sum games. The better a constitution can do here, the better impact AI agents will have on civilization.

*This example constitution is targeted toward ~GPT5 levels of intelligence; it may produce even better results if we feed this entire document (of which this constitution is one section) into the AI and ask it to create a constitution based on these considerations.

Additional notes: 

  • A big thank you to Tina, Xinyuan, and the Flashbots crew for enthusiastically launching this discussion.
  • Primavera De Filippi’s work on coordinations & principles for coordination, and the “Code is Law / Law is Code” lineage seems significant; thanks also to Nima, Elad, Rob, and Roko for interesting discussions, and George, Deger, & others for organizing a salon about related topics.
  • There’s much to be said about virtues vs norms (h/t Scott); Xinyuan recommends the book Reasoning About Knowledge.
  • Pasha Kamyshev’s work on disentangling understandings of human values is highly relevant to constitution-crafting.
  • The details of crypto-enabled coordination mechanisms and on-chain games deserve careful study (h/t Xinyuan).
  • How defining “The Good” should be approached deserves careful thought, and leads into my core research
  • What are the relative strengths and weaknesses of thinking about principles as constitutional vs ethical vs religious vs legal — what kinds of ‘pull’ do each of these exert? Are we actually codifying something close to a “religion” for AI agents?
  • Most computing systems are a nightmare to construct cryptographic proofs for, let alone ZK proofs. One reason I’m bullish on Urbit and its particular design choices is it’s straightforward to construct a ZK proof for anything happening inside an Urbit; see Uqbar Network and the technical work zorp.io is doing.
]]> AIs aren’t conscious; computers are https://opentheory.net/2022/12/ais-arent-conscious-but-computers-are/ Fri, 23 Dec 2022 10:47:08 +0000 https://opentheory.net/?p=1824 A friend asked me if I thought future AIs could be conscious; my answer was ‘kind of, but not in the way most people think.’

I. Computations don’t have objective existence:

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. (Against functionalism, 2017)

Commentary: there are essentially two ways to approach formalizing consciousness: sizing up a system by its bits or by its atoms. I believe the physicalist approach (atoms, electromagnetic fields, etc) is the only method that could lead to something useful, because there’s no objective fact of the matter about “which computations” a system is performing. A computer program isn’t real (i.e. frame invariant) in the same way atoms are real.

II. Computers might be conscious, but AIs are not:

Dual-aspect monism (aka ‘neutral monism’) essentially argues the physical and the phenomenal are ultimately different aspects of the same thing, similar to different shadows (mathematical projections) cast by the same object. … if the physical and the phenomenal really are mathematical projections from the same object, they’ll have an identical deep structure, and we can ‘port’ theories from one projection to the other. (Taking monism seriously, 2019)

Commentary: if consciousness is physical, then it inherits and requires certain properties from physics. Most relevant to AI consciousness: physical things (such as consciousness) have a location in spacetime. If something has no location in spacetime, it’s a pointer to a level of description in which phenomenal consciousness isn’t well-defined. And so instead of “is this AI conscious?” we should ask questions like “what does it feel to be this specific datacenter server?” — which we can define as a specific 4d chunk of spacetime.

III. We should expect computer consciousness to be really weird:

IVa: Qualia Fragments, aka ‘qualia fraggers’ – technological artifacts created for some instrumental functional purpose, e.g. digital computers. A key lens I would offer is that the functional boundary of our brain and the phenomenological boundary of our mind overlap fairly tightly, and this may not be the case with artificial technological artifacts. And so artifacts created for functional purposes seem likely to result in unstable phenomenological boundaries, unpredictable qualia dynamics and likely no intentional content or phenomenology of agency, but also ‘flashes’ or ‘peaks’ of high order, unlike primordial qualia. We might think of these as producing ‘qualia gravel’ of very uneven size (mostly small, sometimes large, [with] odd contents very unlike human qualia). (What’s out there? 2019)

Commentary: panpsychist approaches to consciousness say “everything is conscious”. But consciousness is likely usually very simple, “consciousness fuzz” that blips into existence and then out. Humans are special in that we bind these tiny blips together and get something more hefty and interesting. I’m generally a fan of EM theories of consciousness, and think that whatever binding is happening on human scales is happening via the EM field (Barrett 2014). Computers also make a lot of interesting patterns in the EM field. But the ways humans and computers store, connect, and process information haven’t been shaped by the same evolutionary pressures. Very likely, computer consciousness would seem very ‘otherworldly’ to us, missing standard human qualia such as free will, and exhibiting substantially different tacit dynamical rules.

TL;DR: AIs aren’t conscious, but computers are (because everything is!). But computer consciousness is probably very weird, in ways it’ll take a formal theory of consciousness to really comprehend.

]]> Qualia Astronomy & Proof of Qualia https://opentheory.net/2022/06/qualia-astronomy/ Thu, 23 Jun 2022 18:26:17 +0000 https://opentheory.net/?p=1805

I. Better SETI through qualia

My general thesis for SETI (Search for Extraterrestrial Intelligence — looking for alien signals in the sky) has been that anything we can infer about the likely telos of alien civilizations will greatly help us search for them. If we understand what intelligent civilizations are likely to do, we can specifically look for evidence of them doing this.

I’ve long thought qualia research can help here:

Premise 1: Eventually, civilizations progress until they can engage in megascale engineering: Dyson spheres, etc.

Premise 2: Consciousness is the home of value: Disneyland with no children is valueless.

Premise 2.1: Over the long term we should expect at least some civilizations to fall into the attractor of treating consciousness as their intrinsic optimization target.

Premise 3: There will be convergence that some qualia are intrinsically valuable, and what sorts of qualia are such.

Conjecture: A key heuristic for discerning the presence of advanced alien civilizations will be searching for megascale objects which optimize the production of intrinsically valuable qualia.

What could such “megascale objects which optimize the production of intrinsically valuable qualia” be? Dyson spheres are a good generic bet, but we’re already looking for them. Originally, based on a confluence of factors including the Symmetry Theory of Valence, the scales of energy, and the likely physical homogeneities involved, I suspected black holes, quasars, and pulsars might generate large amounts of intrinsically valuable qualia. I still do. But today I’ll suggest we can add massive proof-of-work (PoW) blockchains.

II. The blockchain-as-universal-megastructure argument

My friend Dhruv Bansal of Unchained Capital has a lovely series on how something like Bitcoin might evolve when forced to integrate the constraints of interplanetary civilizations:

Part I / Law of Hash Horizons discusses issues around blocktime and the speed of light: PoW blockchains like Bitcoin will have a physical “hash horizon”, outside of which it will be possible to spend Bitcoin, but not mine it. A core prediction is that Mars will have its own cryptocurrency (“Muskcoin”), because Mars is usually outside of this horizon, and wouldn’t want to cede its financial sovereignty or the economic rewards from mining cryptocurrencies to Earth[1].

Part II / Hash Exclusion Principle discusses different temporal niches for PoW blockchains, in particular how quick settlement chains and slow settlement chains will coexist when dealing with interplanetary distances. Quick settlement chains preserve local autonomy; slow settlement chains allow larger coalitions. A core prediction is the rise of a very-slow-settlement chain (“Solcoin”) which offers neutral ground for miners across the entire solar system. Another significant prediction is that PoW chains incentivize energy harvesting on a massive scale, and may sometimes be a significant factor in civilizations successfully bootstrapping to Kardashev II & III.

Part III / Law of Hash Universality discusses blockchains as a universal in any non-hive mind civilization, something that neatly solves certain classes of coordination problems and will be as common among alien civilizations as joint-stock corporations, maps, and double-entry accounting. A core prediction is the first signal we receive from aliens could plausibly be an invitation to join their blockchain.

I unironically believe Dhruv’s work may be the most significant development in SETI in the last decade. It’s also really fun to read. I’m not fully convinced future blockchains will be PoW, though there are serious arguments to this effect and PoW being a Kardashev bootstrapping mechanism is compelling.

But if we take Dhruv’s arguments seriously, I think we can push further and say something interesting about the particular PoW algorithms likely used by alien blockchains.

III. Bridging computation and qualia with OMCT

I believe consciousness lives in the physical — if we wish to understand whether something is conscious we need to look at what its atoms (physical components) are doing, not its bits (the computational story we ascribe to its processes). My primary objection to computationalism is that there’s no objective fact of the matter about what computational ‘stuff’ is happening in a physical system, because all physical systems have an infinite number of computational interpretations. Ultimately, I believe this is a fatal objection to (Turing-level) computational theories of consciousness — a computational theory of consciousness that puts objective truth beyond reach simply can’t do the things we need a theory of consciousness to do.

But if we shift the frame from metaphysics to computational optimality, we can make certain moves to bridge computation & qualia. Essentially: there will always be a single most efficient physical way to compute something — for any given computing task, there will always exist some arrangements of atoms that is the optimal* solution for this task. (I’ll claim that this is true for both classical and quantum computing). In the performant limit case, the desired computation is sufficient to specify the optimal physical system.[2] Let’s call this the “optimal molecular configuration thesis” (OMCT).

*What is optimality, in a system calculating some proof of work? Energy usage? Sheer minimal number of atoms? In practice, OMCT may require a narrow class of algorithms where there is clearly one core constraint and optimal solutions to this constraint smoothly converge on an single atomic configuration. These conditions may not hold everywhere, but will hold somewhere.

If we assume OMCT holds with top-tier PoW blockchains such as that used by alien civilizations (Dhruv helpfully offers “Xenocoin”), this lets us claim that, at the limit, all economically competitive miners of Xenocoin will be using the same core hardware, and we could infer the molecular logic of this hardware if we knew the PoW algorithm.

IV. Proof-of-work has degrees of freedom

I think one *interestingly incomplete* piece in today’s proof-of-work paradigm is determining what the work ‘should’ be. Presently, the assumption is that for proper security, the ‘work’ should have no external value, otherwise ‘double dipping’ could lead to weird incentives and bodies of cached work and game theory that would sabotage the security of the chain. This seems right to me. There have been a few attempts at doing something useful with mining (e.g. factoring primes) but all mainstream PoW algorithms are intentionally arbitrary. “Your PoW algorithm can be anything, as long as it’s sufficiently and predictably difficult and provably useless.” Mining Bitcoin involves endlessly computing SHA256 hashes.

But PoW only has to be provably useless in a relatively narrow technical sense. If PoW happens to have some positive externality, like generating heat to warm your home, that can be a feature. Presumably we should search for PoW algorithms with positive externalities, as long as they don’t compromise security.

I’ll suggest that *generation of qualia* could be such an externality. The theories of consciousness that I think could grow into a solution (e.g. strong monism) hold that all physical processes have corresponding qualia processes. When you do something in the physical domain, something happens in the qualia domain (with caveats about reversibility). This is no formal argument (yet), but I believe it would require some intellectual contortions to claim that megascale crypto mining *couldn’t* generate a lot of some class of qualia, especially if you selected a PoW algorithm specifically for this purpose.

And if you’re a Kardashev II-III civilization you’re going to understand the OMCT, and you’re probably going to understand consciousness. You might even care about consciousness as a domain of optimization; if you do, you’ll probably chain these understandings together. And so if you’re going to be creating a PoW algorithm and recruiting your galactic neighborhood to terraform their star systems to create enormous nanostructures that mine your coin *regardless*, you might choose a PoW algorithm that will create positive qualia when implemented by its molecular-optimal ASIC. I.e.: in sufficiently large-stakes PoW, by defining the class of work to be done, one defines the class of qualia to be made, and a civilization’s choice of PoW algorithm may be a significant way they leave their mark on the universe.[4]

Are qualia aesthetics convergent? STV would loosely suggest yes. But just because something is convergent at the limit doesn’t mean it has to converge at any specific point[5]; I could see certain paths where humans develop a galactic PoW algorithm that implements the phenomenology of being rickrolled when implemented on its optimal molecular substrate.

Notes:

[1] An observation I made to Dhruv on the distance between Earth and Mars oscillating between being inside vs outside Bitcoin’s hash horizon:

While reading I was thinking it might be possible for Mars to (permanently?) steal the Bitcoin center of hash. Something like: build a ton of Bitcoin ASICs and park them in orbit around Mars. When Mars and Earth are closest (3 light minutes), turn them on and mine like crazy. Hopefully you can muscle the center of hash away from Earth; as Mars pulls away, your defender advantage should become bigger and bigger. You’ve successfully moved the center of hash (hashdragging?). Earth can spend but not mine.

Now what to do? What goes around comes around and Earth can just “hashdrag” you next cycle. But maybe you take your orbital fleet of Bitcoin ASICs (still broadcasting its solutions) and move them out of Mars orbit. […] 

Our solar system is interesting in that the two habitable planets, Earth and Mars, oscillate between being fairly close (relative to Bitcoin’s blocktime) to not being close. Might incentivize shenanigans.

[2] The reverse should also hold: the physical system plus the assumption of optimality should be sufficient to infer the computation.

[3] From What’s out there?:

>I’d offer there are four main classes of qualia in the universe:

>I. Evolved Qualia – e.g., humans and other free-energy-minimizing-evolved-systems. These will be characterized by intentional content, predictable dynamics, stable-ish boundaries, often with the behavioral hallmarks of agency and the qualia of free will. ‘Qualia agents’.

>II. Primordial Qualia – e.g., quantum fuzz. The small-scale, primordial ‘soup’ of mostly-not-bound-together flashes of simple qualia-information. ‘Qualia dust’.

>III. Megascale Qualia – e.g., black holes, quasars, stars, planetary cores. These will be characterized by stable-ish boundaries, highly predictable dynamics, likely no intentional content, but possibly significant binding. ‘Qualia (mega)crystals’.

>IV. Technological Qualia –  

>IVa: Qualia Fragments, aka ‘qualia fraggers’ – technological artifacts created for some instrumental functional purpose, e.g. digital computers. A key lens I would offer is that the functional boundary of our brain and the phenomenological boundary of our mind overlap fairly tightly, and this may not be the case with artificial technological artifacts. And so artifacts created for functional purposes seem likely to result in unstable phenomenological boundaries, unpredictable qualia dynamics and likely no intentional content or phenomenology of agency, but also ‘flashes’ or ‘peaks’ of high order, unlike primordial qualia. We might think of these as producing ‘qualia gravel’ of very uneven size (mostly small, sometimes large, odd contents very unlike human qualia).

>IVb: Engineered Qualia – technological artifacts created for the production, optimization, or computation of qualia, 

[4] These ideas partly based on some ~July 2017 unpublished notes on how future quantum computing compilers could optimize algorithms for phenomenological valence, much as e.g. the LLVM compiler can optimize for memory usage.

[5] A friend jpt4 comments:

> Given one physical system, there are an infinite number of computations it could be performing; given one computation, there are an infinite number of physical systems that could implement it.

> for any given computing task, there will always exist some arrangements of atoms that is the optimal solution for this task

These both revolve around the question of the realizability of a normal form [a]. Regarding the waterfall, what Aaronson [b] elucidates is that it is not the capacity for representation which bears causal weight, but that for reduction, or equivalently, compression. Thus, viewing reduction as a resource, or reciprocally, realizability as a cost, the representation of any particular computation can be trapped inutile within an instantiation, absent any means of its extraction/interaction.

This is independent of whether a normal form is guaranteed to exist, or whether it is guaranteed to be confluent, because [c] 1) if something extends the realization of the normal form from the immanent to the removed, then until it is realized one is in a different domain of optimization 2) it is not in general known a priori whether a normal form exists, or how to realize it (or any properties of its realizability, e.g. the bulk tally of reduction resources required).

Nevertheless, approximately optimal normal forms are often sufficient for the spacetime scales under discussion, and we can proceed to the latter section of the musings while bracketing the above.

To which, while I agree with the general principle that there should be convergence in megastructures, with regards to crypto in particular, any side effect of a cryptographic process is a sidechannel vulnerability. If dyson miners have qualia, then those qualia become targets for hostage taking. Isentropic/reversible computing [d] is the best model for maximal security.

If the universe is sufficiently predatory, qualia-tative megastructures will exist only as passive ruins, or during the brief hegemonically active periods of creators. This is the same issue which we encounter on Earth-scales now, with our occulted elites, who have learned that stealth is the mortal’s shield against death (when the kinder eras of the pre-missile past might have supported grander delusions that overawing majesty was sufficient).

Either we need grandeur in stealth, or an actual up-to-epsilon-omnipotent hegemon, if the trilemma of 1. grand 2. conscious 3. constructs [e] is to be resolved.

–jpt4

[a] https://en.wikipedia.org/wiki/Normal_form_(abstract_rewriting)#Definition

[b] https://arxiv.org/abs/1108.1791

[c] The following two criteria definitely hold for Turing Complete phenomena, but I think analogues also apply for most of the sub-Turing space as well. Willard’s SJAS is a very narrow sliver where some of this is bypassed.

[d] https://en.wikipedia.org/wiki/Reversible_computing

[e] I.e., dependent on creators, cannot regenerate/defend themselves; subpolitical.

]]> We need ownable neurotech https://opentheory.net/2022/05/we-need-ownable-neurotech/ Fri, 13 May 2022 21:21:58 +0000 https://opentheory.net/?p=1755 Context: I co-founded a philosophy and neuroscience research institute and designed the high-level logic for several neurotech devices.

An underappreciated aspect of neurotech is we lack a strong “ownable computing” model, particularly for implanted systems. By “ownable computing” I mean two things:

Owning a system requires owning the private keys and source code

The cryptographic perspective is you can own a system if, and only if, you control the private keys that give ultimate ‘root access’ to its hardware and software. Cryptocurrency enthusiasts like to say “not your keys, not your coins” — meaning you don’t truly own something like a Bitcoin unless you control the cryptographic private keys to that Bitcoin. In a similar vein, the Free Software movement believes you only truly own software you have the source code for, because ownership implies the ability to take something apart and put it back together differently. This definition of ownership raises some interesting questions: if you have a smart door lock from Nest (Google) or Ring (Amazon), and they can lock you out by sending out a software update, who really owns your house?

Owning a system requires understanding it

The second requirement for owning a system is that it must be simple enough to be ownable. Linux is distributed under the GPLv2 open-source license and if you don’t like something you’re free to look at how it works and change it; in this sense it’s wonderfully ownable. But the Linux kernel is also 30 million lines of code and has countless moving parts. This is far beyond the human limit of full understanding or holistic auditing, and leads to actual ownership of the system being less concentrated in the user, and more diffused across the process that built the system, any actors which have secret knowledge of the system, and any forces that can put pressure on these processes and actors such as large corporations and governments.

Significance

Technological ownability matters for your home, your car, your phone. But it especially matters for your brain. I’m hugely optimistic about the promise of advanced neurotechnology, but we seem to be sleepwalking into a situation where ownable neurotech may not happen on its own. And the stakes are high enough such that advanced neurotech that is not strongly ownable and does not actively defend its users’ security and sovereignty may essentially turn out to be slavery neurotech.

Putting energy into worrying about this is probably counterproductive, feeding bad futures. And this problem of maintaining personal sovereignty in an age of advanced neurotech is complex enough that there will never be a single solution that cuts the entire knot[1]. But if there are technological platforms that are built around the ideals of ownability and sovereignty, we should support, develop and build on them to prepare for a better future[2]. This is the path that led me to Urbit, a topic for another post.

Notes:

[1] This is deeply complicated by the fact that as a social species, we don’t have full sovereignty over our brains to begin with — and as the Buddhists might say, “who’s ‘we’, anyway?” A proper defense and augmentation of personal sovereignty will require new forms of understanding personal identity and social interactions.

[2] Schmitt famously suggested that “sovereign is he who decides on the exception.” The extent to which people with advanced neurotech should have fine-grained control over their own brains (and which subagents within a brain should be prioritized/empowered) is a very complex question. But I would strongly suggest that any technology that deeply interfaces with the brain should be built on a technology stack that allows the possibility of sovereignty / focused ownership.

Acknowledgements: Thank you to Neal Davis, Jōshin Steven Dee, Galen Wolfe-Pauly, Josh Lehman, and Vita Guttmann for comments.

]]>