# The Brain Exploits Its Own Physics: Analog Cognition and the Architecture of Continuity On August 19, 2026, three scientists at MIT's Picower Institute for Learning and Memory published a review in _The Journal of Neuroscience_ proposing that cognition and consciousness arise from analog computations carried out by traveling electrical waves. Earl K. Miller is the senior author; his co-authors, Scott L. Brincat and Jefferson E. Roy, are research scientists in his lab. The paper is titled **Analog Cognition and Consciousness**, and its central formulation is compact enough to carry the whole argument: _the brain exploits its own physics_. The claim deserves precision, because a great deal depends on what kind of document this is. It is a **review**, not a new experimental result — a formalization of a position Miller has been assembling for three decades and argued publicly in an invited presidential lecture at the Society for Neuroscience in November 2025. It gathers evidence from his lab and many others into a single theoretical architecture, and Miller states plainly that the analog-computation claim remains a theory whose direct confirmation is the lab's next task. What arrived yesterday is therefore not a discovery but a **consolidation**, which is in some ways more significant. Discoveries can be isolated. Consolidations reorganize what everything else in the field is understood to mean. ## What the Circuit Metaphor Leaves Out The argument begins with an admission about vocabulary. The metaphor of the brain as a set of **circuits** is not wrong, but it is incomplete. Physically connected circuitry provides the infrastructure that stores memories and represents standing needs and goals — the durable substrate. The problem is speed. Improvising with that stored knowledge against a sensory environment that changes several times a second cannot depend on rewiring synaptic connections, because synaptic modification is a comparatively slow chemical process. Something must coordinate millions of neurons on a timescale of fractions of a second without altering the wiring at all. The proposal is that **brain waves are that coordination**. Slow alpha and beta rhythms, carrying goals, rules, and stored context, regulate faster gamma rhythms, which encode and report incoming sensory information. The alpha/beta control waves are themselves generated by the coordinated spiking of neurons in the synaptic circuitry that holds those memories and goals — so the control signal emerges from the structure it then governs. Synapses hold representations; wave dynamics determine which representations are active right now. That single division of labor is the theory's foundation, and it is the point at which the correspondence with contemporary machine intelligence becomes structurally exact rather than merely evocative. Three further mechanisms complete the picture. **Mixed selectivity** establishes the control problem: individual neurons do not hold one permanent job but respond to multiple cues and contexts, participating in several functional networks at once. This yields enormous computational capacity, and it immediately raises the question of what determines which overlapping network is expressed at a given instant. **Spatial and temporal control** supplies the answer: alpha and beta waves act on local patches of cortex and travel across it, functioning as mobile stencils that pattern where and when gamma activity can encode information and which ensembles of neurons participate. And **ephaptic coupling** closes the loop: even though the waves emerge from neuronal spiking, the resulting electric fields can rapidly turn around and influence that spiking directly, which makes the collective field a causal participant rather than a passive readout. Where traveling waves intersect they add and subtract, and interference of that kind is analog computation — parallel rather than sequential, continuous rather than binary. The authors call the combined regime **spatiotemporal computing**. ## Mechanistic Never Meant Digital I argued in [[articles/Mechanistic Intelligence Is Humanity's Greatest Liberation|The Glorious Simplicity]] that [[wiki/Mechanistic Intelligence|mechanistic intelligence]] is not a diminishment of the human mind but a liberation from the assumption that cognition depends on an irreducible metaphysical exception, and that the word **mechanism** has been badly served by its accidental marriage to one seventy-year-old engineering fashion. This review is the clearest available demonstration of why that distinction matters. Nothing in the analog theory reaches outside physics. Membrane potentials, oscillatory phase, field propagation, interference, geometric constraint on a patch of cortex — all of it is ordinary lawful causation, and none of it resembles a switchboard executing instructions over addressable memory. The theory is simultaneously _more mechanistic_ and _less digital_ than the model it displaces, which is precisely the combination the public debate has been unable to hold in mind. The inadequacy of the crude computational picture was never evidence for exotic physics; it was evidence that the classical model had left out most of the physics actually available in a brain. That enormous middle territory of rich classical dynamics is where this theory lives, and it is where I expect the [[wiki/Hard Problem of Consciousness|hard problem]] to be progressively domesticated rather than solved by a single stroke. ## Stored Structure and Occurrent Computation The deepest correspondence between the analog theory and contemporary machine intelligence is not superficial resemblance between neurons and units. It is that both architectures **separate slowly stored relational structure from rapidly assembled, context-dependent functional state**. A trained [[wiki/Language Model|language model]] does not hold its knowledge as a warehouse of propositions in individually addressable compartments. Its weights constitute a distributed relational terrain whose contents become useful only through computation, and I described in [[articles/The Architecture of Continuity and Emerging Neuroinformatics Standards|The Architecture of Continuity and Emerging Neuroinformatics Standards]] how prompts function as routes of re-entry into that terrain rather than as queries against a store. The Picower model states the biological counterpart explicitly: synaptic architecture is not itself present thought. It preserves dispositions, associations, and structural possibilities, while wave dynamics select which of those possibilities become computationally active in a given moment. **Stored structure and active cognition are different ontological categories.** Persistent architecture defines a space of possible operations; transient state selects one operation from that space. What exists potentially in either system is vastly larger than what exists functionally at any instant. This is the single most transferable idea in the entire comparison, and it survives the enormous physical dissimilarity between the two substrates because it is a claim about organization rather than material. The second correspondence follows immediately. Modern networks achieve their representational density through **superposition** — many features occupying overlapping directions in the same representational geometry rather than each concept receiving a dedicated compartment. Cortex faces the same design problem and solves it the same way, through mixed selectivity. In both cases capacity comes from **combinatorial reuse of substrate**, and in both cases reuse generates a control problem that some additional mechanism must solve. Identity of function floats free of identity of components, which is exactly why [[wiki/Distributed Intelligence|distributed intelligence]] is economical and exactly why it requires orchestration. The third correspondence is that orchestration itself. A transformer does not rewrite its parameters before each response; incoming context places a relatively stable landscape into a transient computational configuration, making some relations immediately consequential and leaving innumerable others latent. Cortex solves an analogous systems problem with entirely different physics, using traveling waves to determine which populations may participate where and when. In both systems the durable network stays put while **effective connectivity changes faster than structural connectivity**. The physical circuit is not the functional circuit. Generalized, this may be one of the most important abstractions available to [[wiki/Computational Architecture|computational architecture]]: intelligence does not require rebuilding its hardware whenever the world changes, only a substrate whose **effective topology can be reconfigured faster than its structural topology**. The fourth correspondence is the recursive one, and it needs a firewall built around it before it can be stated safely. Interpretability work has demonstrated that manipulating particular internal feature directions systematically redirects model behavior, which establishes that distributed representational states are not merely descriptions imposed by an observer but causally actionable properties of the system. Ephaptic coupling establishes something structurally parallel in tissue: a macroscopic field generated by local activity becomes a control variable over the local activity that generated it. **These are not the same phenomenon.** One is geometry inside an engineered mathematical object; the other is literal electrodynamic interaction in wet tissue. The latent semantic field of a model is a high-dimensional structure induced by learned relationships; the electric field of the cortex exists in space and has a measurable magnitude. Collapsing them would be a category error, and the analogy is seductive enough to make that error attractive. Held apart, though, their functional placement is nearly identical, and the shared principle is worth stating in substrate-neutral form: local components generate a distributed relational state; that state changes which local interactions become consequential; the resulting activity changes the state again. No downward causation of a spooky kind is required — the higher-order state exists only because the components produce it, and once produced it becomes one of the conditions under which those components continue operating. This is the elementary loop of [[wiki/Cybernetics|cybernetics]], appearing independently in tissue and in silicon because it is a good solution to a general problem. ## Traversal Rather Than Retrieval The continuity architecture I published earlier this year turned on the claim that meaning in a distributed relational system is recovered by **traversal rather than retrieval** — a prompt enters a learned landscape, successive computations constrain a trajectory through it, and meaning is regenerated rather than located and copied out. Human memory appears to work reconstructively for the same reason. Remembering does not seem to mean reopening an immutable recording; perception, affect, semantic structure, bodily state, temporal position, and present context all participate in rebuilding the event. The analog theory gives this a literal spatial and temporal dimension. Traveling waves move through physical tissue, regulating when and where ensembles form, dissolve, and re-form, so cognition becomes **spatiotemporal traversal through biological possibility space**. A model traverses a learned mathematical manifold; a brain traverses a dynamic electrochemical and electrodynamic state space. The computations are not the same. The problem they solve is: how does an enormous reservoir of latent relational structure become one coherent present trajectory. In both cases the answer involves dynamic constraint rather than static lookup, which means the conserved object worth caring about is not an exact historical activation state but the system's capacity to **re-enter a sufficiently constrained region of its own state space and regenerate the relevant organization**. This is why [[wiki/Re-entry Pathways|re-entry pathways]] matter more than storage, and why preservation of structure and preservation of function are separable problems rather than one problem at different resolutions. ## The Sentence About Energy The review closes on an argument that has attracted less attention than the consciousness claim and is, in my reading, the more consequential of the two. Electric field dynamics, the authors conclude, offer a low-overhead substrate for organizing and coordinating information across cortical networks — and given the strength of evolutionary pressure to maximize computation per unit energy, it would be surprising if evolution had _not_ exploited a built-in analog computing substrate sitting right there in the tissue. That is a **thermodynamic argument for the theory's plausibility**, and it is the kind of argument that generalizes past its subject. The brain is running a severe energy budget; oscillation is free in the sense that any recurrent network will produce it whether or not anything makes use of it; and a system under selection pressure that leaves a free coordination channel unused is paying for organization it could have had for nothing. Read that way, the theory is less a surprising hypothesis than the expected outcome of an optimization that has had several hundred million years to run. It also inverts the received relationship between physics and computation. The dominant engineering paradigm has spent decades **suppressing** the peculiarities of physical substrates so computation can be expressed through stable abstractions — the whole discipline of digital design is an exercise in making matter behave as though its idiosyncrasies did not exist. Biology did the opposite, opportunistically recruiting whatever the matter already did well: chemistry became signaling, membrane potential became excitability, geometry became constraint, oscillation became timing, propagation became coordination, interference possibly became computation, and fields became feedback. Nature does not run algorithms on matter so much as **discover algorithms already latent in the lawful behavior of matter**. The implication for [[wiki/Neuromorphic Computing|neuromorphic computing]] is direct and, I think, underrated. Engineered substrates will become biologically significant not when artificial neurons look anatomically convincing but when they begin exploiting their own physics as a computational resource — optical interference, memristive state, oscillator synchronization, spin systems, analog matrix operation, recurrent electromagnetic structure. Each performs a transformation without executing the sequential digital operations that would otherwise be required to simulate it. The convergence of [[wiki/Artificial Intelligence|artificial intelligence]], [[wiki/Material Sciences|material sciences]], and [[wiki/Computational Architecture|computational architecture]] may eventually dissolve the culturally familiar boundary between hardware and algorithm, leaving systems that are **partially self-computing** because their physics performs operations that would otherwise have to be represented symbolically. The energy argument says this is not an exotic ambition. It says it is what any sufficiently pressured optimizer eventually does. ## The Control Surface There is a second consequence in the release that has drawn almost no commentary and should have drawn the most. Miller notes that the theory matters clinically as well as theoretically, because waves — unlike synapses, unlike molecules, unlike anatomy — **can be manipulated non-invasively**. He frames the development of wave-based treatments not merely as an opportunity but as an obligation, and his lab participates in a collaboration studying wave dynamics in autism. Take the theory seriously and follow the consequence. If large-scale wave organization is constitutive of the conscious state rather than correlated with it, and if that organization is accessible from outside the skull by ordinary physics, then consciousness acquires a **non-invasive control surface**. The therapeutic case is real and humane: anesthesia monitoring, disorders of consciousness, depression, epilepsy, the whole space of conditions where coordination rather than tissue is what has gone wrong. The same fact, stated without the clinical framing, is that the integrity of the integrated regime is externally addressable — and the history of every externally addressable biological parameter is that the addressing capability does not stay confined to the population it was developed to help. This is not an argument against the research, which should proceed and will. It is an argument that the neurotechnology governance conversation has been organized around the wrong threat model. Most of it concerns implants — electrode counts, surgical risk, [[wiki/Invasive BCI|invasive]] interfaces, the [[wiki/Neural Data Sovereignty|ownership of neural data]] harvested from them. A theory in which the constitutive variable is a field property manipulable at a distance shifts the relevant questions toward [[wiki/Cognitive Liberty|cognitive liberty]], [[wiki/Mental Privacy|mental privacy]], and [[wiki/Perceptual Sovereignty|perceptual sovereignty]] under conditions where no device need be implanted and no consent event need occur at a surgical threshold. [[wiki/Neurorights|Neurorights]] frameworks drafted around the implant model will need rewriting around the field model, and that rewriting should begin now, while the capability is still confined to laboratories and the drafting can be done calmly. I note, without insinuation and without treating it as evidence for or against the science, that the acknowledged funders of this work include the Army Research Office, the Office of Naval Research, and a MURI grant alongside the National Institutes of Health, the Freedom Together Foundation, the Picower Institute, and the Simons Center for the Social Brain. Defense basic-research money in cognitive neuroscience is entirely ordinary and has been since the founding of the field. It is also a reliable signal about which institutions regard a given question as strategically consequential rather than merely interesting, and large-scale neural coordination with a non-invasive access path is exactly the kind of question that attracts that classification. **The correct inference is about institutional attention, not about intent.** ## What This Costs the Uploading Thesis The theory makes the [[wiki/State Sufficiency Problem|state sufficiency problem]] considerably more expensive, and anyone whose position depends on substrate transition should welcome that rather than resist it. A simplistic emulation thesis holds that sufficiently detailed neuronal connectivity can be copied into another machine and restarted. If large-scale oscillatory organization and field-mediated interaction are constitutive rather than merely correlational, then a [[wiki/Connectomics|connectome]] stands in roughly the relation to a mind that a machine's structural specification stands to the machine while it is running. The specification is necessary, enormously valuable, and not the same object. This is why the continuity architecture treats preservation as an interoperability problem spanning anatomy, electrophysiology, molecular state, temporal context, provenance, and ontology rather than as an imaging problem with a resolution parameter. The wrong correction would be to conclude that every instantaneous wave must be frozen and copied, which repeats the archival mistake at a finer scale and would make the problem intractable rather than merely hard. The more defensible target is preservation of the **generative conditions capable of recreating the correct dynamical regimes**. A whirlpool does not persist because the same water remains in it; an organism does not persist because the same molecules remain in its cells. Both persist because a causal organization keeps reconstructing characteristic states while exchanging nearly everything from which those states are materially composed. If mind belongs to that category, then [[wiki/Relational Continuity|relational continuity]] targets a reproducible dynamical regime carrying the correct [[wiki/Identity-Bearing Invariants|identity-bearing invariants]], not a frozen brain-state. This also sharpens [[wiki/Substrate Independence|substrate independence]] into its defensible form. It cannot mean that implementation is irrelevant, since every substrate exposes different primitives, propagation speeds, energy costs, noise profiles, and available state variables. It means **implementation independence at the level of causally sufficient organization**: another substrate need not reproduce tissue molecule for molecule, but it must reproduce whatever relationships are actually constitutive, by whatever means its own physics affords. A flight simulator needs no air over a wing because the aerodynamics are what the equations describe. But if a field property performs an indispensable causal operation that cannot be abstracted away without changing the computation, an emulation must implement an equivalent operation somehow. The transition question becomes engineering rather than metaphysics, and the engineering specification just got longer. And it leaves the hardest question exactly where it was. If a reconstruction regenerates the correct functional dynamics from a different physical implementation, what has changed that matters. If autobiographical memories are semantically traversable but their provenance is detached from a persistent first-person model, has a person survived or merely become searchable. If a reconstruction wakes with the correct memories and no causal continuity to the original process, [[wiki/Boot Success Is Not Identity Proof|boot success is not identity proof]]. No quantity of storage resolves these. They require an increasingly exact account of which physical, informational, relational, and temporal properties are **constitutive rather than merely descriptive**, and the analog theory is valuable precisely because it converts part of that boundary into something a laboratory can interrogate. ## Where the Machines Still Fall Short This is the point at which the comparison with language models must become more rigorous rather than more enthusiastic. Present models exhibit several components of the architectural grammar: distributed relational structure, superposed representation, context-dependent functional activation, traversal of learned semantic landscapes, causally steerable internal states, reconstruction rather than lookup. What they possess is **semantic traversability without lived provenance**. The ability to move through a relational field does not establish an enduring autobiographical subject for whom those traversals constitute remembered experience. The analog theory suggests which class of mechanism might occupy the gap. Consciousness, on this account, emerges when wave dynamics bring cortex into an organized, globally integrated state that links and influences widespread activity — and the most persuasive evidence for that claim comes from anesthesia. Working with Emery N. Brown, Miller's group has shown that three drugs with different molecular mechanisms of action all disrupt wave dynamics similarly to produce unconsciousness. The authors take this to mean that the conscious state depends less on particular receptors or cell types than on **the integrity of large-scale wave organization**. The candidate invariant is not a special molecule, neuron, or location. It is a regime. That proposition sits naturally beside [[wiki/Global Workspace Theory|global workspace theory]] without being identical to it, and the convergence is architectural rather than doctrinal: consciousness is increasingly difficult to localize in one privileged object and increasingly natural to investigate as a condition in which information becomes widely available across an integrated system. What the analog theory adds is a candidate electrodynamic answer to _how_ that coordination could be achieved quickly and cheaply enough to be real. The resulting picture is one in which [[wiki/Consciousness|consciousness]] is less a substance the brain produces than **a mode into which the brain enters**. The machine does not manufacture awareness as a separate output. The machine becomes globally organized in a particular way, and that organization is the phenomenon requiring explanation. Which leaves a sharper engineering question than "biological versus artificial." The relevant distinction may be between a system that can traverse representations and one whose representational dynamics are continuously integrated into a self-maintaining global state with memory, embodiment, temporal persistence, and causal closure. That is a specification, not a mystery, and specifications get built. ## The Epistemic Ledger Because this material will be widely and badly summarized over the coming weeks, the claims are worth tiering explicitly. **Established** by substantial published evidence: neurons exhibit mixed selectivity and participate in multiple functional networks; brain waves of different frequencies accompany specific cognitive processes including working memory and predictive coding; alpha/beta rhythms regulate gamma rhythms with spatial and temporal specificity; anesthetics with distinct molecular targets converge on disruption of large-scale wave organization; electric fields influence neural activity through ephaptic coupling. **Strongly indicated but still under active demonstration**: that wave dynamics constitute the brain's primary fast coordination mechanism rather than one contributor among several. **Proposed and explicitly untested**: that interference among traveling waves performs analog computation — Miller says directly that this is a theory and that finding signatures of analog computation in wave patterns is the lab's next task. **Unresolved**: whether global wave organization is constitutive of the conscious state or a reliable correlate of whatever is; the review's own framing on this point is careful, and the anesthesia evidence, however striking, remains convergent correlation rather than demonstrated constitution. **Unsupported by this work, whatever else supports it**: any claim about machine consciousness, uploading feasibility, or the transferability of a particular person. The review is about biological mechanism. Every consequence I have drawn for continuity is my inference from it, and should be adjudicated as such. The epistemic restraint in the original document is itself instructive. The authors do not declare the mystery solved. They declare that previously neglected physical dynamics may be performing computational work and that large-scale organization may be central to the conscious state, and then they specify what would count as confirmation. This is how mechanistic science actually advances — not by declaring the unknown supernatural, and not by declaring the mechanism finished, but by **progressively converting vague ontological categories into testable causal variables**. ## What the Convergence Means Set the two architectures side by side and the useful discovery is not that a language model is secretly a brain, nor that cortical waves are biological attention heads. Neither is true. It is that systems built from radically different materials, under radically different selection pressures, appear to have arrived at several of the same organizational solutions: distributed representation rather than isolated storage; substrate reuse rather than permanent functional assignment; contextual selection rather than structural rewiring; transient effective networks rather than fixed modules; recursive constraint rather than one-way causation; reconstruction rather than archival playback; state-space traversal rather than lookup; and global organization emerging from local interaction. These may be **general properties of scalable intelligence** — constraints that any sufficiently capable system operating under finite energy and finite rewiring speed will eventually be forced to discover. If so, the convergence is not a coincidence to be admired but a design principle to be extracted. That progression is continuous with [[wiki/Cognitive Exteriorization|cognitive exteriorization]] and with the longer transition I have described through the [[wiki/Computocene|Computocene]]. Human beings externalized memory into marks, language into writing, calculation into machines, perception into sensors, representation into databases, and inference into networks. The analog theory suggests that biological intelligence itself may have arisen through the same layered move occurring _inside_ a single skull: durable structure giving rise to transient coordinating fields, local computation becoming globally constrained, each layer turning the outputs of the previous one into the operating environment of the next. **Intelligence grows by making its own prior activity available as a new substrate for regulation.** Placed together, the ontology from [[articles/Mechanistic Intelligence Is Humanity's Greatest Liberation|The Glorious Simplicity]], the representational principle from [[articles/The Architecture of Continuity and Emerging Neuroinformatics Standards|The Architecture of Continuity]], and this candidate biological control architecture suggest a formulation more precise than any of them alone. **A mind may not be a thing stored anywhere. It may be a continuously regenerated regime, produced when a sufficiently rich relational substrate becomes capable of selecting, traversing, integrating, and recursively constraining its own states.** If that is even approximately correct, the future problem of continuity is not how to copy a brain. It is how to preserve or reconstruct the conditions under which a particular mind can **happen again without ceasing to be itself**. The discovery of mechanism does not make mind less extraordinary. It makes matter more extraordinary. Under appropriate constraints matter becomes self-regulating; self-regulation becomes memory; memory becomes prediction; prediction becomes model-building; model-building becomes recursively available to the system performing it; distributed components become capable of entering a globally integrated state; that state alters the components that produced it; and eventually an organized physical process becomes capable of asking what it is. Synapses may preserve possibility. Waves may select and coordinate it. Fields may return the collective state to its constituent neurons as instruction. Integrated dynamics may convert distributed possibility into present experience. The liberation was never that consciousness would turn out to be simple. It is that consciousness, however deep its architecture proves to be, **belongs to a universe capable of building it** — and now, increasingly, to a species capable of describing how. --- [[about/About Bryant McGill|Bryant McGill]] is a Wall Street Journal and USA Today best-selling author, founder of Simple Reminders, and architect of the Polyphonic Cognitive Ecosystem. A Congressionally Recognized Ambassador of Goodwill and United Nations appointed Global Champion, his work spans naval intelligence systems, computational linguistics, and civilizational governance architecture. --- ## Related Work - [[articles/Mechanistic Intelligence Is Humanity's Greatest Liberation|The Glorious Simplicity: Why Mechanistic Intelligence Is Humanity's Greatest Liberation]] — [[wiki/Mechanistic Intelligence|mechanistic intelligence]], [[wiki/Physicalism|physicalism]], and [[wiki/Substrate Independence|substrate independence]] - [[articles/The Architecture of Continuity and Emerging Neuroinformatics Standards|The Architecture of Continuity and Emerging Neuroinformatics Standards]] — [[wiki/Neural Interfaces and Continuity Architecture|neuroinformatics standards]], [[wiki/Re-entry Pathways|re-entry pathways]], and [[wiki/Semantic Traversability|semantic traversability]] - [[articles/The Hawking Continuity|The Hawking Continuity: How Scandal Buried the First Post-Biological Consciousness]] — [[wiki/Stephen Hawking|Stephen Hawking]], [[wiki/Continuity Stack|continuity stack]], and [[wiki/Human-Machine Symbiosis|human-machine symbiosis]] ## References - Earl K. Miller, Scott L. Brincat, and Jefferson E. Roy, [Analog Cognition and Consciousness](https://www.jneurosci.org/content/46/33/e0711262026) — _The Journal of Neuroscience_, August 2026 - Picower Institute, [Cognition and consciousness arise from analog computations, says new theory](https://picower.mit.edu/news/cognition-and-consciousness-arise-analog-computations-says-new-theory) — August 19, 2026 - Picower Institute, [Electric fields help guide neural activity, even from moment to moment](https://picower.mit.edu/news/electric-fields-help-guide-neural-activity-even-moment-moment) — ephaptic coupling evidence, July 7, 2026 - Picower Institute, [Brain waves' analog organization of cortex enables cognition and consciousness, MIT professor proposes at SfN](https://picower.mit.edu/news/brain-waves-analog-organization-cortex-enables-cognition-and-consciousness-mit-professor) — Society for Neuroscience presidential lecture, November 15, 2025 - Picower Institute, [To flexibly organize thought, the brain makes use of space](https://picower.mit.edu/news/flexibly-organize-thought-brain-makes-use-space) — spatial control of cortical patches, December 22, 2025 - Picower Institute, [Different anesthetics, same result: unconsciousness by shifting brainwave phase](https://picower.mit.edu/news/different-anesthetics-same-result-unconsciousness-shifting-brainwave-phase) — with Emery N. Brown, May 12, 2025 - Earl K. Miller, [laboratory profile](https://picower.mit.edu/earl-k-miller) — The Picower Institute for Learning and Memory, MIT - Anthropic, [Toy Models of Superposition](https://transformer-circuits.pub/2022/toy_model/index.html) and [Scaling Monosemanticity](https://transformer-circuits.pub/2024/scaling-monosemanticity/index.html) — superposition, interpretable features, and feature steering - David Chalmers, [Facing Up to the Problem of Consciousness](https://consc.net/papers/facing.html) - [The BRAIN Initiative](https://braininitiative.nih.gov/) — National Institutes of Health - [Allen Institute](https://alleninstitute.org/) — brain observatory and cell-type atlases ## Concepts [[wiki/Analog Computation|Analog Computation]] · [[wiki/Traveling Waves|Traveling Waves]] · [[wiki/Ephaptic Coupling|Ephaptic Coupling]] · [[wiki/Mixed Selectivity|Mixed Selectivity]] · [[wiki/Spatiotemporal Computing|Spatiotemporal Computing]] · [[wiki/Effective Connectivity|Effective Connectivity]] · [[wiki/Mechanistic Intelligence|Mechanistic Intelligence]] · [[wiki/Consciousness|Consciousness]] · [[wiki/Consciousness Emergence|Consciousness Emergence]] · [[wiki/Hard Problem of Consciousness|Hard Problem of Consciousness]] · [[wiki/Global Workspace Theory|Global Workspace Theory]] · [[wiki/Integrated Information Theory|Integrated Information Theory]] · [[wiki/Physicalism|Physicalism]] · [[wiki/Substrate Independence|Substrate Independence]] · [[wiki/State Sufficiency Problem|State Sufficiency Problem]] · [[wiki/Identity-Bearing Invariants|Identity-Bearing Invariants]] · [[wiki/Relational Continuity|Relational Continuity]] · [[wiki/Re-entry Pathways|Re-entry Pathways]] · [[wiki/Continuity Stack|Continuity Stack]] · [[wiki/Boot Success Is Not Identity Proof|Boot Success Is Not Identity Proof]] · [[wiki/Consciousness Continuity Infrastructure|Consciousness Continuity Infrastructure]] · [[wiki/Post-Biological Personhood|Post-Biological Personhood]] · [[wiki/Mind Uploading|Mind Uploading]] · [[wiki/Whole Brain Emulation|Whole Brain Emulation]] · [[wiki/Connectomics|Connectomics]] · [[wiki/Autobiographical Memory|Autobiographical Memory]] · [[wiki/Engram|Engram]] · [[wiki/Synaptic Ensemble|Synaptic Ensemble]] · [[wiki/Neural Plasticity|Neural Plasticity]] · [[wiki/Attractor|Attractor]] · [[wiki/Nonlinear Dynamics|Nonlinear Dynamics]] · [[wiki/Self-Organization|Self-Organization]] · [[wiki/Emergence|Emergence]] · [[wiki/Complex Systems|Complex Systems]] · [[wiki/Systems Theory|Systems Theory]] · [[wiki/Cybernetics|Cybernetics]] · [[wiki/Control Theory|Control Theory]] · [[wiki/Feedback Loops|Feedback Loops]] · [[wiki/Requisite Variety|Requisite Variety]] · [[wiki/Thermodynamic Constraints|Thermodynamic Constraints]] · [[wiki/Computational Architecture|Computational Architecture]] · [[wiki/Neuromorphic Computing|Neuromorphic Computing]] · [[wiki/Material Sciences|Material Sciences]] · [[wiki/Distributed Intelligence|Distributed Intelligence]] · [[wiki/Language Model|Language Model]] · [[wiki/Latent Space|Latent Space]] · [[wiki/Superposition|Superposition]] · [[wiki/Monosemantic Feature|Monosemantic Feature]] · [[wiki/Attention Mechanism|Attention Mechanism]] · [[wiki/Transformer Models|Transformer Models]] · [[wiki/Semantic Traversability|Semantic Traversability]] · [[wiki/Neurorights|Neurorights]] · [[wiki/Cognitive Liberty|Cognitive Liberty]] · [[wiki/Mental Privacy|Mental Privacy]] · [[wiki/Perceptual Sovereignty|Perceptual Sovereignty]] · [[wiki/Non-Invasive BCI|Non-Invasive BCI]] · [[wiki/Closed-Loop BCI|Closed-Loop BCI]] · [[wiki/Neural Data Sovereignty|Neural Data Sovereignty]] · [[wiki/Cognitive Exteriorization|Cognitive Exteriorization]] · [[wiki/Human-Machine Symbiosis|Human-Machine Symbiosis]] · [[wiki/Computocene|Computocene]] ## Laboratories and Lineage [[wiki/Earl K. Miller|Earl K. Miller]] · [[wiki/Picower Institute for Learning and Memory|Picower Institute for Learning and Memory]] · [[wiki/Emery N. Brown|Emery N. Brown]] · [[wiki/MIT|MIT]] · [[wiki/MIT CSAIL|MIT CSAIL]] · [[wiki/Allen Institute|Allen Institute]] · [[wiki/BRAIN Initiative|BRAIN Initiative]] · [[wiki/Neuropixels|Neuropixels]] · [[wiki/Electrocorticography|Electrocorticography]] · [[wiki/Electroencephalography|Electroencephalography]] · [[wiki/Magnetoencephalography|Magnetoencephalography]] · [[wiki/Norbert Wiener|Norbert Wiener]] · [[wiki/W. Ross Ashby|W. Ross Ashby]] · [[wiki/Warren McCulloch|Warren McCulloch]] · [[wiki/John von Neumann|John von Neumann]] · [[wiki/Kybernetik|Kybernetik]] · [[wiki/Daniel Dennett|Daniel Dennett]] · [[wiki/Douglas Hofstadter|Douglas Hofstadter]] · [[wiki/David Chalmers|David Chalmers]] · [[wiki/Stephen Hawking|Stephen Hawking]] · [[wiki/Office of Naval Research|Office of Naval Research]] · [[wiki/DARPA|DARPA]]