Sources

Papers

Books arrive through Kindle and the post. Papers arrive through Google Scholar, OpenAlex and PubMed — a century of argument, most of it a search away and a good deal of it free. Accumulation has never been cheaper, and what follows is the trace of some of it: what one reader gathered, in the order the questions came up.

There are two accounts of how a mind works, and for twenty years they barely spoke.

One begins in 1982, when John Hopfield treated a network of neurons as a physical system with an energy — a landscape of hills and valleys, so that remembering stopped being a filing problem and became a rolling-downhill problem. The other begins with the claim that cortex does not receive the world at all: it guesses the world, compares the guess against what arrives, and passes on only the difference. Karl Friston turned that into a general theory of self-organising things; Andy Clark carried it into philosophy of mind.

By the 2010s there were two vocabularies. One camp said attractor, basin, energy landscape, connectome. The other said prediction error, precision, generative model, free energy. They were working on one organ, in one century, with overlapping funding and adjacent conference hotels, and almost nothing passed between them — not because anyone was wrong, but because they had no shared words for an arriving idea to land on.

Since about 2021 that has been closing, and three of the papers below do most of the closing. One sets out a single framework — similarity, separation, projection — that every single-shot associative memory turns out to be a special case of. Another shows a temporal predictive-coding network, in its simplest form, falling inside it: an asymmetric Hopfield network whose comparison step is done on whitened inputs. A third derives the same attractor dynamics from the free-energy principle itself rather than importing them — Hopfield-style memory falling out of the blanket formalism as a special case. Not one camp absorbed by another. All three cases of the same third thing.

The Integral of Life is a top-down hypothesis built on accumulation, resonance and imagination — in individual minds, compounding across the group. Below are five sets of papers I have found useful, grouped by where each one bears on the argument rather than by field — a book belongs to a discipline, but a paper is a single result, and what counts is which claim it touches. (The bibliography and conceptual foundations are grouped by field.) They are not exhaustive and they are not conclusive, and a few are here because they push against the argument rather than for it. Titles link through to the paper.

A note on method. I used AI as a research instrument — to search, to retrieve, and to check. Every citation here has been verified against the bibliographic record rather than against somebody’s reference list. Author lists longer than about six names are shortened; the link carries the full one. The selection is mine, and so is the judgement about what each paper does and does not do for the argument. I have assembled the model’s pieces from others; but the idea to do that, and the pursuit of the checking, testing and verifying to the degree possible, has been a human quest.

Accumulation19

AC-01 Neural networks and physical systems with emergent collective computational abilitiesJ. J. HopfieldPNAS 79(8), 2554–2558, 1982The paper that made remembering a rolling-downhill problem rather than a filing one, and handed the study of memory to physicists.
AC-02 Neurons with graded response have collective computational properties like those of two-state neuronsJ. J. HopfieldPNAS 81(10), 3088–3092, 1984The same result taken off the two-state idealisation and onto neurons that fire at a rate, which is what real ones do.
AC-03 Learning representations by back-propagating errorsDavid E. Rumelhart, Geoffrey E. Hinton & Ronald J. WilliamsNature 323, 533–536, 1986Knowledge distributed across a weight matrix rather than filed anywhere in particular — the engineering form of the claim this book makes about a life.
AC-04 Learning to predict by the methods of temporal differencesRichard S. SuttonMachine Learning 3(1), 9–44, 1988The update rule as a running total: incremental accumulation of prediction error, discounted with age. The integrand, in a page of arithmetic.
AC-05 Children creating language: how Nicaraguan Sign Language acquired a spatial grammarAnn Senghas & Marie CoppolaPsychological Science 12(4), 323–328, 2001Deaf children with no shared language between them built one, and grammar appeared across the first cohorts. There was no deposit to inherit — the structure came out of the receivers.
AC-06 Language as shaped by the brainMorten H. Christiansen & Nick ChaterBehavioral and Brain Sciences 31(5), 489–509, 2008Language changes far faster than genes can track, so the fit between the two is better explained by language having been shaped to the receiver than by the receiver having been rebuilt for language.
AC-07 Cumulative cultural evolution in the laboratory: an experimental approach to the origins of structure in human languageSimon Kirby, Hannah Cornish & Kenny SmithPNAS 105(31), 10681–10686, 2008Structure appears in a language passed down a chain of learners, with no designer and no genetic change — compositionality emerging because it has to fit through finite minds.
AC-08 A scale-invariant internal representation of timeKarthik H. Shankar & Marc W. HowardNeural Computation 24(1), 134–193, 2012A bank of leaky integrators with a spread of decay rates computes the Laplace transform of its own input, and a second layer approximates the inverse — the transform falling out of membrane leakage and a population rather than being computed by anything. The variable here is real throughout: a spectrum of decay rates encoding how long ago, with no phase and no oscillatory component.
AC-09 Hierarchical process memory: memory as an integral component of information processingUri Hasson, Janice Chen & Christopher J. HoneyTrends in Cognitive Sciences 19(6), 304–313, 2015Memory is not a store the processing consults. It is a property of the processing, with a different time constant at every level.
AC-10 Cultural evolutionary theory: how culture evolves and why it mattersNicole Creanza, Oren Kolodny & Marcus W. FeldmanPNAS 114(30), 7782–7789, 2017Culture as an inheritance system with dynamics of its own — the accumulation that runs above the individual and outlasts them.
AC-11 Evolutionary neuroscience of cumulative cultureDietrich Stout & Erin E. HechtPNAS 114(30), 7861–7868, 2017What cumulative culture did to the brain that carries it. The loop between the two accumulations, argued from toolmaking.
AC-12 Attention is all you needAshish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser & Illia PolosukhinNeurIPS 2017; arXiv:1706.03762Incoming signal routed through what the parameter set already holds. The mechanism this book calls resonance, built for an entirely different reason.
AC-13 Dense associative memory is robust to adversarial inputsDmitry Krotov & John HopfieldNeural Computation 30, 3151–3167, 2018Sharpen the energy landscape and capacity climbs far above the classical bound. The accumulation holds a great deal more than anyone had assumed.
AC-14 Computation through neural population dynamicsSaurabh Vyas, Matthew D. Golub, David Sussillo & Krishna V. ShenoyAnnual Review of Neuroscience 43, 249–275, 2020The review that moved the field off single-cell tuning and onto population trajectories — a state moving through an accumulated space, which is the altitude this book works at.
AC-15 The Tolman–Eichenbaum machine: unifying space and relational memory through generalization in the hippocampal formationJames C. R. Whittington, Timothy H. Muller, Shirley Mark, Guifen Chen, Caswell Barry, Neil Burgess & Timothy E. J. BehrensCell 183, 1249–1263, 2020Structure learned once and reused everywhere. What accumulates is not the places but the relationships between them.
AC-16 Hopfield networks is all you needHubert Ramsauer et al.ICLR 2021; arXiv:2008.02217The transformer's attention layer turns out to be a modern Hopfield network. Two literatures that had never spoken had been writing the same equation.
AC-17 Default mode network connectivity predicts individual differences in long-term forgetting: evidence for storage degradation, not retrieval failureAndrea StoccoPLOS Computational Biology 21(9), e1013485, 2025A decay rate, per person, measured. Each participant's speed of forgetting fitted from their own accuracy and response times, then predicted from resting-state connectivity alone. What differs between people is how fast storage degrades, not whether retrieval succeeds.
AC-18 Network memory consolidation under adaptive rewiringSergey A. Lobov, Alexey Zharinov, Dmitry Kurganov & Viktor B. KazantsevEuropean Physical Journal Special Topics 234, 3669–3681, 2025Rewiring holds a trace far longer than changing the weights does, and holds it by changing the anatomy. The consolidation happens during spontaneous activity, after the learning has stopped. Forgetting is the price of the plasticity that makes remembering possible.
AC-19 Verbalizable Representations Form a Global Workspace in Language ModelsWes Gurnee et al.Transformer Circuits Thread, 6 July 2026LLM being used as a telescope, complete with Jacobian terminology.

Resonance18

RE-01 Balanced amplification: a new mechanism of selective amplification of neural activity patternsBrendan K. Murphy & Kenneth D. MillerNeuron 61(4), 635–648, 2009Selective amplification with every mode stable. Because an asymmetric network’s modes are not orthogonal, an arriving pattern can be amplified by orders of magnitude without anything running away. Resonance as a consequence of asymmetry rather than of instability.
RE-02 Speaker–listener neural coupling underlies successful communicationGreg J. Stephens, Lauren J. Silbert & Uri HassonPNAS 107(32), 14425–14430, 2010Two brains converge, and how far they converge predicts how well the communication actually worked. Coupling, measured.
RE-03 The functional role of cross-frequency couplingRyan T. Canolty & Robert T. KnightTrends in Cognitive Sciences 14(11), 506–515, 2010Rhythms nested inside rhythms, as a general mechanism — and the licence for expecting more than one orthogonal oscillatory dimension.
RE-04 Tagging the neuronal entrainment to beat and meterSylvie Nozaradan, Isabelle Peretz, Marcus Missal & André MourauxJournal of Neuroscience 31(28), 10234–10240, 2011The brain builds the beat rather than receiving it — a periodicity in the signal that is not in the sound. The oldest coupling technology, caught in the act.
RE-05 Brain-to-brain coupling: a mechanism for creating and sharing a social worldUri Hasson, Asif A. Ghazanfar, Bruno Galantucci, Simon Garrod & Christian KeysersTrends in Cognitive Sciences 16(2), 114–121, 2012The general statement: minds couple through a shared physical carrier, and the interesting part happens on arrival.
RE-06 The theta–gamma neural codeJohn E. Lisman & Ole JensenNeuron 77(6), 1002–1016, 2013Content in the fast rhythm, order in the slow one. The clearest case of a brain using phase to carry something.
RE-07 Human brain networks function in connectome-specific harmonic wavesSelen Atasoy, Isaac Donnelly & Joel PearsonNature Communications 7, 10340, 2016Brain activity as a superposition of standing waves on the connectome. Resonance not as metaphor but as the eigenmodes of an actual structure.
RE-08 Cortical travelling waves: mechanisms and computational principlesLyle Muller, Frédéric Chavane, John Reynolds & Terrence J. SejnowskiNature Reviews Neuroscience 19(5), 255–268, 2018Cortical activity travels. Once that is granted, phase is a real variable and complex numbers are its ordinary description.
RE-09 The easy part of the hard problem: a resonance theory of consciousnessTam Hunt & Jonathan W. SchoolerFrontiers in Human Neuroscience 13, 378, 2019The most ambitious resonance claim in print. Here as the far end of the range rather than as agreement — this book keeps consciousness at its boundary.
RE-10 Universal Hopfield networks: a general framework for single-shot associative memory modelsBeren Millidge, Tommaso Salvatori, Yuhang Song, Thomas Lukasiewicz & Rafal BogaczICML 2022, PMLR 162, 15561–15583Every single-shot associative memory turns out to be one template — similarity, separation, projection. A tidy-up that makes the family visible.
RE-11 Geometric constraints on human brain functionJames C. Pang, Kevin M. Aquino, Marianne Oldehinkel, Peter A. Robinson, Ben D. Fulcher, Michael Breakspear & Alex FornitoNature 618, 566–574, 2023The shape of the cortex predicts its activity better than the wiring diagram does. Uncomfortable for the connectome reading this book leans on, and here for that reason.
RE-12 Sequential memory with temporal predictive codingMufeng Tang, Helen Barron & Rafal BogaczNeurIPS 2023; arXiv:2305.11982A single-layer temporal predictive-coding network, in linear form, retrieves as an asymmetric Hopfield network does — except that it compares whitened inputs, which is why it holds up on correlated material where the Hopfield versions collapse.
RE-13 Input-driven dynamics for robust memory retrieval in Hopfield networksSimone Betteti, Giacomo Baggio, Francesco Bullo & Sandro ZampieriScience Advances 11, 2025The arriving signal reweights the landscape it arrives into, deepening the matching valley. The accumulation recognising itself in the input rather than being corrected by it.
RE-14 Dynamics of Continuous Attractor Neural Networks With Spike Frequency AdaptationYujun Li, Tianhao Chu & Si WuNeural Computation 37(6), 1057–1101, 2025Spike-frequency adaptation acts as a slow negative feedback that destabilises the network’s own resting bump — the same architecture holds a stable representation and, once adaptation engages, updates it. Accumulation and moment-to-moment computation from one mechanism, not two.
RE-15 Inertial asynchronous computationDoruk Efe Gökmen, Michel Fruchart, Dmitrii Zendrikov, Giacomo Indiveri, Giulio Biroli & Vincenzo VitelliarXiv:2607.21965, 2026Split the hardware into two asymmetrically coupled parts, the way position couples to momentum, and the inertia that results lets many units compute together with nothing orchestrating them. Validated on neuromorphic silicon. The asymmetry is not a defect to be corrected; it is what carries order.
RE-16 Brain modes of resonance estimated by a biophysical multi-compartment finite elements modelInês Gonçalves, Dulce Oliveira, Catarina Rocha, Joana Cabral & Marco ParenteScientific Reports 15, 33845, 2025The same brain, different modes. Eigenmodes computed from tissue rigidity and viscosity rather than geometry alone, shifting while shape and connectivity stay fixed — so the resonant profile is a state and not only a structure. The Atasoy caveat holds: real eigenvalues on a physical body, an analogy for the matrix rather than evidence about it.
RE-17 Learning to Generate Reviews and Discovering SentimentAlec Radford, Rafal Józefowicz & Ilya SutskeverarXiv:1704.01444, 2017A single unit in a byte-level language model, trained only to predict the next character of Amazon reviews, tracks sentiment almost perfectly — and the feature is corpus-shaped: retrain on books and it is gone. A contribution to this book's line that a trained model integrates without resonating — the axis is the cheapest regularity in the distribution, not a receiver changed by what it read.
RE-18 Dynamic embeddedness within the structural connectome characterizes macroscale functional organization of the human brainWeiyang Shi, Congying Chu, Yu Zhang & Tianzi Jiang et al.Science Bulletin, in press, 2026Score every frame of activity for how closely it conforms to the connectome’s own harmonics, and the stable picture turns out to be assembled from the minority of moments that conform most — the rest is flux. The claim that only a small part of what is stored is live at any instant, measured rather than asserted. Real eigenmodes on a symmetric graph, and it reads the basis as the connectome where RE-11 reads it as geometry.

Imagination15

IM-01 The cognitive neuroscience of constructive memory: remembering the past and imagining the futureDaniel L. Schacter & Donna Rose AddisPhilosophical Transactions of the Royal Society B 362(1481), 773–786, 2007Remembering and imagining run on the same constructive machinery. The founding statement of the continuity this book depends on.
IM-02 Patients with hippocampal amnesia cannot imagine new experiencesDemis Hassabis, Dharshan Kumaran, Seralynne D. Vann & Eleanor A. MaguirePNAS 104(5), 1726–1731, 2007Damage the hippocampus and the patient loses not only the past but the ability to imagine a new scene at all. The shared machinery shown by breaking it rather than by imaging it.
IM-03 Constructive episodic simulation of the future and the past: distinct subsystems of a core brain network mediate imagining and rememberingDonna Rose Addis, Ling Pan, Mai-Anh Vu, Noa Laiser & Daniel L. SchacterNeuropsychologia 47(11), 2222–2238, 2009The regions credited to thinking about the future turn out to support imagining in general. The first crack in the mental-time-travel framing.
IM-04 Default and executive network coupling supports creative idea productionRoger E. Beaty, Mathias Benedek, Scott Barry Kaufman & Paul J. SilviaScientific Reports 5, 10964, 2015Idea production couples the default network first to salience, then to executive control. A sealed room does not need a door that keeps opening.
IM-05 Right temporal alpha oscillations as a neural mechanism for inhibiting obvious associationsCaroline Di Bernardi Luft, Ioanna Zioga, Nicholas M. Thompson, Michael J. Banissy & Joydeep BhattacharyaPNAS 115(52), E12144–E12152, 2018Reaching a remote association runs by inhibiting the obvious one — and you cannot inhibit what is not already running.
IM-06 Neuroscience of imagination and implications for human evolutionAndrey VyshedskiyCurrent Neurobiology 10(2), 89–109, 2019Binding by synchrony: an imagined scene is the phase relation between ensembles the mind already holds, not a picture fetched from anywhere.
IM-07 Imagined speech can be decoded from low- and cross-frequency intracranial EEG featuresTimothée Proix et al.Nature Communications 13, 48, 2022Recorded from inside human heads: imagined speech carries a low- and cross-frequency signature distinct from spoken speech. A different band, one substrate.
IM-08 Predictive processing and perception: what does imagining have to do with it?Dan Cavedon-TaylorConsciousness and Cognition 106, 103419, 2022A philosopher's attack on the assumption that imagining must be sealed off from the world. Projected imagery runs onto the real, not instead of it.
IM-09 Modeling the contribution of theta-gamma coupling to sequential memory, imagination, and dreamingGabriele Pirazzini & Mauro UrsinoFrontiers in Neural Circuits 18, 2024Cut a phase-code model off from its input and it replays and recombines. Loosen the recombination further and the model dreams.
IM-10 Episodic recombination and the role of time in mental travelJohannes B. Mahr & Daniel L. SchacterPhilosophical Transactions of the Royal Society B 379(1913), 20230409, 2024Recombination is one capacity; pointing it at a time is a second. Which means the out-of-phase dimensions need not be about the future at all.
IM-11 The role of alpha oscillations in free- and goal-directed semantic associationsIoanna Zioga, Yoed N. Kenett, Anastasios Giannopoulos & Caroline Di Bernardi LuftHuman Brain Mapping 45, e26770, 2024Intending to be creative shifts alpha phase synchronisation from left temporal to right. Intention showing up as a phase relation rather than as activation.
IM-12 Possible worlds theory: how the imagination transcends and recreates realityDacher Keltner & Eftychia StamkouAnnual Review of Psychology 76, 329–358, 2025Reality and possibility as a blend rather than a switch, and imagination as the thing that builds shared worlds — money, law, gods.
IM-13 A neural basis for distinguishing imagination from realityNadine Dijkstra, Thomas von Rein, Peter Kok & Stephen M. FlemingNeuron 113, 2536–2542, 2025Imagined and perceived content intermixed in one substrate, told apart by strength rather than by kind. Which is a complete account of the boundary that needs no phase story at all.
IM-14 Constructing future behavior in the hippocampal formation through composition and replayJacob J. W. Bakermans, Joseph Warren, James C. R. Whittington & Timothy E. J. BehrensNature Neuroscience 28, 1061–1072, 2025Compose a map out of pieces already held and an animal behaves correctly in a place it has never been, with nothing learned there. Replay does the composing.
IM-15 Enhancing deep neural networks through complex-valued representations and Kuramoto synchronization dynamicsSabine Muzellec, Andrea Alamia, Thomas Serre & Rufin VanRullenTransactions on Machine Learning Research, June 2025; arXiv:2502.21077Amplitude for presence, phase for membership. Beats the real-valued network, and beats the complex one with the phases held still — so the phase is doing work.

Meaning and the Second Law10

Schrödinger’s What Is Life? belongs at the head of this group but is a book rather than a paper, and sits in the bibliography at MP-06 — in the edition that carries Mind and Matter and the autobiographical sketches alongside it.

MS-01 A mathematical theory of communicationC. E. ShannonBell System Technical Journal 27(3), 379–423, 1948Information measured, and deliberately stripped of meaning. Everything this book says about meaning is said in the gap Shannon left open.
MS-02 Information theory and statistical mechanicsE. T. JaynesPhysical Review 106(4), 620–630, 1957Entropy as inference rather than as a property of the world — the move that lets thermodynamics and information theory be one subject.
MS-03 Irreversibility and heat generation in the computing processR. LandauerIBM Journal of Research and Development 5(3), 183–191, 1961Forgetting costs energy. The physical price of erasure, and the reason decay is not free.
MS-04 Time, structure, and fluctuationsIlya PrigogineScience 201(4358), 777–785, 1978The Nobel lecture. Structure maintained by dissipation — order that exists because it is spending, which is the shape of every claim in Part Three.
MS-05 The thermodynamics of computation — a reviewCharles H. BennettInternational Journal of Theoretical Physics 21(12), 905–940, 1982The review that made Landauer's principle load-bearing, and charged Maxwell's demon for what it forgets.
MS-06 The free-energy principle: a unified brain theory?Karl FristonNature Reviews Neuroscience 11(2), 127–138, 2010The most-read statement of the position this book takes the update rule from and declines the summit of. Here to be read rather than to be agreed with.
MS-07 Statistical physics of self-replicationJeremy L. EnglandThe Journal of Chemical Physics 139(12), 121923, 2013A thermodynamic lower bound on self-replication. The second law not as what life resists but as what drives it.
MS-08 Causal entropic forcesA. D. Wissner-Gross & C. E. FreerPhysical Review Letters 110(16), 168702, 2013Entropy maximised over future paths, producing behaviour that looks like foresight. Contested, and here because the contest is the interesting part.
MS-09 Semantic information, autonomous agency and non-equilibrium statistical physicsArtemy Kolchinsky & David H. WolpertInterface Focus 8(6), 20180041, 2018Semantic information defined as the information a system holds that is causally necessary for maintaining its own existence far from equilibrium. Meaning made measurable, and grounded in self-maintenance rather than in mind.
MS-10 Assembly theory explains and quantifies selection and evolutionAbhishek Sharma, Dániel Czégel, Michael Lachmann, Christopher P. Kempes, Sara I. Walker & Leroy CroninNature 622, 321–328, 2023Measure how much history an object requires to exist. A physics of accumulated structure, arrived at from chemistry.

Free Energy and Active Inference8

Karl Friston’s active inference is the closest existing formal account to this book’s territory — the update rule for a single organism, where this book asks about what a lifetime of updating accumulates into. The principle itself, two papers arguing it claims more generality than it has earned, and the multi-agent extensions that come nearest to resonance without quite reaching it. It’s interesting to see the Markov blanket morph over time to more closely resemble the Hopfield network.

FE-01 Answering Schrödinger’s question: A free-energy formulationMaxwell J. D. Ramstead, Paul B. Badcock & Karl J. FristonPhysics of Life Reviews 24, 1–16, 2018Negentropy, then surprise, then the imperative to keep existing — Schrödinger’s question, formalised. The floor this book’s information axis builds from, not its ceiling.
FE-02 A Free Energy Principle for a Particular PhysicsKarl FristonarXiv:1906.10184, 2019Markov blankets nested across scales, quantum to classical, each one a boundary an internal state infers across. The single-organism frame this book’s coupled integrals depart from.
FE-03 How particular is the physics of the free energy principle?Miguel Aguilera, Beren Millidge, Alexander Tschantz & Christopher L. BuckleyPhysics of Life Reviews 40, 24–50, 2022The derivation goes through only for a narrow class of systems, not the general case claimed. One of two papers here arguing the principle overreaches its own scope.
FE-04 The Markov Blanket Trick: On the Scope of the Free Energy Principle and Active InferenceVicente Raja, Dinesh Valluri, Edward Baggs, Anthony Chemero & Michael L. AndersonPhilSci Archive #18843, 2021FEP generalises Bayesian inference to anything with a statistical boundary; active inference presupposes the perception and action it claims to explain. The other half of the overreach case.
FE-05 Shared Protentions in Multi-Agent Active InferenceMahault Albarracin, Riddhi J. Pitliya, Toby St. Clere Smithe, Daniel A. Friedman, Karl Friston & Maxwell J. D. RamsteadEntropy 26(4), 303, 2024Agents converging on a shared generative model through phenomenology and category theory — a common estimate, arrived at together. Coordination, in this book’s terms, not resonance.
FE-06 As One and Many: Relating Individual and Emergent Group-Level Generative Models in Active InferencePeter Thestrup Waade, Christoffer Lundbak Olesen, Jonathan Ehrenreich Laursen, Samuel W. Nehrer, Conor Heins, Karl Friston & Christoph MathysEntropy 27(2), 143, 2025A collective with a group-level Markov blanket behaves as one larger agent. The nearest Friston’s cluster comes to coupled integrals — still short of it, since nothing here has one agent’s arriving signal preferentially activate what another already holds.
FE-07 What the flock knows that the birds do not: exploring the emergence of joint agency in multi-agent active inferenceDomenico Maisto, Davide Nuzzi & Giovanni PezzuloarXiv:2511.10835, 2026Flocking modelled directly: birds minimising free energy, coupled to their neighbours, the flock itself acquiring a Markov blanket of its own. What the flock knows about the predator is a property of the coupling, not of any one bird’s history with the signal — the clearest case yet of coordination without resonance.
FE-08 Self-orthogonalizing attractor neural networks emerging from the free energy principleTamas Spisak & Karl FristonNeurocomputing 682, 133472, 2026Applies the blanket recursively — internal states subdivided into overlapping sub-particles, each with its own blanket — and Hopfield-style attractor dynamics fall out as a special case, orthogonalised and Hebbian. The boundary, nested deeply enough, starts to look like memory.