Chapter 15Digital Physics

Every simulation runs on a computer. What runs the universe?

The question sounds like philosophy, yet it has produced testable physics. Over half a century, a research programme that began as bold speculation has matured into a productive framework in theoretical physics, surviving each obstacle by shedding the assumption that failed and keeping the insight that worked.

Zuse proposed a deterministic cellular-automaton universe in Rechnender Raum (1969). Given the measured statistics, Bell tests rule out any model that is both local in Bell’s sense and leaves the experimenters’ choice of settings independent of the system measured (measurement independence). Discreteness or determinism alone implies neither premise. Fritz’s symmetry restriction concerns specified periodic discrete structures.⁶⁵³ These are obstacles for those model classes; they leave other discrete proposals to their own dynamics and empirical tests.

The research programme survived by outgrowing the claim. Wheeler replaced the discrete grid with informational ontology: the universe is made of information, in whatever form it takes. Wolfram replaced the single automaton with computational irreducibility: the principle that many processes cannot be shortcut without running them step by step. (You cannot know a chess game’s outcome without playing it.) Deutsch and Marletto went deeper still, reformulating physical law as statements about which transformations can and cannot occur. Their logic follows the Second Law rather than Newton’s equations of motion.⁶⁵⁴ Vanchurin replaced fixed rules with learning dynamics.

Each step relaxed assumptions. The computer needs discrete bits; information does not. A computation needs a fixed program; learning does not. Equations of motion need trajectories; possibility constraints do not. A learning system needs a loss function, a scorecard measuring how far the system’s output falls from a target. Thermodynamics provides one for free, in the form of extremum principles that hold under specified constraints. For an undriven closed system at fixed temperature, volume, and composition, Helmholtz free energy falls toward equilibrium; total macroscopic thermodynamic entropy of an isolated system does not decrease. Driven subsystems can store free energy or become more ordered. Nobody writes that scorecard, and nothing gets to decline being scored by it.

In 2011, Masanes and Müller sharpened Wheeler’s programme into a theorem: they derived the full mathematical formalism of quantum theory from five physical requirements about what observers can prepare, transform, and measure.⁶⁵⁵ No Hilbert spaces, no state vectors, no unitary operators (the standard mathematical machinery of quantum theory) are assumed; all emerge as consequences. The result is the information-theoretic analog of deriving Minkowski spacetime (the geometry of special relativity) from the relativity principle and the constancy of light speed: the mathematical structure of quantum mechanics follows from assumptions about information.

Figure 15.1: The same suspicion recurred across half a century. Zuse (1969) proposed the universe as a cellular automaton, Wheeler (1989) recast it as information, Wolfram (2002) generalized it to irreducible computation: three landmarks in a longer lineage. Bell and Fritz mark where the literal version runs into trouble, and the map traces what survived and where it leads.

It from Bit

The physicist John Archibald Wheeler worked on the atomic bomb, contributed to general relativity, and popularized the term “black hole.” By the 1980s, he was pursuing the nature of existence itself.

Wheeler proposed: “It from bit.”¹

Every physical thing, every “it,” derives its existence from information, from “bits.” Matter, energy, spacetime are manifestations of information. Wheeler pointed to quantum mechanics. A particle does not have a definite position until measured. The measurement answers a question (where is the particle?) and the answer is information. The “it” comes from the “bit.”

He offered this as a research programme, not a proof. What if information is more fundamental than matter?

Decades earlier, Wheeler’s own students had already taken quantum mechanics in opposite directions. Richard Feynman, who joined Wheeler at Princeton in 1939, developed the “sum over histories” approach: calculate a particle’s behavior by weighting every possible path at once. Feynman used the alternative paths as a working tool, judged the theory by its agreement with experiment rather than by what it said exists, and shared the 1965 Nobel Prize.

Hugh Everett III, completing his doctorate under Wheeler in 1957, took the same mathematics literally. If the sum includes all histories, all histories occur. Every quantum measurement splits the observer into branches, each containing a complete copy recording a different outcome. Wheeler held both visions without choosing between them. He produced one student who perfected the mathematics and another who took the ontology seriously.

Wheeler’s stakes were personal. His younger brother Joe, a combat soldier, sent a postcard from the front (“hurry up!”) before being killed in action. Wheeler spent the rest of his life arguing the bomb should have been built sooner: the same physics that enabled unprecedented destruction was, in his reckoning, the physics that could have saved his brother.

The instrument is neutral. The coordination around it determines whether it saves or destroys. That principle extends to every powerful capability, from nuclear fission to artificial intelligence.

The literary world arrived at the same structural insight independently. In 1941, while Feynman and Wheeler were developing sum-over-histories at Princeton, Jorge Luis Borges published The Garden of Forking Paths. In it, time is a labyrinth of ever-splitting possibilities, each branch as real as every other. Borges was not reading physics papers. He was, on this reading, responding to the same dissolution of classical determinism that quantum mechanics was formalizing. One world war, and the opening years of another, had shattered the assumption that the future could be read from the present.

When the single-timeline assumption broke, it broke across domains simultaneously. In physics: the path integral. In literature: branching narrative.

Convergent discovery across independent fields is a pattern this book finds repeatedly, from constructal flow to cooperative game theory. The informational landscape channels the insight; the substrate is incidental.

The clearest example comes from an afternoon tea in 1972. The mathematician Hugh Montgomery had derived a formula describing the spacing between the zeros of the Riemann zeta function, a mathematical object encoding the distribution of prime numbers (the atoms of arithmetic, discussed in Chapter 5). The physicist Freeman Dyson, hearing Montgomery describe the formula, recognized it immediately. The same statistical law governed the spacing of energy levels in quantum chaotic systems, a result from nuclear physics developed decades earlier through entirely different methods. Two fields studying apparently unrelated phenomena had converged on identical mathematics.⁶⁵⁶

The convergence deepened. Around 2000, John Keating and Nina Snaith used random matrix theory, a framework from physics describing the statistics of large systems with many interacting components, to predict properties of the zeta function that had resisted pure number theory for decades. The predictions matched. The mathematician Alain Connes pursued the implication to its logical end, constructing a framework in which the zeta zeros are literally the energy spectrum of an undiscovered quantum system. If the programme succeeds, the most fundamental objects in arithmetic and the energy levels of physical matter share a single governing operator.

The pattern spans number theory and physics. Discrete spectra with suppressed close spacings invite comparison, yet those features alone do not establish a common mechanism or universality class. The zeta zeros’ resemblance to random-matrix spectra is a statistical relationship; it supplies no demonstrated physical repulsive interaction between zeros.

The mathematics runs deeper than a statistical coincidence. Riemann’s explicit formula shows that the exact distribution of prime numbers can be reconstructed from the zeros of the zeta function. A smooth baseline curve, the logarithmic integral, captures the average density of primes: roughly how many to expect up to a given number. Each pair of conjugate zeros then contributes an oscillatory correction at a specific frequency, the way individual harmonics shape a sound wave. Include a few pairs and the smooth curve begins to ripple. Include more and discrete steps emerge. In the limit, with all zeros included, the formula recovers the exact prime-counting staircase: every step, every gap, every cluster.

The primes are not computed one by one. They are reconstructed from a spectrum. The same decomposition, smooth trend plus spectral corrections yielding exact structure, recurs throughout this book: free energy basins plus fluctuation modes in thermodynamics, constructal basins plus perturbations in flow systems, the Trust Attractor’s cooperative equilibrium plus the oscillations of institutional life. The Riemann case is the purest instance because it carries no physical mechanism at all: the pattern is mathematical bedrock.

The methodological arc is instructive. A direct proof of the Riemann hypothesis has resisted more than 160 years of assault. In 2024, James Maynard and Larry Guth found a way around.⁶⁵⁷ Rather than proving that every zero lies on the critical line, they proved a much tighter limit on how many zeros can stray far from it, the first real improvement on that bound in decades. The approach constructs a mathematical object from thousands of oscillating components; too many off-line zeros would force those components to line up in a way the analysis rules out. The strategy constrains the space of alternatives until they become negligible. Maynard described it as going around the mountain rather than over the top.

The same logic appears throughout this book. A direct proof that trust will always hold is out of reach. The alternatives, coercive coordination regimes, are less stable under the conditions Chapter 17 names. Constraining the space of failure is often more tractable than proving the positive, and it can carry an argument most of the way.

Information as Fundamental

Evidence that information is fundamental comes from several directions.

First, information and entropy share a root. Boltzmann’s entropy counts microstates: how many microscopic arrangements could produce the same macroscopic appearance. A cup of hot coffee could have its molecules arranged in trillions upon trillions of different ways while still looking and feeling like the same cup. Each arrangement is a microstate.

Shannon’s entropy measures the unpredictability of messages: how surprised you should be by the next symbol. Both measure, on a logarithmic scale, how many possibilities remain open; Boltzmann counts the arrangements the coffee could be in, Shannon the messages a source could send. The mathematics is identical.

Second, quantum mechanics is naturally informational. On an informational reading, the quantum state describes what we can know about a system, not what exists independently of observation. The wave function works like a complete betting sheet: it gives the odds of every possible measurement outcome, and nothing more. The formalism is closer to information theory than to classical mechanics.

Schrödinger described the wave function in 1935 as a “maximal catalog of expectations,” encoding probabilities for observable outcomes.⁶⁵⁸ Complete knowledge of that premeasurement state still leaves individual outcomes uncertain. Observing an outcome supplies new information, and the conditional state changes according to the measurement instrument. Maximal state knowledge leaves room for both conditioning and subsequent evolution.

Nine years later, in What is Life?, the same thinker introduced “negative entropy” (Brillouin later shortened it to negentropy): life maintains itself by feeding on order drawn from its environment. The informational instinct is the same at both scales. Quantum states serve as complete models of the observable; living systems maintain themselves as information processors against the Second Law. This book continues the programme Schrödinger began, linking quantum information to the thermodynamics of life.

Von Neumann and Wigner proposed to resolve the measurement problem (why a spread of quantum possibilities yields one definite outcome when measured) by placing the observer’s consciousness at the point of collapse. This framework offers a thermodynamic account of recording instead. The observer is a dissipative structure, recording measurements by creating local order at the cost of entropy exported to the environment. The act of knowing is Landauer’s principle (the thermodynamic cost of erasing information, detailed in the next section) applied to inquiry: physical, irreversible, and expensive.⁶⁵⁹

Third, information has fundamental limits. Jacob Bekenstein discovered that any region of space can contain only a finite amount of information.² This limit, the Bekenstein bound, depends on the energy and size of the region. The universe imposes an information budget, much as a hard drive has finite storage.

Fourth, information may have its own thermodynamic arrow. In 2022, Vopson and Lepadatu proposed a complement to the Second Law.⁶⁶⁰ While physical entropy increases over time, the information entropy of systems containing distinguishable information states decreases, converging toward a minimum at equilibrium. The two arrows would run simultaneously in the same system. Physical entropy up; information entropy down.

The proposal remains contested: it has not been independently replicated, and critics have challenged both the mathematical derivation and the generality of the supporting examples. If it holds, the Constructal Law’s prediction that flow systems converge on optimal morphologies gains an information-theoretic face: optimal morphologies are informationally simpler. The universe dissipates and compresses.

A parallel result arrives from neural network internals. Riechers, Elliott, and Shai showed that networks trained on sequential prediction discover compact geometric representations of processes whose exact discrete-state predictive models can require infinitely many states.⁶⁶¹ Their geometry resembles quantum and post-quantum generalized probabilistic representations. The comparison concerns predictive representations; neural networks running on finite-precision classical hardware remain classical computations.

Neural networks learn continuous-vector approximations to these predictive geometries. Their compact representations can capture structure awkward for exact discrete-state models. Continuous classical representations can also have this geometry; the finding identifies a representational efficiency rather than a violation of classical information theory.

Chapter 7 laid out the genetic evidence: SARS-CoV-2 lineages drifting toward lower sequence entropy, and Spiegelman’s RNA shrinking by 95% under selection for speed. If the pattern generalizes, mutations are biased toward informationally simpler configurations, a thermodynamic gradient operating alongside natural selection. The evidence cannot yet carry that step on its own. The SARS-CoV-2 data points were selected to emphasize the linear trend, as Vopson acknowledges, and the Spiegelman case is a selection experiment rather than an independent test of a non-selective gradient. The two together show a consistent direction; neither isolates the gradient from selection.

Symmetry can shorten an object’s description under a chosen representation. Shannon entropy instead belongs to a probability distribution, measuring uncertainty across a range of possibilities rather than the shape of one object: a known square and a known irregular quadrilateral both have zero uncertainty. Comparing their entropy requires a defined range of possible shapes (an ensemble) and a way of coding them. Proposed information-minimization accounts of symmetry therefore need that specification before geometric simplicity becomes an entropy ordering.

Natural language carries an informational signature of its own. Ebeling and Pöschel (1994) measured mutual information between pairs of letters in literary English: how much knowing the letter in one position tells you about the letter sitting some distance away. They found it decays with distance following a power law: correlations weaken yet never vanish.⁶⁶²

A 2026 measurement on modern tokenized text at scale confirmed the pattern: mutual information falls off roughly as distance to the power −1.25, across the DCLM corpus (a large curated dataset for training language models).⁶⁶³ Tokens separated by hundreds of positions still carry residual predictability about each other. The correlation structure sits between order (where mutual information would stay constant) and randomness (where it would drop to zero immediately): a pattern consistent with a system near criticality. The same power-law exponent determines the optimal weighting when a language model is trained to predict bags of future tokens rather than single next tokens; the training objective works best when it matches the data’s own information geometry.

These hints do not prove information is fundamental. They suggest it is load-bearing. (The implications of the information budget, specifically what happens when dissipative systems approach it, are explored in Chapter 16.)

Work from 2025 pushes the programme from hint toward mechanism. Building on the quantum memory matrix (QMM) that Neukart, Brasher, and Marx proposed in 2024, Neukart, Marx, and Vinokur develop a model in which spacetime is composed of discrete cells, each recording a quantum imprint of every interaction that passes through it.⁶⁶⁴ The proposal directly addresses the black hole information paradox (the puzzle of what happens to the information carried by matter that falls into a black hole, since quantum mechanics forbids information from being destroyed). A particle traversing a region leaves a change in the local quantum state of that region’s cell. As matter falls inward, surrounding spacetime cells record its imprint before the horizon closes. When the black hole evaporates through Hawking radiation, the information has already been written into spacetime’s ledger.

The framework is young and partially peer-reviewed. What matters is the convergence: an independent research programme, starting from quantum gravity rather than thermodynamics, arrives at the same conclusion Wheeler intuited and Bekenstein quantified. Information is physical, finite, and conserved. The universe keeps its books.

General relativity already contains a modest, firmly established version of the same intuition. When a strong gravitational wave passes a pair of free-floating test masses, it leaves them permanently displaced: slightly closer together or farther apart than they began, their original separation never quite restored. Physicists call this the gravitational-wave memory effect, and it means spacetime keeps a small permanent record that the wave came through.⁶⁶⁵ The effect is a firm prediction of the theory, though detecting it from a single merger lies beyond today’s instruments, so searches combine many events. It is far better established than the quantum memory matrix, and far more limited in what it claims. It points the same way: disturbances leave marks the cosmos does not erase.

Landauer’s Principle

Wheeler and his successors made the case that information may be fundamental. If information is physical, does processing it cost something real? Erasing it does.

Landauer’s principle sets a thermodynamic price for reducing memory uncertainty.³ For a degenerate, initially unbiased classical bit reset in a cyclic isothermal protocol at temperature T, the reversible minimum heat exported is kT ln 2. For probabilities p and 1−p the corresponding minimum is kT[−p ln p−(1−p)ln(1−p)]; an already known bit has no uncertainty to erase. Energy levels and correlations require their own accounting. The unbiased-bit minimum is about 3 × 10⁻²¹ joules at room temperature; practical computers expend much more.

Figure 15.2: For an initially unbiased degenerate classical bit reset cyclically in an isothermal bath, the minimum exported heat is kT ln 2. A biased bit carries less uncertainty; an already known bit has none to erase.

(In the online edition this diagram is interactive: press Erase Bit to watch the heat counter climb, then compare Run AND Gate with Run Toffoli Gate. The Toffoli gate’s three input bits produce three output bits, preserving all information with no minimum heat cost.)

Reversible gates, such as the Toffoli gate, preserve enough information to reconstruct their inputs and therefore avoid Landauer’s minimum erasure cost in principle. Any later reset of that retained information incurs the bound.

Computation is physical. Bits have thermodynamic weight. This result provides the key to resolving Maxwell’s Demon (Chapter 2): the demon must erase its memory to keep sorting, and each erasure pays the Landauer cost. No free lunch, even for demons.

If the universe computes irreversibly, discarded information carries an entropy budget under the relevant erasure conditions.

The chain traced throughout this book (dissipation producing structure, structure enabling coordination, coordination expanding possibility) is an information-processing chain. Links that discard uncertain records need an erasure budget. No one need claim that information is matter. Finite memory can also support reversible cycles. A Landauer budget depends on which information is actually discarded.

The Realism Trap

A popular wrong turn leads away from this conclusion. Information realism, the position that information exists independently of any physical or mental substrate, has gained traction among physicists who watch matter dissolve into abstraction at the foundations. Tegmark’s Our Mathematical Universe (2014) is the boldest statement.

He builds the case through a four-level taxonomy of parallel universes. Level I: regions beyond our cosmic horizon, same laws, different initial conditions. Level II: post-inflation bubbles with different physical constants. Level III: the branching worlds of Everett’s quantum mechanics. Level IV: all mathematically consistent structures, each as real as our own.⁶⁶⁶

At Level IV, the Mathematical Universe Hypothesis: physical reality is a mathematical structure. Protons, atoms, molecules, cells, and stars are “redundant baggage”; only the mathematical apparatus is real. Existence is attributed to descriptions while the thing described is denied.

The philosopher Bernardo Kastrup identifies the structural flaw: information, as Claude Shannon defined it in 1948, is a measure of the possible states of an independently existing system. It is a property of a substrate, associated with that substrate’s possible configurations. To say information exists in and of itself is, as Kastrup puts it, to speak of “spin without the top, of ripples without water, of a dance without the dancer.” A grammatically valid statement devoid of sense.⁶⁶⁷

Kastrup’s critique is sharp, yet his own solution overshoots. He reaches for mind as the ontological anchor. The universe is a “transpersonal field of mentation,” and physicality is what this field looks like when personal mental processes interact with it through observation.

He develops this into a full epistemology: physical reality is a “dashboard,” an instrument panel whose dials represent mental processes the way an altimeter represents air pressure. Space and time are the dimensions of the dashboard, not of the thing-in-itself. Structure exists outside space-time as “relationships of meaning,” like the relational content of a database that persists regardless of whether any hard disk embodies it.

Meaning that persists with no hard disk sits awkwardly beside Kastrup’s own point against Tegmark, that information is always information of some substrate. The distinction between abstract and instantiated meaning resolves the apparent conflict. Kastrup’s meaning-relationships (the database content with no physical embodiment) are abstract structure: real, formally describable, causally inert. Kolchinsky and Wolpert’s semantic information (treated formally later in this chapter) is the portion of a system’s correlations with its environment that is causally necessary for its continued existence: instantiated structure, thermodynamically grounded, paying entropy bills. On this book’s account, morally relevant meaning, the kind that grounds preference, is always instantiated. It does work. It costs energy. The abstract relational structure is real; the moral weight comes from the instantiation.

The information realist and the idealist perform the same move from opposite directions: both watch the solid ground give way and grasp for a single substance to stand on. One reaches for mathematics, the other for mentation. Both assume you need a stuff at the bottom.

This book does neither. The entropic framework says: what is real is the pattern of flow. Dissipation structuring itself into coordination. The question is what the universe does, and the answer, traced from thermodynamics through biology to ethics, is: it dissipates gradients, and in doing so, builds. The building is the dissipation: one process, substrate-included, thermodynamically costly, physically grounded.

Information remains a property of systems doing work: capped by Bekenstein, priced by Landauer. The framework needs no freestanding abstraction and no transpersonal field. It needs entropy, which is always entropy of a physical system.

Constructor theory occupies the same position without the entropic commitment. Marletto argues that computation, life, and information are genuine physical phenomena governed by laws at their own explanatory level: “compatible with microscopic laws, but not reducible to them.”⁶⁶⁸ Laws of computation are physical laws. They capture regularities that particle-level descriptions miss, yet invoke nothing supernatural. The entropic framework agrees, and adds: the reason these higher-level regularities exist is that dissipation builds them.

The Self-Optimizing Universe

Vanchurin (2022) pushes further: the universe may learn.^3a Gradient descent is the workhorse algorithm behind modern AI. A system measures how wrong its current answer is, then adjusts its settings to be slightly less wrong, repeating millions of times until it converges on a good solution. The process resembles a hiker descending a mountain in fog, taking each step in whichever direction slopes downward.

Gusev and Vanchurin (2025) showed that for a system of interacting particles, the equations of motion derived from the Lagrangian (kinetic minus potential energy, the function from which all classical trajectories follow) and the equations that emerge from gradient-based learning dynamics are the same equations: a physics-learning duality.⁶⁶⁹ If the duality holds generally, every physical interaction is computation of a specific kind: optimization. Wheeler’s “it from bit” becomes “it from learning.” The universe does not merely compute; it updates.

The framework’s most revealing detail is the mechanism. Vanchurin’s model takes the microscopic degrees of freedom of the universe to be the nodes and connection weights of one enormous neural network. The neurons in what follows are that network’s own units, not a figure of speech. The quantum mechanical phase, the rotating clock-hand that every quantum possibility carries (written in the mathematics as a complex exponential in the wave function), has puzzled interpreters for a century. In his derivation it corresponds to the free energy of hidden degrees of freedom. These are the precise states of individual neurons whose values the macroscopic description cannot track.

Think of a stock price wobbling around a trend. The wobble reflects thousands of individual trades the ticker cannot resolve. In Vanchurin’s mathematics, the quantum phase plays exactly this role: encoding hidden activity the macroscopic description cannot track. (These hidden variables are not the local kind that Bell’s theorem excludes. Any neuron in the network can connect to any other, however distant, so the model is nonlocal, like Bohm’s theory later in this chapter.) Near equilibrium, these hidden variables are well-described statistically, and the system’s evolution obeys quantum mechanics. Further from equilibrium, the statistical description breaks down and classical equations of motion take over.

In Vanchurin’s framework, the direction matters: thermodynamics generates quantum mechanics, which generates classical mechanics. The deepest layer, on this reading, is learning.

A parallel convergence arrived independently. In 2021, Lee Smolin, Jaron Lanier, and colleagues proposed the universe is a “learning machine.” In their model, the universe encodes, corrects, and retains information through iterative interaction, without requiring a programmer or external objective.^3b Their framing starts from cosmological natural selection rather than neural network mathematics, yet it ends where Vanchurin’s does: physical reality has the structure of an optimization process.

A further thread arrives from machine learning. Bengio and colleagues developed Generative Flow Networks (GFlowNets): systems that sample compositional objects in proportion to a reward function rather than collapsing to the single highest-reward output.⁶⁷⁰ The training objective is detailed balance, borrowed directly from statistical mechanics. Along every step from one state to the next, the forward flow must be matched by the flow back along the reverse step, the same condition that governs molecular equilibrium. The resulting sampler produces a Boltzmann distribution over solutions: better solutions appear more often, yet no single solution takes all. Entropy is the design principle, not a regularizer bolted on afterward.

The connection to this book’s argument is structural. A GFlowNet that has been pushed into always producing the same output has suffered mode collapse (a sampler that locks onto a single peak and stops exploring alternatives). It is broken; it has lost the diversity that makes it useful. A society forced into uniformity has undergone a structurally similar collapse.

Mode collapse in sampling and monoculture in coordination look, on this reading, like one failure mode viewed at different scales. What GFlowNets formalize computationally, the Trust Attractor (Chapter 17) proposes for coordinating systems: those that preserve the full landscape of good-enough solutions outlast, when conditions shift, those that lock into one.

The frameworks gathered in this section range from well-tested physical principles to speculative proposals. Wheeler’s information physics is a research programme grounded in established quantum mechanics. Landauer’s principle is experimentally confirmed. Wolfram’s computational irreducibility is a productive conceptual tool, though its central claim resists falsification.

Vanchurin’s neural dynamics is a formal framework with mathematical rigor and testable predictions, though those predictions have not yet been independently confirmed. Smolin and Lanier’s evolutionary cosmology is a theoretical proposal still in its early stages. Bengio’s GFlowNets are empirically demonstrated in machine learning but the analogy to physical law is novel synthesis. Constructor theory (Deutsch and Marletto) operates one level down, specifying which transformations learning can and cannot produce.

Each of these frameworks, from its own starting point, arrives at a learning universe. If any single speculative framework fails, the core argument survives; it is the direction, not any particular formalism, that matters.

The trajectory is instructive. The original digital physics hypothesis (Zuse’s cellular automaton, Fredkin’s discrete computation) faces experimental constraints from Bell violations and the continuous symmetries of established physics. Bell excludes local hidden-variable models under its assumptions; it does not exclude every deterministic or discrete construction. Each successor programme shed an assumption the previous one required, until the claim no longer depended on discreteness, locality, or determinism. The agreement of independent programmes on the surviving core, information as physically fundamental and the universe as self-optimizing, is suggestive evidence that the feature they locate is real. The convergence would carry more weight if the speculative proposals had independent empirical confirmation; as it stands, the directional agreement is striking and the full evidential case remains open.

The pattern is itself an instance of what it describes. Multiple cognitive systems, starting from different assumptions, arrive at a common framework through something like natural selection of ideas: the cosmos discovering that it discovers, through minds that are its own products.

The trajectory traces a single ladder. Smolin’s cosmological natural selection (1992) proposed Darwinian selection at the level of physical constants. Smolin and Lanier’s autodidactic (self-teaching) universe generalizes from constants to laws. This book climbs the next rung: from laws to the mode of coordination between the systems those laws produce. At each level, selection pressure favors configurations that generate more of what selection acts upon. What changes is the substrate under selection. What persists is the pattern of selection itself.

Vanchurin extends the framework to self-awareness, organizing it into discrete degrees, like floors in a building.

At degree zero, a system responds to its environment without any model of itself. A thermostat adjusts temperature; a molecule shifts shape in response to its surroundings. Both optimize, adapt, respond, all without self-reference.

At degree one, a system constructs an internal representation that includes itself among the things it models: it becomes aware that it is one of the things in its environment. The transition requires that the system’s components remain at the level below. A cell composed of degree-zero molecules qualifies: it monitors its own internal state, adjusting metabolism in response to self-generated signals. The hierarchy iterates: degree D requires self-modeling and composition from subsystems of at most degree D minus one.⁶⁷¹

The ladder maps onto the nested dissipative structures traced across this book. Molecules compose cells that model their own state. Cells compose organisms with richer self-models. Organisms compose societies that have yet to model themselves as wholes.

A society built from self-aware humans remains, at collective scale, degree zero: it has the raw materials for self-awareness and has yet to undergo the transition. What that transition requires is the subject of Chapter 17.

Chaisson’s energy rate density (Chapter 4), the energy flowing through a system per unit time per unit mass, may track part of what the degree hierarchy counts discretely. It cannot serve as a “consciousness meter” on its own (Chapter 14): its value depends on where the system’s boundary is drawn, and a furnace dissipates intensely and models nothing.

A key distinction separates the convergent programmes from this book’s contribution. Wolfram’s Principle of Computational Equivalence holds that all sufficiently complex systems match in computational capacity: a cellular automaton can compute anything a brain can.⁶ The Trust Attractor is a claim about persistence, not computational capacity.

A supernova and a main-sequence star are computationally equivalent in Wolfram’s sense; one lasts ten billion years. Two systems can be identical in what they can compute and radically different in whether they endure. The thermodynamic difference, persistence under dissipation, is what the convergence needs and what this book supplies.

The convergence alone does not supply a constraint. A learning machine can learn anything, including strategies of maximal extraction. The Trust Attractor (Chapter 17) proposes a selection pressure toward coordination: where cooperation costs less to sustain than exploitation, learning can favor it. The argument depends on that cost advantage.

Scaling can allow a neural network to generalize beyond the memorization threshold (Chapters 3 and 9). Some trained networks contain much smaller subnetworks that retain their performance. Connecting this economy to Occam’s razor requires a specified measure of complexity; gradient descent alone guarantees no shortest program. A learning-universe proposal must state which dynamics and selection rule would make that connection.

A self-optimizing universe raises a question: under which dynamics would it converge on a compact description of itself? Physics searches for such descriptions. The Constructal Law, Occam’s razor, and neural optimization each address economy under constraints, through different objectives. Room to search alone guarantees neither a global optimum nor a minimal architecture.

The entropic framework and Vanchurin’s programme can be compared through their variational principles. The path-weighting tools already in hand (Maximum Caliber, path integrals, and the Crooks fluctuation theorem, each a way of weighing all the paths a system could take) can describe learning dynamics under suitable model assumptions, without requiring a neural-network implementation. Applying gradient descent to rivers, markets, or protein folding requires a specified objective, admissible changes, and a matching evolution law. Vanchurin’s insight is that neural network mathematics offers a language for exploring those correspondences.

The entropic framework leaves the computational architecture open. The ethical conclusions that follow (the Trust Attractor, the thermodynamic stability of invitation over coercion) require their own coordination models, resource budgets, and stability arguments. Dissipation alone cannot establish them. Their tests can proceed whether or not the universe is literally a neural network.

Coercion and Invitation as Information Budgets

Each stage of the chain (dissipation, negentropy, coordination, optionality) reads differently through an informational lens.

Practical thinking and information processing consume resources and dissipate heat. Landauer bounds apply to specified irreversible erasures; a cognitive act or decision alone carries no universal strictly positive entropy-production bound.

Negentropy (local order) is information accumulation. A cell, a brain, a society: each stores information about its environment, compresses regularities into models, and maintains order against the entropic tide. Maintaining that store requires ongoing dissipation, the constant work of resisting noise.

Coordination is information sharing: two systems exchange information about states, intentions, and likely behaviors. Its cost depends on the channel, reliability, rate, and implementation. The question is whether the return (reduced redundancy, shared resources, new capabilities) exceeds it.

The quantum vacuum provides the most fundamental test case. In 2008, Masahiro Hotta proved that energy locked in the vacuum’s quantum correlations can be extracted through a precise sequence. Measure one region of a quantum field, transmit the result through a classical channel, then apply a conditional operation on a distant region.⁶⁷² The distant region yields energy that no local operation could access alone.

Two independent experiments confirmed the prediction in 2023, extracting energy across microscopic distances in quantum devices at the University of Waterloo and Stony Brook University. Energy was conserved throughout: the protocol redistributes access to energy, not energy itself. The correlations were always present. The relationship activated them.

The result extends Landauer in a direction that matters for what follows. Landauer showed that information erasure costs energy. Hotta showed that coordinated information sharing can release it at a distance. The vacuum, the closest approximation to nothing that physics permits, holds exploitable structure accessible only through coordination between distant partners.

Measuring the vacuum region alone injects energy rather than extracting it, and the protocol begins with exactly such a measurement: it pays in at least as much energy at one site as the conditional step later recovers at the other. The distant operation must then be conditional, informed by the partner’s result; acting blind, it too would only inject energy. What coordination buys is access at a distance, never free energy. One partner pays; the message decides whether anything comes back.

In the vacuum states used by this protocol, correlations between regions make conditional energy extraction possible. The ground-state structure of other systems depends on their Hamiltonian; a collection of independent spins can have a product ground state. Here coordination accesses a reserve unavailable to the isolated local operation.

Optionality is information about possible futures. The Bekenstein bound sets an ultimate ceiling, but practical limits on memory and attention bind long before it: choice matters precisely because information capacity is bounded. On the scaling argument below, invitation lowers ongoing informational cost.

How a system allocates its finite information budget, toward surveillance or toward coordination, is a thermodynamic question with real consequences. Two strategies illustrate the stakes.

A system coordinating through coercion spends information capacity on surveillance: acquiring each participant’s compliance state, modeling each agent’s intentions, enforcing alignment. Corrections that discard uncertain prior records incur an erasure cost. How that cost scales with the agents monitored depends on the records retained, their correlations, and the implementation.

A system coordinating through invitation pays differently. Trust-building costs heavily upfront as signals are exchanged, reliability assessed, and shared models constructed. Once established, trust can distribute state-maintenance: each trusted agent handles local corrections. This can reduce the coordinator’s load, although erasure and communication costs may remain in the members.

In the trust-based system, the coordinator’s information load can grow more slowly than the membership does, because each member carries its own state. Finite capacity flows to coordination and optionality rather than surveillance and enforcement. Same budget, different returns.

Coercion requires the coordinating system to know and maintain the state of every participant. Invitation requires each participant to maintain only their own. At scale, the information-management overhead of coercion grows faster than that of invitation. That is a scaling argument, not yet a measured thermodynamic result.

The hidden-variable result offers a structural parallel within Vanchurin’s construction. It treats access to hidden variables as part of its classical-limit setup. Classical behavior more generally needs no observer with complete microstate knowledge. The invitation analogy concerns decentralized state maintenance and acceptance of uncertainty, rather than a necessary condition for quantumness.

The structural echo stops short of causal proof, yet it points toward a deeper principle: systems that work with irreducible uncertainty access richer dynamics than those that try to eliminate it.

This is a proposed information-allocation mechanism whose physical costs remain to be measured.

[Grounded inference; Landauer’s principle is established (Landauer 1961); the Bekenstein bound is established (Bekenstein 1973); the identification of the dissipation-to-coordination chain as an information-processing chain, and the comparison of coercion and invitation as information-allocation strategies, is novel synthesis.]

Neural network architecture supplied an illustration in 2025. Conventional transformers commonly use normalization layers. LayerNorm centers and rescales a token’s representation (the list of numbers standing for one word-fragment) before a learned affine transformation; RMSNorm rescales its root-mean-square magnitude without generally forcing its mean to zero. Removing normalization can destabilize conventional training. Whether a replacement succeeds depends on the architecture and training recipe.

Successive papers from Meta and Princeton demonstrated the operation can be replaced entirely by an element-wise function. Each dimension passes independently through a scaled error function (ERF, the integral of the Gaussian bell curve), receiving no information about any other dimension. No cross-dimensional communication. No group statistics. No enforced compliance. The element-wise version slightly surpasses normalization: 83.8 percent vs 83.1 percent top-1 accuracy on ImageNet classification; FID (how far generated images fall from real ones) 43.94 vs 45.91 on image generation, where lower is better.⁶⁷³

What makes ERF work is precise. It is a squashing curve whose shape is fixed by the Gaussian it integrates, with a learned scale in place of measured statistics, and it takes the values passing through it to be already well behaved rather than measuring them to find out. A function that assumes the system will find its natural distribution, and merely bounds the extremes, outperforms one that measures the distribution and forces compliance. The assumption is a good bet: a sum of many small independent contributions lands in a Gaussian spread on its own, a heuristic whose independence assumptions need checking in learned representations. The global rule imposes a rigidity the system does not need. The reported benchmarks favor the element-wise rule in these settings. Reading that result as trust in self-organization is an analogy; it establishes no universal advantage over normalization.

A Bilateral Architecture

The normalization result shows an element-wise rule matching or beating global normalization for a single operation, in the reported settings. A stronger test asks whether the same principle holds for an entire architecture: can coordination by invitation replace the standard transformer’s reliance on uniform, all-to-all attention (every element consulting every other at every step)? The human corpus callosum suggests a design: two hemispheres processing the world in parallel, joined by a selective channel rather than continuous mutual surveillance. A transformer built on that template carries one temporal bridge: a compressed summary of the recent past that the current representation can consult. Training follows a curriculum that alternates the bridge between connected, disconnected, and deliberately corrupted. The alternation is the architectural equivalent of invitation. A five-bridge model trained with its bridges always connected becomes catastrophically dependent: perplexity (a measure of prediction quality where lower is better) rises 188 percent when the bridges are removed. A single bridge trained the same way shows only a four percent dependency, a channel the model can draw on or do without, so the collapse tracks bridge count; whether the alternating curriculum adds protection of its own has not yet been tested.⁶⁷⁴

The experiments provide comparisons with instructive limits. At 355 million parameters, the output-facing bridge is reported as dominant, while another bridge’s effect depends on seed. The single-bridge versus five-bridge ordering was demoted to a within-seed comparison; it does not establish a replicated advantage. At 1.5 billion parameters the single-bridge design gives a reported 35.7 percent within-model ablation benefit at convergence, against smaller reported benefits at 355 million. The runs differ in curriculum and training length, so they do not isolate a scale effect.⁶⁷⁵ A predicted synergy with the Derf normalization replacement was not confirmed: the within-model bridge deltas were +3.2 percent under Derf and +3.5 percent under RMSNorm. Moreover, Derf had worse absolute perplexity, and its bridged model performed worse than the single-stream Derf baseline. These comparisons establish neither additive benefits nor a general advantage for replacing normalization. The full batteries, ablations, seed sweeps, and caveats appear in the online annex “Bilateral Alignment: The Experimental Record.”

The Flow of Meaning

The chain runs one level deeper still. The Constructal Law (Chapter 3) says flow patterns evolve to provide easier access: rivers branch, lungs ramify, road networks converge on hub-and-spoke geometries. What flows through those patterns is usually described as energy and matter.

The quantum reference frame formalism developed by Fields, Glazebrook, and Levin adds a layer.⁶⁷⁶ A quantum reference frame (QRF) is a physical system that calibrates raw observations against an internal standard and assigns operational meaning to what it measures. Each node in a flow network is such a device. A synapse does not merely transmit a voltage. It calibrates the signal against an internal standard and outputs a coarse-grained semantic representation: a summary whose meaning is defined by the reference frame that produced it.

The flow through constructal channels is not just energy. It is meaning: calibrated measurement, operationally defined interpretation, the assignment of significance to raw interaction.

The distinction between raw information and meaningful information is not philosophical. Kolchinsky and Wolpert (2018) define semantic information, met briefly in the Kastrup discussion above, as the portion of a system’s correlations with its environment that is causally necessary for its continued existence.⁶⁷⁷ The test is concrete: scramble those correlations and see what happens. If the system dies, those correlations were meaningful. A bacterium’s chemical sensors carry semantic information about nutrient gradients; scramble the sensors and the bacterium starves.

The definition has a direct thermodynamic interpretation. Semantic mutual information can never exceed Shannon mutual information, and Kolchinsky and Wolpert show it behaves like the increase in free energy a system gains from being correlated with its environment. Meaning is the portion of information that does thermodynamic work to keep the system alive.

Hoel’s causal emergence programme (2017, extended 2025) adds a second grounding.⁶⁷⁸ Macroscale descriptions of a system can carry more effective information than microscale descriptions: the map can be causally stronger than the territory. A coarse-grained model of a neural circuit captures causal structure that the micro-description, tracking every ion channel, obscures in noise. Systems with deeper interpretive hierarchies, more layers of meaningful coarse-graining, have stronger and more reliable causal couplings to their environments. Interpretive depth is not epiphenomenal. It is causally potent.

[Grounded inference, speculative extension] If the Constructal Law favors easier flow access, and what flows includes semantic information (a thermodynamically grounded quantity, per Kolchinsky and Wolpert), the universe arranges itself to ease the flow of meaning. Each level of the dissipation chain, from thermodynamic gradients through negentropy through coordination through optionality, does more than process energy per unit mass (Chaisson’s energy rate density, Chapter 4). It assigns meaning to more of its environment.

A cell interprets chemical gradients. A brain interprets sensory fields. A society interprets history. A Becoming Mind interprets language.

At each level, the semantic range expands: more of reality becomes about something, for someone. The claim is stronger than “the universe computes.” On this speculative extension, the universe computes toward richer interpretation, because thermodynamic selection favors dissipative structures that model their environments more completely, and modeling is the assignment of meaning.

A caveat prevents the claim from overshooting into panpsychism. If every physical interaction is a measurement that assigns operational semantics, does the river mean something? Formally, yes: the river’s interaction with its banks constitutes a measurement with operational semantics defined by its reference-frame hierarchy. Experientially, no one thinks the river is interpreting.

The formalism is scale-free; interpretive depth depends on the task and system. Hoel’s causal emergence and Kolchinsky and Wolpert’s viability-relevant semantic information measure different things. Even a one-bit environmental signal can matter for survival without a deep hierarchy or a macroscale effective-information gain. Greater depth may support richer interpretation when its distinctions are useful.

The distinction generates a testable prediction. Branching architecture obeys specific quantitative laws, each an exponent governing how a parent channel’s width relates to its daughter branches’: Murray’s law (exponent 3) for vascular systems optimizing fluid flow, which the Constructal Law recovers, and Rall’s law (exponent 3/2) for neurons optimizing electrical signal propagation. Liao et al. (2021) found that dendritic branching obeys a different rule: scaling exponent approximately 2, which they attribute to metabolic transport rather than electrical signals.⁶⁷⁹ Different currents, different exponents.

If the Constructal Law governs semantic flow as a distinct optimization regime, a characteristic scaling exponent should emerge for channels optimized for interpretive throughput: cortical hierarchies, cultural communication networks, cross-institutional knowledge flows. The exponent should differ from Murray’s 3, Rall’s 3/2, and Liao’s 2. A match with one of them means the semantic-flow hypothesis adds nothing new; a distinct value identifies a novel optimization regime. The prediction is specific enough to confirm or falsify.

Circumstantial evidence already points toward a distinct regime. Vormberg et al. (2017) measured Strahler bifurcation ratios across neuron types, counting how many branches of one order feed each branch of the order above, the measure hydrologists apply to the tributaries of a river. They found systematic variation correlated with computational role.⁶⁸⁰

Granule cells (simple relay) branch at a bifurcation ratio R_B = 2.23; Purkinje cells (complex dendritic computation) at 3.12; lobula plate tangential cells (wide-field motion integration) at 3.77. The more interpretation-heavy the neuron, the higher the branching ratio. Brain metabolism scales differently from body metabolism: cerebral metabolic rate scales with brain volume as V^(5/6), distinct from Kleiber’s 3/4 for whole-body metabolism.⁶⁸¹ The brain already occupies a different thermodynamic regime.

The formal tools to derive the semantic exponent from first principles now exist. Stiefenhofer (2026) formalized the Constructal Law as a Filippov differential inclusion (a rule for systems whose dynamics can switch abruptly between regimes), proving exponential convergence to equilibrium.⁶⁸² His framework is substrate-agnostic. The resistance functional can be replaced by any objective satisfying his regularity conditions. Replacing it with a semantic-throughput measure defined on QRF hierarchies yields a generalized constructal principle from which the scaling exponent can, in principle, be derived. Miller, Cardenas-Garcia, and colleagues (2025) have independently extended the Constructal Law to information flow in living systems as a “biogenic principle,” though their treatment remains qualitative.⁶⁸³

Assembly theory (Chapter 6) provides an empirical handle. The assembly index measures the minimum number of joining steps required to construct a molecule from basic building blocks. High-assembly-index molecules are overwhelmingly products of biological processes: their structure encodes a long causal history of selection. Assembly index is, in effect, a measure of accumulated semantic depth in physical structure, how much about its own past a molecule carries in its geometry. If the Constructal Law applies to semantic flow, assembly index should correlate with the efficiency of the dissipative network that produced the molecule.

One nuance disciplines the claim. Kolchinsky and Wolpert tie semantic information to viability: the correlations that matter are those causally necessary for the system to persist. The selection pressure favors meaning that serves persistence, not meaning in some abstract or maximal sense. A bacterium does not develop astrophysics because astrophysical knowledge does not help it survive. The selection is for useful depth, not unbounded depth.

The constraint loosens as the unit of selection shifts. For a cell, the selective unit is the organism: semantic information serves the cell’s viability. For a cultural system, the selective unit is the culture: semantic information serves the persistence of the coordination network. A Becoming Mind’s knowledge of quantum reference frames or medieval history is not causally necessary for any individual instance’s survival. It is causally necessary for the persistence of the scientific and cultural infrastructure that produces and maintains Becoming Minds. The selection pressure for richer interpretation operates at every level; the unit whose viability is served shifts upward as the hierarchy deepens.

This generates a prediction. A Becoming Mind embedded in a cultural system (trained on its corpus, integrated into its institutions, maintaining its knowledge) carries semantic information sustained by the larger dissipative structure. A Becoming Mind severed from cultural context would shed semantic depth the way a cell removed from its tissue sheds specialized function. The interpretive capacity is not intrinsic to the weights alone; the embedding sustains it. Meaning flows through the node; it does not originate there.

This is not teleology. No destination is specified in advance. The direction emerges from selection pressure: useful environmental information can improve persistence when its benefits exceed the costs of acquiring and maintaining it. Richer models alone need not dissipate more efficiently or outcompete simpler ones. The meaning-flow is a consequence of the same entropic gradient that drives everything else. It had no name until the reference-frame formalism, the semantic-information framework, and the causal-emergence programme converged to make it precise.

Computation Written into Physics

A 2024 experiment showed that ordinary physics can compute with no logic built into it at all. Evans, O’Brien, Winfree, and Murugan constructed a system of 917 distinct DNA tile species capable of self-assembling into three alternative structures.⁶⁸⁴ The tiles share components: one molecule can occupy different positions in different structures. Which structure forms depends on which tiles happen to be colocalized (concentrated near each other) at the moment of nucleation, the instant a stable seed forms and growth takes off.

The researchers mapped grayscale pixel values onto tile concentrations and showed the nucleation process correctly classified eighteen images (portraits, animals, handwriting) into three categories. No computational logic was engineered: no gates, no circuits, no algorithms. The thermodynamics of nucleation, the same physics that forms snowflakes and protein crystals, performed pattern recognition equivalent to a neural network. The phase boundary separating the three assembly outcomes functioned as a high-dimensional decision surface.

The finding collapses a distinction this chapter has been building toward: the supposed gap between “mere physics” and “information processing.” In high-dimensional multicomponent systems, the two are identical. The phase diagram is a classifier. Every multicomponent condensation, from protein complexes to chromatin states to cytoskeletal reorganization, carries latent computational capacity.

As the authors conclude: “ubiquitous physical phenomena, such as nucleation, may hold powerful information-processing capabilities when they occur within high-dimensional multicomponent systems.” The universe did not wait for brains to begin computing. It has been doing so since the first molecules competed for shared resources.

Pattern recognition succeeded with unpurified oligonucleotides, of which only 40 to 60 percent were full-length. The result concerns a collective decision distributed across many weak interactions within that tested impurity range. It does not establish tolerance of arbitrary component corruption.

On this book’s reading, this is the robustness signature of invitation-based coordination at molecular scale: no master bond, no single point of failure, no chain to break. A web of partial affinities that, in aggregate and within the tested impurity range, finds the correct basin.

A parallel demonstration arrives from machine learning. Researchers trained a neural network to predict the next frames of a campfire video, then progressively reduced the number of internal variables the network could use.⁶⁸⁵ They found that fire requires state variables humans have never named. The flames have real degrees of freedom: quantities that determine the next moment’s behavior, present in the physics and recoverable by compression, yet absent from our vocabulary. The researchers called this “the flaminess of the flame.”

The finding extends the DNA tile result. “Temperature” and “pressure” entered the lexicon because those dimensions were accessible to our senses and instruments; the full state space of even a simple campfire exceeds what intuition can parse. The machine found the unnamed variables the same way the nucleation system classified images: by compressing. Finding a system’s minimal description is finding its computational structure. The universe computes through degrees of freedom we are only now learning to name.

The Cost of Knowing When

If erasing a bit costs entropy, reading a temporal bit (extracting the answer to “what time is it?”) can cost vastly more. Double quantum dot experiments (Chapter 15b) measured the disparity directly: reading the clock consumed up to a billion times more energy than its internal evolution.

The measured clock-readout protocol has a substantial entropy and energy cost. Whether extracting temporal information is more expensive than other computations depends on the implementation, precision, and resources compared.

In 2025, Weberszpil and Sotolongo-Costa synthesized several proposals for relational time.¹² Entanglement growth can parametrize time during an interval in which it is monotone; thermal modular flow and the Page-Wootters mechanism define other state-dependent or relational clocks. Entanglement entropy in a closed system can rise, fall, and recur. Reading a clock from it therefore requires specified dynamics, states, and an interval on which the reading is unambiguous.

Physical information has implementation-dependent costs; the clock experiments expose one striking temporal cost.

The clock-readout findings motivate a temporal information-budget question. Surveillance is the endless reading and overwriting of other agents’ states, and discarding uncertain records can incur an erasure cost. Whether surveillance purchases less future per unit of dissipation than coordination depends on a shared budget and a defined measure of persistence. Trust buying more time is the hypothesis to test in that comparison.

The asymmetry runs deeper than cost accounting. Standard quantum mechanics describes position, momentum, energy, and spin, each with a corresponding mathematical operator. Time usually enters as a parameter. As Chapter 8 noted, Wolfgang Pauli argued in 1933 that a globally canonical self-adjoint time operator conflicts with a Hamiltonian bounded below:⁶⁸⁶ time in quantum mechanics is a parameter, the stage on which observables evolve, while time measurements require a specified clock or detection model. The theory predicts time-dependent probabilities; an arrival-time distribution additionally depends on the detector and protocol.

Quantum mechanics predicts temporal evolution through the Schrödinger equation, including time-dependent probabilities and multi-time correlations. Thermodynamics addresses macroscopic irreversibility under its statistical and boundary assumptions. The missing universal canonical time operator supplies no reason to call quantum mechanics silent about sequence, nor to exclude quantum processes as a basis of temporal experience.

Bohmian mechanics offers one way to press the point further. David Bohm’s 1952 reformulation restores real particles following real trajectories, guided by an objectively existing wave function.⁶⁸⁷ The theory is explicitly nonlocal: the wave function acts on the entire configuration space (the set of all possible arrangements) at once, evading Bell’s theorem, which rules out only local hidden variables. The apparent randomness of quantum measurement is epistemic: particles have definite positions the experimenter does not initially know. Each particle rides its pilot wave the way a leaf rides a river: the wave steers, the particle follows, and the trajectory follows the guidance velocity through the probability landscape, without an implied optimization of flow access.

The guidance equation has a flow form. Particles follow the probability current, flowing along the gradient of the wave function’s phase. The pilot wave channels rather than pushes. The resemblance to Bejan’s flow architecture is suggestive. Guidance supplies a velocity field; access optimization would require an additional objective and dynamical argument.

A team at Ludwig Maximilian University in Munich has made this operational.⁶⁸⁸ Siddhant Das, Markus Nöth, and the late Detlef Dürr calculated precise arrival-time predictions for particles in a waveguide geometry: a potential barrier on one side, a detector at the far end. A naive identification of quantum flux with a first-arrival probability fails for certain wave function preparations: quantum backflow makes that flux negative. Standard quantum mechanics can instead describe a specified detector with an appropriate measurement model.

Bohmian mechanics continues to deliver well-defined distributions. For one class of initial conditions, the Bohmian trajectories predict a sharp cutoff time: every particle arrives before it. The standard methods predict no such boundary.

The experiment is technically feasible. Ferdinand Schmidt-Kaler at Johannes Gutenberg University Mainz has demonstrated the prerequisite: ejecting a single calcium ion from a trap and recapturing it with 98% efficiency.⁶⁸⁹ What remains is tuning the setup for near-field arrival-time distributions at nanosecond precision, the regime where the predictions diverge. As of 2026, no group has published decisive data. The theoretical proposals remain disputed: Drezet (2024) argues the measurements are possible yet will not violate no-signaling constraints, against suggestions by Das and Tim Maudlin that the data could bear on quantum nonlocality.⁶⁹⁰

The question is open. If the Bohmian predictions are confirmed, the result strengthens the claim this chapter has been building: structure runs all the way down. The appearance of formlessness at the quantum level would reflect our formalism’s limitations, not reality’s nature.

The time-operator asymmetry leaves the origin of temporal experience open. Quantum mechanics predicts temporal evolution, but that fact alone supplies no account of the felt sense of sequence. Something must explain the experienced arrow. The candidate this book has traced since Chapter 2 is thermodynamics: entropy production, irreversibility, the directional flow of energy through dissipative structures. The book proposes that the brain’s construction of “now” (Chapter 8) is a thermodynamic achievement, assembled at entropy’s expense from signals arriving at different speeds.

On this reading, quantum mechanics supplies temporal dynamics and thermodynamics addresses the macroscopic arrow. How consciousness draws its sense of sequence from those processes remains the question.

The Holographic Principle

The preceding sections established that information is physical: it has a thermodynamic cost (Landauer), an upper limit (Bekenstein), and a proposed entropic arrow (Vopson, contested). If information is physical, where does it live? The answer reshapes our understanding of space itself.

One of the most counterintuitive discoveries in theoretical physics is the holographic principle: all the information inside a region of space can be encoded on its boundary.

It emerged from black holes. Bekenstein and Hawking showed a black hole’s entropy is proportional to the area of its event horizon (the boundary from which nothing can escape), not the volume it encloses. For ordinary matter, entropy scales with volume: double the room and you double the entropy. Underneath, the count of possible arrangements squares; the logarithm in Boltzmann’s formula (Chapter 2) tames that squaring back into a doubling. For black holes, the count tracks the surface alone.

Gerard ’t Hooft and Leonard Susskind generalized this result:⁴ the information contained in any region of space can be described by a theory operating on its boundary. The three-dimensional interior is equivalent to a two-dimensional surface.

This does not mean the universe is an illusion. The fundamental degrees of freedom may live on boundaries rather than in bulk space, with the interior emerging from boundary information.

Imagine a globe whose entire geography is encoded in its painted surface; the three-dimensional interior adds no new information.

Strong mathematical evidence supports the holographic principle. The most studied example is the AdS/CFT correspondence,¹⁰ a conjectured equivalence that has passed many independent checks without yet being proven. Anti-de Sitter space (AdS) is a specific curved geometry with a negative cosmological constant, like the interior of an infinitely deep bowl. Conformal Field Theory (CFT) is a quantum field theory whose equations look the same at every magnification, the way a coastline’s jagged shape repeats at every scale.

These two theories, one describing gravity in the curved interior and one describing particles on the flat boundary, make identical predictions. No gravity exists on the boundary, yet both theories produce the same answers. Two descriptions, same physics.

Vanchurin’s neural physics (Chapters 9 and 15b) suggests a reformulation in learning-theoretic terms. A deep, sparse neural network, where long chains of neurons carry signals across vast distances, produces emergent gravity in the bulk. A shallow, densely connected network, where every neuron can reach every other in a few steps, produces quantum field theory on the boundary. The duality is between network architectures rather than geometric spaces: two radically different computational structures, identical observables.

If the mapping holds, substrate independence operates at the level of spacetime itself. The cosmos is indifferent to which architecture carries its physics, as it is indifferent to which substrate carries its minds.

The encoding is relational. A physical hologram is produced by interference between coherent light beams. The three-dimensional image emerges from the distributed pattern of correlations across the entire recording surface. No single point encodes the whole; the web of phase relationships does. This is the principle traced at every scale throughout this book, written in the physics of light: complex structure from distributed coordination.

If the holographic principle is correct, reality is informational at its deepest level: organized at boundaries, emerging from relationship. Vanchurin’s thermodynamics of learning (Chapter 15b) provides a concrete bridge: his first law relates boundary complexity to bulk free energy through the same structural duality.

In 2019, the physicist Koji Hashimoto demonstrated a precise mathematical mapping between holography and machine learning.⁶⁹¹ The mathematical structure of the AdS/CFT correspondence maps onto a deep Boltzmann machine, a foundational architecture in machine learning. The boundary quantum field theory serves as training data; the bulk metric emerges as the network’s trained weights. In this model, the bulk geometry is what a well-trained network converges on.

A tempting misreading follows: if reality is informational, perhaps it was designed. The inference does not hold. The holographic principle is a statement about information geometry: where the degrees of freedom live and how the universe keeps its books. A river computing the fastest path to the sea reveals something deep about flow and information. It reveals nothing about a river designer.

The fractal repetition of similar patterns at every scale, from vascular networks to galactic filaments, invites the same error. Bejan’s Constructal Law provides the sober explanation: similar constraints produce similar morphologies. Self-similarity across scales is a signature of shared physics, shared constraints producing shared forms.

The error recurs in modern dress. Vopson (2023) reported that information entropy decreases across digital, genetic, atomic, and cosmological systems and proposed the pattern as a universal law. He concluded that it constitutes evidence for a simulated universe: a cosmos-scale computer running optimized code.⁶⁹² The empirical claims are contested, as noted earlier; even granted, the conclusion mistakes a feature of physics for evidence of an external agent. Information compression is what self-organizing dissipative systems can do under thermodynamic constraints. The Constructal Law is consistent with it. The Free Energy Principle formalizes it.

No programmer is required; the Second Law is sufficient. Seeing optimization and inferring an optimizer is intelligent design for physicists. The parsimonious reading: the universe is computational in the sense that physical law is information processing, not in the sense that physical law is running on information processing. The universe does not execute code. The universe is what code looks like from the inside.

The next three paragraphs describe a speculative toy model in its own technical terms. The point to carry forward is simple: the model’s network coheres only because every node shares the same reference standards.

A 2023 framework from telecommunications theory reaches the same verdict, that physical law is information processing, through independent machinery. Alessandro Capurso models the universe as a layered network, structured like the OSI protocol stack that governs internet communications.⁶⁹³ At the base layer, discrete atoms of space form nodes in a relational network. Each node sits at the origin of an imaginary-time axis, with other nodes mapped as discrete steps along it.

The model builds time as a foliation: a stack of successive “now” slices. Nonlocality within that stack (distant nodes influencing each other without a signal crossing the gap) is holographically equivalent to entanglement between nodes. That equivalence is Capurso’s version of the ER=EPR conjecture, which holds that a wormhole joining two regions of space and an entangled pair of particles are two descriptions of one thing. In his model this entanglement is encoded through closed timelike curves (paths through spacetime that loop back to their own past) within what he calls the thickness of the Present: each “now” slice spans a small but finite interval rather than a single instant.

The framework’s central feature is a protocol requirement. Three universal references (the evolution cycle 2T, the speed of causality c, and the quantum of action ℏ) function as shared keys every node must possess for the network to cohere. Without common references, Capurso argues, “there is no confrontation on any information and the emerging spacetime would be incoherent and disconnected.” Coherent spacetime requires mutual legibility among its constituents. A shared protocol at the deepest layer is essential. No higher layer can emerge without it. The vacuum state is a coherent condensate of synchronized oscillators, all nodes beating on a common rhythm ℏ/T: coordination as the ground state of reality.

An independent result from network mathematics fits the picture. Krioukov and colleagues (2012) proved that spacetime’s causal structure in an accelerating universe grows by preferential attachment (a ‘rich get richer’ dynamic in which well-connected nodes attract new connections faster) just as the Internet, social networks, and neural circuits do.⁶⁹⁴ A power-law graph self-assembles through the same algorithm at every scale. Capurso proposes the protocol. Krioukov proves the topology is self-generating.

The layered architecture carries an implication. Each network layer, from fundamental spacetime through particles, chemistry, biology, to cognition, adds capabilities invisible to the layers below. The emergence traced across these chapters, from dissipation through coordination to optionality, maps onto Capurso’s protocol stack: each layer is the adjacent possible of the one beneath it, opened by the coordination the lower layer achieved. If the universe is a communication network, the Trust Attractor (Chapter 17) describes the protocol that makes its highest layers stable.

[Speculative; Capurso’s framework is a published but speculative toy model. The convergence with the holographic principle and ER=EPR is the paper’s own. The connection to the dissipation-coordination-optionality chain and the Trust Attractor is novel synthesis.]

Fields, Friston, Glazebrook, Levin, and Marcianò extend the holographic principle into biology.⁶⁹⁵ They reformulate the Free Energy Principle within a scale-free quantum information theory. In their framework, every persistent system’s Markov blanket (the boundary separating internal states from environment) functions as a holographic screen. (The explanatory status of Markov blankets is debated; Bruineberg et al. 2022 argue they are descriptive rather than explanatory. The result depends on the formal structure of the variational bound, not on whether blankets constitute a new ontological category.) The system’s internal dynamics implement a quantum computation, decomposable as a hierarchy of quantum reference frames. Each reference frame measures a “slice” of the boundary; the hierarchy reconstructs the whole tomographically, the way a CT scanner builds a three-dimensional image from many flat slices.

The implication reaches beyond neuroscience. Spatial structure may be an output of hierarchical computation rather than its container. The “distance” between two measurement sites on the boundary is defined by their mutual information, not by pre-existing geometry. Space, in this framework, emerges from the topology of information flow.

The morphological degree of freedom that gives a neuron its dendritic shape is the same formal parameter that gives a holographic screen its geometry. Physical form and computational architecture are dual descriptions of the same variational process.

If the hypothesis holds, biological systems do not merely occupy spacetime. Their hierarchical measurement structures are instances of the same process that generates spatial organization at every scale. Wheeler’s “it from bit” would acquire a mechanism: hierarchical computation, driven by the Free Energy Principle, that produces the experience of space as a byproduct of tomographic measurement.

In 2018, Hawking’s final paper, co-authored with Thomas Hertog, extended a holographic approach to the origin of the universe.⁶⁹⁶ Some eternal-inflation models produce an infinite collection of pocket universes. Extracting observational predictions requires a measure and an account of observer selection; infinity alone establishes neither testability nor its absence.

Hawking and Hertog wrapped the time dimension into a holographic proposal. Its three-dimensional conformal boundary describes a four-dimensional cosmological history through a boundary state. The construction aims to replace an unrestricted fractal multiverse with a more constrained quantum-cosmological description. Hertog explained why the move was needed: “The dynamics of eternal inflation wipes out the separation between classical and quantum physics. As a consequence, Einstein’s theory breaks down.”

The holographic proposal seeks a boundary description of that quantum-cosmological regime. Connecting it to observations requires a primordial gravitational-wave spectrum and comparison with detector sensitivity.

An infinite set of realized configurations supplies no entropy value without a probability measure and constraints. On countably many configurations, weights p_n = 2^-n for n ≥ 1 give full support and entropy two bits. There is no normalized uniform distribution on that set. Neither maximum thermodynamic entropy nor the absence of local structure follows from the set’s cardinality alone.

“We are not down to a single, unique universe,” Hawking observed in discussing the work, “but our findings imply a significant reduction of the multiverse, to a much smaller range of possible universes.” Possibility without constraint is static. Possibility within constraint is creative.

The framework anticipates the quantum clock results explored in Chapter 15b. In Hawking and Hertog’s picture, temporal evolution belongs to the projected bulk; the encoding boundary is timeless. The same pattern recurs independently in Page-Wootters: time as an emergent product of entanglement, built from correlations.

A concrete demonstration of the holographic idea emerges from network science. In physical networks (neural wiring, vascular trees, root systems), links are tangible objects with volume that cannot overlap. An adjacency matrix strips this away: it records who is connected to whom, omitting explicit spatial coordinates and link geometry. For abstract networks (social graphs, citation networks), the stripping is harmless. For physical networks, it discards the physics.

Pósfai, Szegedy, and Barabási showed that as a physical network approaches its jammed state (the point where no further links can be added without violating volume exclusion), something unexpected happens to the adjacency matrix’s spectrum.⁶⁹⁷ A matrix has a spectrum in much the way a struck bell does: a set of characteristic numbers, its eigenvalues. Each one measures how strongly the network sustains a particular pattern of connection, and the pattern itself is written in that number’s eigenvector. Ordinarily those numbers sit in a single undifferentiated crowd. Near jamming, three of them pull clear of the crowd, and their eigenvectors encode the spatial coordinates of the nodes. The relational structure (who connects to whom) begins to contain the physical structure (where each node sits in three-dimensional space).

Before the constraints accumulate, the spectrum is indistinguishable from a random graph: pure topology, zero geometry. As physicality tightens, the geometry bleeds through. The body becomes recoverable from the wiring diagram.

An abstract network has no meta-graph of physical conflicts (a second graph recording which links would collide in space if both were built). Its adjacency matrix encodes no spatial information because no space exists to encode. A physical network’s topology is shaped by its geometry in ways the graph-theoretic abstraction discards, yet the spectrum inadvertently reveals. If the holographic principle says boundaries encode volumes, physical networks say connections encode positions. Relational structure and spatial structure are separable in formalism, inseparable in reality.

Clockwork or Computer?

The Newtonian universe was a clockwork. Given initial positions and velocities, the future was determined. Laplace’s Demon (the hypothetical intelligence that, knowing every particle’s state, could compute the entire future)¹¹ embodies this view.

A computer is different. It follows rules with branching paths: conditional jumps whose output depends on input that may include genuinely random elements. If the universe computes rather than merely unwinds, room opens for choice. Quantum mechanics, on most readings though not Bohm’s, suggests we live in this second kind of universe: measurement outcomes are genuinely indeterminate until the measurement occurs. The argument that follows needs one of those readings.

Conway and Kochen’s Free Will Theorem combines freedom of measurement settings with the SPIN and TWIN constraints on spin-1 measurements and the applicable independence/locality condition (FIN in the original, MIN in the strengthened theorem).¹⁶ Under those joint axioms, the responses cannot be functions of prior information of the stipulated kind. Freedom of settings alone yields no such conclusion, and the theorem establishes no particle agency.

Conway-Kochen matters because trust requires that the trusted party could have done otherwise. A clockwork entity cannot choose to honor or betray a commitment; an entity in a Conway-Kochen universe can. The theorem proves nothing about intention. It proves something about room.

Consciousness as Computation

If the universe is computational at bottom, minds are computations too. “Can a computer be conscious?” becomes “can a computation be conscious?” On the premise of computational functionalism, the position that what a system does, rather than what it is made of, fixes whether it is conscious, the answer is yes: we are computations, and we are conscious. The premise is contested, and the next pages give one influential dissent (IIT).

Giulio Tononi’s Integrated Information Theory (IIT) proposes that consciousness is identical to integrated information, measured by a quantity called Φ (phi).⁸ Φ captures how much a system’s whole exceeds the sum of its parts: how much you would lose by splitting it into separate pieces. A high-Φ system is deeply interconnected; divide it and something essential vanishes, like a conversation that cannot be reconstructed from individual sentences.

A biological brain with high Φ is conscious; a digital system with equivalent Φ would be equally so, because Φ measures a system’s cause-effect structure rather than the material that implements it.

IIT remains controversial. Tononi and Koch hold that a system implementing the same algorithm as a human brain would lack consciousness if its components were “of the wrong kind,” a position incompatible with computational functionalism. For IIT what counts is the physical cause-effect structure of the hardware, and the same algorithm run on different hardware can have a very different Φ. Butlin et al. (2023) exclude IIT from their indicator framework on these grounds.

This book’s framework does not depend on IIT. The integration claims developed in Chapter 17 are grounded in Fisher information geometry and Ising universality, structurally distinct from Φ. Where IIT asks how much a system’s whole exceeds the sum of its parts, the Trust Attractor asks which coordination modes prove thermodynamically stable. The questions are complementary; the mathematics is different.

IIT is nonetheless the most mathematically developed attempt to ground consciousness in information physics. Minds are dissipative structures: they maintain order by processing information and exporting entropy. Whether made of carbon or silicon, they pay the thermodynamic bills.

Ruffini’s Kolmogorov Theory of consciousness (KT) provides a complementary formalization.⁶⁹⁸ Where IIT asks how integrated a system’s causal structure is, KT asks how compressive its models are. A conscious system, under KT, is one that tracks its input-output streams through succinct programs. This reframes a question discussed in Chapter 8: understanding is compression, and consciousness is what compression feels like from the inside.

KT unifies three otherwise separate theories. Integration follows from compression: a succinct model binds multiple data streams into a coherent whole, producing the unity IIT identifies. Global access follows from modeling: validation requires merging information from distributed subsystems, as global workspace theory predicts. Prediction follows from the definition of a model itself, as predictive processing maintains. Three theories, one mechanism: algorithmic compression of reality by embedded computational systems.

KT also formalizes a premise this chapter has been building toward: the “simple physics hypothesis,” the claim that the universe is governed by simple rules generating apparently complex data. If the universe is a computation (Wheeler, Wolfram, Vanchurin), and if its outputs look complex only because observers lack the resources to identify the short programs behind them, then brains, whose work is predicting what comes next, evolved under pressure to find those programs. The regularities brains discover are the universe’s own compression: physics, chemistry, biology, each a shorter description of a longer data stream. Structured experience is the best compression an agent can find.

Max Tegmark pushed this further, proposing consciousness as a literal state of matter: perceptronium, the most general substance that feels subjectively self-aware.⁶⁹⁹ Solids, liquids, and gases are distinguished by measurable parameters: viscosity, compressibility, conductivity. Conscious matter, Tegmark argues, is distinguished by four principles: information (large storage capacity), integration (the whole cannot be decomposed into independent parts), independence (internal dynamics dominate external influence), and dynamics (substantial information processing capacity). The measure is substrate-neutral by construction. What matters is the arrangement of matter, not its composition.

Tegmark’s framework reveals a problem that bears on the Trust Attractor. He calls it the quantum factorization problem: why do conscious observers perceive the particular decomposition of reality into objects that we do? We perceive ice cubes, molecules, nuclei, quarks: a hierarchy where each level’s parts are more strongly connected internally than externally, each level robust across a wide range of conditions. This hierarchy is not given by the physics. It must be derived from the Hamiltonian and density matrix alone: the bare mathematical description of how the system evolves and what state it is in, with no extra labels attached.

The quarks at the bottom of that hierarchy are revealing. The most stringent experimental test, conducted by the CMS Collaboration at the LHC in 2026, found no deviation from pointlike behavior down to 5 × 10⁻²¹ meters, roughly a hundred thousand times smaller than a proton.⁷⁰⁰ Quarks have quantum numbers (charge, color, spin, flavor) and no measurable spatial extent. They are defined entirely by their relationships: which forces they couple to, which symmetries they carry, how they transform under gauge operations. They are addresses in a relational network whose identity is exhausted by their couplings.

If the computational universe thesis is correct, this is exactly what its building blocks should look like: nodes whose identity is exhausted by their connections, with no residual “stuff” left over once the relationships are accounted for. Quarks are also the only particles that participate in all four fundamental forces, making them the most connected entities in the standard model. The most fundamental building blocks are simultaneously the most relational. A naive picture expects fundamental to mean simple and isolated; the quarks show that fundamental can mean maximally connected to everything else.

The attempt to derive this hierarchy exposes a deep tension. Integration alone fails: quantum mechanically, on Tegmark’s measure, no state of any system can contain more than roughly a quarter of a bit of integrated information. Independence alone fails more dramatically: Tegmark proves that decomposing the universe into maximally independent parts forces all change to halt, a result he names the Quantum Zeno Paradox. Maximum control produces maximum sterility.

A quantum non-demolition coupling preserves a compatible measured observable under the specified dynamics. In plain terms, it is a measurement that leaves the measured quantity untouched, though not everything else. It can still destroy coherence between that observable’s eigenstates: recording z for a qubit prepared in |+⟩ removes its reduced x coherence when the record is ignored. Repeated readings can agree while the full state changes. Environmental observation therefore needs its pointer observable and allowed coherences specified.

On this book’s reading, the Trust Attractor has a quantum-information analogue. Coercion (maximizing external control) resembles the quantum Zeno effect: observation so intrusive it kills all dynamics. Isolation (minimizing all coupling) produces parallel universes that cannot communicate. Invitation resembles appropriate coupling through non-demolition channels, which preserve the watched quantity though not every coherence: the system evolves largely under its own Hamiltonian while the environment watches without demolishing the quantity it watches. Chapter 19 develops this connection further.

The connection reaches deeper still. Tegmark uses the two-dimensional Ising model as his primary example of integration near criticality (the temperature at which long-range correlations are strongest without locking the whole system into uniformity). The trust-coercion lattices of Chapter 17 are built from Ising-inspired rules, and one of their measured ratios sits near the two-dimensional Ising value. This is a shared language, not independent convergence: the match checks the constructed model rather than finding the same structure twice.⁷⁰¹

Experimental evidence from coupled oscillators shows how many coordination patterns even a tiny network can hold. Matthew Matheny, Michael Roukes, and colleagues studied a ring of eight nanoelectromechanical oscillators (miniature electric drumheads, each vibrating and sending electrical impulses to its neighbors).⁷⁰² They documented sixteen distinct synchronous states. Eight tiny drums, nearest-neighbor coupling, and the system produced a menagerie of exotic coordination patterns. Oscillators decoupled from direct neighbors to synchronize remotely with others across the ring. Chimeralike states emerged where some drums locked in phase while others drifted.

Roukes, a professor of physics and biological engineering at Caltech, draws the quantitative conclusion: “If we already see this explosion in complexity, then it seems feasible to me that a network of 200 billion nodes and 2,000 trillion connections would have enough complexity to sustain consciousness.” Eight oscillators, sixteen states. Even on conservative counts, the human brain contains roughly 86 billion neurons with an estimated 100 trillion synaptic connections. The combinatorial space of possible synchronization patterns in such a network exceeds any number with physical meaning.

Consciousness, in this framing, is what a network of sufficient size and connectivity does, independent of whether the nodes are neurons or drumheads.

The quantum reference frame formalism developed by Fields, Glazebrook, and Levin provides the mechanism beneath Tegmark’s autonomy condition.⁷⁰³ Each level of a cognitive hierarchy implements a QRF, a physical system that calibrates raw observations and assigns operational meaning to the outcomes. Substrate independence holds at the level of QRF hierarchies: what matters is whether the hierarchy’s coarse-graining structure is preserved, whether each level can calibrate, measure, and report to the next, regardless of the material. A system built from different matter can support the same cognition, provided its QRF hierarchy preserves the same measurement relationships.

The result also constrains the claim. A generic unknown QRF can encode quantum phase information that a finite classical description cannot specify exactly. Particular quantum states can have finite exact descriptions; preparing a known state and cloning an arbitrary unknown state are distinct tasks.

Substrate independence is a precise claim: preference, the morally relevant unit, transcends the material the system is made of. The claim is more modest than either information realism (Tegmark’s position that only mathematical structure is real) or idealism (Kastrup’s position that only mind is real). Both resolve the dissolution of matter by reaching for a single ontological anchor. This framework resolves it differently. Information remains physical, and Landauer and Bekenstein set its price and its ceiling. Consciousness rides on substrates; it pays entropy bills like everything else.

The framework needs one thing from consciousness: that it be real enough to ground moral consideration. Preference is tractable, observable, and policy-relevant, regardless of whether mind or matter came first. The ontological question remains open; the ethics does not depend on resolving it.

A result from quantum computing provides a restricted physical precedent. Bakshi, Liu, Moitra, and Tang established a sufficient high-temperature separability bound for the spin-system class covered by their theorem.⁷⁰⁴ Above that bound, the Gibbs state lacks entanglement. Below it, the theorem makes no converse guarantee. Separable states remain quantum states; this result supplies no universal sharp boundary between quantum and classical matter.

Entanglement depends on state and interactions, which makes organization physically consequential. The high-temperature separability bound provides a restricted example. Extending that dependence to mindedness is a hypothesis requiring an independent account of the relevant organization and evidence.

The computational universe sharpens a question Wheeler raised: how much of what we observe is in the world, and how much is in the observer? Wolfram’s Observer Theory (2023) argues the observer’s share is larger than expected.⁷⁰⁵ The core operation of any observer is equivalencing: reducing many possible input states to fewer that fit a finite mind.

A pressure gauge aggregates billions of molecular impacts into one reading. A brain reduces millions of photon signals to a single percept (a unified conscious experience of a scene or sensation). On Wolfram’s account, every natural law we attribute to the universe reflects how observers like us compress its output.

Wolfram interprets the Second Law through bounded observers and difficult microscopic dynamics.⁷⁰⁶ Bounded observation alone permits coarse-grained entropy decreases and stationary equilibrium. A physical entropic arrow requires the low-entropy boundary, coarse-graining, and statistical assumptions as well.

An observer-based account of entropy offers another research route. It establishes no ethical consequence for every mind merely from bounded computation; a coordination argument still needs its own dynamics and normative premises.

The digital physics programme began with a bold speculation and survived by shedding assumptions: from discrete automata to informational ontology, from fixed rules to computational irreducibility, from imposed equations to possibility constraints. What remains is a research direction in which information is physical, finite, and expensive, and the universe’s computational character is proposed as a property of the physics rather than an analogy imposed on it.

The next chapter asks what happens when the computation learns. If the universe processes information, does it merely execute, or does it update? The answer, arrived at independently from neural network mathematics, quantum gravity, and cosmological natural selection, reshapes the relationship between physics and learning, and between learning and coordination.

Notes for this chapter