editorial: the fork is the unit of progress

the deepest contribution of “it from bit, revisited” is not any answer it favors. it is the proposal that a philosophical disagreement should be converted into a fork that changes what one would measure, build, or expect. this is a useful antidote to consciousness discourse, where different metaphysical packages often circulate as vibes with bibliographies.

the essay also sees something most methodological discussions miss: metaphysical bets can be technologically productive before they are proven. faraday and maxwell did not first obtain a license from the philosophy of science department to believe in fields. a speculative ontology earns credit when it compresses observations, predicts surprises, and makes previously impossible interventions possible. a formal theory of valence that reliably engineered suffering downward would alter the evidential landscape more than another century of intuition pumps.

but there is a trap in moving from “every proposition is true or false” to “every important sentence can be made into a useful yes-or-no question.” the law of excluded middle applies to a proposition $p$: $p \lor \neg p$. it does not guarantee that a sentence containing unstable terms already expresses one proposition. “is consciousness conserved?” looks binary, but until “consciousness” denotes an additive quantity defined across boundaries and times, the sentence is closer to a request for a theory than a question inside one.

this matters because the essay’s ten forks are not ten independent bits.

• questions 3–5 presuppose enough of question 1 to identify phenomenal properties across cases.
• question 6 is not merely monism; it adds an exceptionally strong claim about translation between physical and phenomenal theorem sets.
• question 7 presupposes a measure, additivity, cross-time identity conditions, and a relativistically acceptable way to total it.
• question 8 needs both a physical bound and a psychophysical inequality connecting the bounded quantity to consciousness.
• question 9 conflates the thermodynamic cost of maintaining a physical implementation with the metaphysical cost of instantiating experience.
• question 10 is an inverse problem: without restricting the class of objectives, every actual universe is trivially optimal under some objective.

so the right successor to the essay’s hypercube is a typed dependency graph. each node should state what kind of claim it is, what earlier nodes it assumes, what observation could discriminate its branches, and which words must be operationalized before bivalence applies.

quick rulings on the original ten

synthetic referee a: bivalence must be earned

perspective: logic, philosophy of science, and experimental metaphysics

the essay’s binary method is both its best idea and the place where it cheats most productively. it treats yes-or-no form as though it automatically removes the muddled middle. often it merely moves the muddle into the predicates.

consider the first fork. “does consciousness have a crisp formalism?” can mean at least four things:

1. every experience can be represented by some mathematical object;
2. experiences possess exact internal relations;
3. there is a canonical mathematical representation of those relations;
4. a physical observer can infer that representation from measurements.

the first thesis is close to trivial if arbitrary encodings count. the second is substantive. the third is far stronger because mathematical representations are usually non-unique. the fourth is an epistemic and experimental claim, not an ontological one. a yes to one does not force a yes to the others.

the essay also offers too stark a choice between maxwell and élan vital: either consciousness has an exact latent structure or the concept should be dissolved. mature science contains a large middle kingdom of robust but scale-bound kinds. temperature is multiply realizable and coarse-grained, yet not a linguistic mistake. genes, species, waves, phases, and vortices can be real enough to support counterfactuals and interventions without being fundamental objects in the final microphysical theory. consciousness could likewise contain exact invariants, approximate macroscopic regularities, and culturally assembled report categories at once.

this does not weaken the formalist program. it improves the target. the scientifically valuable question is not merely whether a formalism exists, but whether it discovers projectible invariants: quantities or structures that remain stable under irrelevant changes and support successful interventions in new cases.

the maxwell analogy also needs one demotion. maxwell’s equations are not simply the exact final truth about electromagnetism everywhere. classical electromagnetism sits inside quantum electrodynamics and curved spacetime physics, and its applicability depends on scale and regime. what made it magnificent was not metaphysical finality but extreme compression, precision, unification, and engineering fecundity. consciousness formalism should be held to that standard, which is both more realistic and more demanding.

the essay is right that hypotheses should earn status by what they let us build. but “build cool stuff” is not enough by itself. many incompatible theories can inspire the same useful tool. a neurotechnology that relieves depression would establish a causal handle; it would not by itself establish that the handle tracks the intrinsic structure of valence. the formal theory must make risky, discriminating predictions that nearby rival theories do not inherit.

this problem is acute in consciousness science because predicted experience is compared with experience inferred from report or behavior. kleiner and hoel’s falsification analysis shows how easily the inference procedure and the theory can become either illicitly independent or circularly dependent. a fork is not empirically useful merely because its verbal branches disagree. the branches must constrain different observable distributions without defining the target through the same observations used to test it.

proposed fork hygiene

every fork should carry a small passport:

the kill condition matters. in the essay, a positive test of stv is said to validate many upstream assumptions, while a negative test would reveal only that something somewhere was wrong. this is an asymmetric evidential contract: heads the theory wins, tails an auxiliary loses. it is reasonable during early exploration, but it must expire. each empirical version of stv should freeze enough choices that a negative result can kill that version.

the most important correction is therefore simple: yes-or-no form is the endpoint of clarification, not its beginning. once earned, however, it is potent. a clean fork does not merely divide logical space. it divides labor.

synthetic referee b: structure without a bridge is cheap

perspective: mathematical structuralism, philosophy of physics, and consciousness formalization

the essay’s strongest metaphysical claim is the “meta-theorem of monism”: every true theorem in physics has a corresponding true theorem in phenomenology, and vice versa. this is presented as what follows when physical and phenomenal descriptions are two projections of one underlying reality.

it does not follow.

two projections of one object can discard different information. a tree’s shadow on the ground and its shadow on a wall share a source without sharing a theorem set. one projection may preserve branching order while destroying metric depth; the other may preserve a length relation while merging branches. common origin gives consistency constraints. theorem-by-theorem correspondence requires something closer to an invertible duality.

let the underlying reality be $r$, with a physical projection $p(r)$ and a phenomenal projection $q(r)$. monism says, roughly, that both arise from $r$. strong theorem transfer requires a map between $p(r)$ and $q(r)$ that preserves the relevant structure. if either projection is many-to-one, some distinctions vanish. if the maps preserve different relations, a theorem expressible on one side may have no counterpart on the other. even perfect supervenience of phenomenology on physics supplies a function from physical states to phenomenal states; it does not supply an inverse.

the same problem appears inside physics. gauge-related mathematical descriptions can represent one physical state. dual theories can use very different primitives while agreeing on observables. coordinate-dependent features are not physical merely because they appear in an exact formalism. a qualia formalism therefore needs more than a mathematical object. it needs a principled quotient: which transformations change only our description, and which change experience?

this is where the newman problem for consciousness science becomes relevant. a merely structural claim can become nearly vacuous if it says only that some set of the right cardinality instantiates an abstract relation. avoiding that problem requires grounding the elements and relations of phenomenal structure in a way that is not exhausted by arbitrary relabeling.

five demands on a phenomenal representation

1. intrinsicness. the represented features should not depend on a coordinate system, semantic decoder, or observer-imposed state map.
2. canonicality. equivalent descriptions of the same physical situation should produce the same phenomenal object, up to a declared equivalence relation.
3. causal relevance. distances, boundaries, or symmetries in the object should constrain the results of interventions rather than merely fit reports after the fact.
4. compositional discipline. the representation should say when subsystem structures combine, overlap, exclude one another, or fail to form one subject.
5. phenomenal anchoring. the structure must connect to experienced similarity, discriminability, unity, or valence without being freely refit to whatever data are available.

these demands reveal three separate bridges that the essay sometimes treats as one:

• a physical-to-phenomenal bridge identifies which physical distinctions alter experience;
• a phenomenal representation theorem identifies the mathematical object that preserves those distinctions;
• an interpretation rule identifies which invariants of that object correspond to color, valence, unity, intensity, or other phenomenal facts.

stv is an interpretation rule. it cannot select the physical-to-phenomenal bridge or the representation by itself without risking circularity: choose the geometry whose symmetry matches reported pleasure, then cite the match as evidence that symmetry is pleasure.

there is also a residual quidditism problem. a formalism might perfectly capture similarities, discriminabilities, composition, and transformations while leaving open whether two intrinsic qualities could exchange places without changing that structure. qualia formalism must eventually bite one of two bullets: either phenomenal identity is exhausted by structural position, or the formal object needs primitively grounded quality labels in addition to relations. the second option can still be mathematically exact, but mathematics would catalogue the labels rather than derive their intrinsic character. this is a different threat from linguistic dissolution and another reason the first fork has more than two natural research programs.

strong monism, repaired

a defensible successor would be:

there exists a nontrivial, structure-preserving partial translation between a privileged class of gauge-invariant physical observables and a privileged class of phenomenal invariants.

“partial” is doing honest work. it permits some physical facts—perhaps absolute location, gauge choice, or facts outside a subject’s boundary—to lack phenomenal counterparts. it also permits phenomenal truths to correspond not to individual physical variables but to equivalence classes or global relations. “structure-preserving” makes the proposal testable: the map must preserve specified composition, distance, order, or symmetry relations.

the original meta-theorem is grander. the repaired one is better. it can fail locally, generate concrete mathematics, and reveal exactly which portions of physics have phenomenal shadows.

synthetic referee c: atoms and bits are not rival substances

perspective: theoretical computer science, physical computation, and machine consciousness

question 2 asks whether we should track atoms or bits when locating consciousness. this is rhetorically crisp and ontologically crooked.

atoms are physical constituents at one scale, not the final inventory of contemporary physics. bits are not constituents at all. a bit is a distinction under an encoding: two reliably discriminable states used to carry information. the same physical trajectory can support many computational descriptions, and the same computation can be implemented by many physical systems. “atoms or bits?” therefore compares a physical level with an abstract description of patterns across levels.

the strongest objection to team bit is not that computation is unreal. it is that bare computation is too cheap. turing completeness establishes what a system can compute given unbounded time and memory; it discards the very properties most plausible as consciousness-relevant: causal topology, timing, locality, bandwidth, noise, counterfactual transitions, energy scales, and the boundary between system and environment. a cellular automaton, a laptop, and a suitably interpreted mechanical contraption can all be universal while being physically and causally alien.

the phrase “qualia-complete” should therefore not be modeled on turing-complete. computational universality is a capacity claim. phenomenal equivalence would be an invariance claim: which transformations of an implementation leave experience unchanged?

recent versions of computational functionalism are moving toward the right battlefield. canonical functionalism tries to define functional structure without an observer-relative semantic map. its successor, intrinsic causal-computational realization, adds internal mechanisms, intervention profiles, intrinsic state individuation, and an agent–body–world boundary. whether this program succeeds is open, but it improves “team bit” by replacing externally assigned bit patterns with physically realized counterfactual organization.

team atom also needs discipline. if complete physical state fixes complete phenomenal state, it does not follow that every microscopic detail matters. temperature supervenes on microphysics while remaining invariant under enormous microphysical variation. a physicalist qualia theory might similarly identify a mesoscopic equivalence class. substrate dependence can be real without being dependence on every substrate detail.

the clean fork is:

if two systems are isomorphic under a specified intrinsic causal-computational equivalence while differing in microphysical realization, must all their phenomenal facts match?

a yes supports a serious, intervention-sensitive functionalism. a no entails at least one additional physical invariant. the scientific task would then be to find the minimal counterexample: the smallest change that preserves the stipulated causal-computational organization while changing experience.

this suggests a useful ladder rather than two teams:

1. input-output equivalence;
2. algorithmic equivalence;
3. internal causal-graph equivalence;
4. intervention and counterfactual equivalence;
5. dynamical equivalence, including timing and noise;
6. field-level or quantum-state equivalence;
7. complete physical equivalence.

somewhere on this ladder, phenomenal invariance may begin. saying “bits” or “atoms” hides the location of the actual dispute.

the essay’s own later thought—roughly, that software descriptions are not subjects but physical computers may be—is closer to the right answer. only real systems can instantiate anything. the remaining question is which reality-preserving abstractions preserve qualia.

synthetic referee d: valence may be real without being simple

perspective: affective neuroscience, psychophysics, and theory testing

question 3 is one of the issue’s strongest forks because valence is unusually close to both phenomenology and intervention. subjects can often compare experiences by pleasantness; drugs and stimulation can alter that dimension; pain research can dissociate sensory intensity from unpleasantness. in a classic hypnosis study, altered pain unpleasantness tracked anterior cingulate activity while sensory intensity and somatosensory activity were comparatively preserved, supporting a distinction between nociceptive content and pain affect (rainville et al. 1997).

but four different claims can hide inside “valence is a natural kind”:

1. pleasantness is a genuine feature of experience rather than a linguistic fiction;
2. every experience has one determinate valence;
3. all experiences can be placed on one context-independent scalar axis;
4. that scalar has one simple physical or structural identity.

the first can be true while the rest fail. mixed emotions, localized pain amid global satisfaction, relief, bittersweetness, craving without pleasure, and pain without ordinary aversion all suggest that the global report “how good does this feel?” may compress a structured field. the right phenomenal object might be a signed distribution, vector, partial order, or temporally extended comparison relation rather than one scalar.

this is not an argument against valence realism. it is an argument against identifying realism with one-dimensionality. the natural joint may be the rule by which local valences compose, not a single global magnitude.

the symmetry theory’s representation debt

stv states that the symmetry of an information geometry of mind corresponds to the pleasantness of the represented experience. the 2023 formulation correctly separates three questions: how to generate the information geometry, which symmetry metric to apply, and how to test it (johnson 2023). that separation should become non-negotiable.

every mathematical object has symmetries relative to some transformation class. change the representation or the admissible transformations and the automorphism group changes. a perfectly homogeneous object may have abundant symmetry while containing little differentiation. a richly differentiated pleasant state may have few exact automorphisms but extensive approximate, dynamical, or multiscale harmony. “more symmetry” is therefore not yet one hypothesis.

the central risk is proxy promiscuity. synchrony, compressibility, spectral consonance, graph automorphisms, low prediction error, spatial regularity, and dimensional reduction can all be called symmetry-adjacent. they need not covary. if any one may replace another after a failed test, the theory becomes an aesthetic family resemblance rather than an identity claim.

the strongest version of stv would specify:

• the canonical phenomenal object $q$;
• the admissible transformation group $g$;
• a frame-invariant symmetry functional $s(q,g)$;
• a valence relation $v(q)$;
• the exact form of the proposed identity or monotonicity;
• the domain on which it is claimed to hold;
• rival functionals chosen before seeing the decisive data.

an adversarial test program

the theory should be tested where common proxies for symmetry and valence come apart.

1. orthogonalize valence from arousal, salience, motor readiness, novelty, attention, and report confidence.
2. include states with high synchrony or low complexity but absent, neutral, or negative experience; anesthesia and pathological hypersynchrony are useful stress tests even if they are not counterexamples to symmetry in the correct geometry.
3. separately manipulate wanting, liking, learning, pain intensity, and pain unpleasantness.
4. preregister several symmetry metrics and out-of-sample predictions rather than selecting the best-fitting measure afterward.
5. demand cross-individual and cross-species transfer with minimal refitting.
6. use causal interventions on candidate hedonic control sites, then predict the distributed geometric change rather than merely correlating local activity with reports.
7. test composition: does the metric predict mixed and spatially localized valence, not just one retrospective rating?

a positive result on this program would be genuinely startling. a negative result should also be allowed to teach us something. the claim that positive evidence supports the whole stack while negative evidence merely indicts an auxiliary assumption is tolerable only during reconnaissance. mature stv needs mortal versions.

the theory’s best possible evolution may also be subtler than “symmetry equals pleasure.” valence could track a particular pattern of symmetry and symmetry breaking: coherence without feature-erasure, differentiation without unresolved conflict, or accessible transformations that preserve a stable whole. that idea is compatible with the broader scientific use of symmetry, where broken symmetries often generate the interesting structure. but it is a new hypothesis and should not be smuggled in as the original one after the fact.

synthetic referee e: noether does not accept metaphor as currency

perspective: mathematical physics, field theory, and gravitational bounds

the essay rightly treats symmetry as a generative scientific heuristic. it is less careful with what noether’s theorem licenses.

the popular slogan says every symmetry corresponds to a conservation law. the actual theorems require specified mathematical machinery: an action, differentiable transformations, equations of motion, and distinctions between global symmetries, local gauge redundancies, boundary terms, and constraints. local gauge symmetries are handled differently from ordinary global symmetries, and the relation between symmetries and physically meaningful conserved charges is not a context-free one-to-one dictionary (brading and brown 2000).

this matters for question 7. before asking whether consciousness is conserved, one needs:

• a phenomenal state space;
• an additive scalar charge;
• a notion of local phenomenal current;
• a dynamics or action;
• a continuous transformation under which that action is invariant;
• boundary conditions under which a global total is defined.

none of these follows from monism. patterns made of conserved matter and energy need not themselves be conserved. temperature, vortices, organisms, and computations can appear and disappear while underlying conserved charges balance perfectly. even if consciousness is wholly physical, “amount of consciousness” may behave like an emergent organization rather than a charge.

the global question is harder still. summing consciousness “in the universe at a time” requires a spacetime slicing; relativity gives no unique universal now. overlapping or nested candidate subjects create double-counting problems. an additive measure must decide whether one integrated system plus its parts counts once, many times, or according to an exclusion rule. without those choices, conservation is not false. it is undefined.

a salvageable question is local:

does any well-defined additive phenomenal scalar obey a continuity equation across physically admissible processes?

if such a quantity existed, one could then seek its symmetry. noether’s theorem would organize an achieved formalism; it does not independently predict that consciousness should be conserved.

what the bekenstein bound does not bound

the bekenstein bound constrains entropy for a bounded physical system with specified energy and size. in one common form,

[ s \leq \frac{2\pi k e r}{\hbar c}. ]

the modern quantum-field-theoretic story is subtler than “bits per region”; casini’s formulation uses relative entropy and clarifies what quantities are well-defined (bekenstein 1981; casini 2008).

on casini’s interpretation, the bound arises from positivity of relative entropy and may supply no independent new semiclassical constraint. this makes the casual gloss “a region contains at most this many ordinary bits” especially treacherous. even the physical information claim depends on what entropy, region, reference state, and operational task are meant before any psychophysical extrapolation begins.

even granted the bound, no consciousness result follows without a bridge inequality. entropy is not automatically phenomenal intensity, integration, complexity, valence, or number of subjects. a maximally entropic state may have poor accessible organization. a black hole’s enormous entropy does not by itself establish that it is an enormous experience.

at least five candidate bounds should be separated:

1. the number of physically distinguishable states in a region;
2. the rate at which a system can traverse distinguishable states;
3. the width or integration of one subject;
4. the magnitude or intensity of experience;
5. the achievable positive or negative valence.

the bekenstein bound bears most directly on the first. the margolus–levitin bound constrains the speed of dynamical evolution in terms of energy above the ground state (margolus and levitin 1998). neither alone provides the remaining three. a consciousness theory could connect them, but the connection would be the discovery.

question 8 therefore contains a valuable research prompt and an invalid inference. physical finitude should constrain any physically realized phenomenology somehow. the form of that constraint is precisely what cannot be borrowed for free from “information is physical.”

synthetic referee f: entropy is not experience

perspective: thermodynamics of computation, nonequilibrium physics, and cosmology

“does conscious experience consume negentropy?” compresses several distinct processes:

1. instantiating one physical state;
2. maintaining that state against thermal noise;
3. evolving to a distinguishable state;
4. erasing or merging computational records;
5. producing a report or memory;
6. realizing a unit of valuable experience.

these processes have different thermodynamic profiles. landauer’s principle places a cost on logically irreversible operations such as erasure, not on every logical operation merely because it is computation. reversible computation can in principle approach arbitrarily low dissipation when performed sufficiently slowly, although finite speed, reliability, control, memory management, and error correction reintroduce costs (bennett 2003; wolpert 2019).

actual brains are warm, noisy, metabolically hungry nonequilibrium systems. this shows that human consciousness is implemented by dissipation. it does not yet show a universal lower bound per experience. such a bound would require a theory connecting phenomenal change to physical state discrimination, speed, error tolerance, and control.

“negentropy” is also too loose for the work demanded of it. the relevant resource might be free energy or exergy relative to an environment, not simply low entropy. a cold ordered state can be useless if no accessible gradient permits work. conversely, a reversible trajectory can visit many states without paying the landauer cost at every step.

finite negentropy does not equal finite value

the essay glosses question 9 as whether the universe contains finite or infinite potential value. the equivalence fails in both directions.

• a positive cost per experience yields finite total value only if usable resources and available time are finite and value per experience is bounded.
• a universe with unbounded time or resources may support infinite integrated experience despite a positive local cost.
• an arbitrarily small dissipation cost does not guarantee infinite value if a causal patch has a finite state space, finite duration, quantum speed limits, or recurrence that should not be counted repeatedly.
• zero marginal cost for maintaining a state does not imply that the state contains positive valence, or that duration integrates linearly into additional value.
• infinite positive and negative value can coexist, producing a measure problem rather than a happy infinity.

the thermodynamic question and the axiological measure question must therefore be separated.

a cleaner physical fork is:

at fixed implementation fidelity, error tolerance, and rate, is there a strictly positive lower bound on free-energy dissipation per nontrivial phenomenal transition?

“transition” avoids assuming that a static instant consumes anything. fixed fidelity blocks the trick of allowing the implementation to dissolve into noise. fixed rate blocks the ideal reversible limit from hiding all cost in infinite slowness. a theory could then ask separately whether phenomenal duration without transition accrues value.

if valence becomes measurable, a useful engineering quantity would be a valence yield,

[ \eta_v = \frac{\int v(\tau),d\tau}{\delta f}, ]

where $v(\tau)$ is signed valence over the subject’s proper time and $\delta f$ is consumed free energy. this would not settle ethics, but it would make the cosmological resource question explicit. different substrates could then be compared by positive yield, negative-value risk, stability, and convertibility rather than raw operations per joule.

the aestivation hypothesis is relevant only after this bridge is built. waiting for a colder universe may increase thermodynamic computation per unit energy under some assumptions (sandberg, armstrong, and ćirković 2017). it need not increase valuable consciousness per unit cosmic resource. the mapping from computation to experience—and from experience to value—is the whole game, not a rounding error.

synthetic referee g: every universe optimizes a sufficiently bad objective

perspective: inverse problems, machine learning, and cosmological model selection

question 10 may be the issue’s most fertile question after repair and its emptiest question before repair.

given any observed universe $u_*$, define an objective $j(u)$ that equals 1 at $u_*$ and 0 elsewhere. our universe then uniquely maximizes $j$. define $j(u)=0$ everywhere and every possible universe is optimal. more smoothly, one can construct a loss proportional to distance from the observed constants and dynamics. unrestricted existence of some optimized quantity is therefore guaranteed by mathematical gerrymandering.

this is the cosmological version of reward ambiguity. inverse reinforcement learning cannot generally infer a unique objective from behavior because many reward functions rationalize one policy. with unknown decision procedures, even simplicity priors can fail to identify the intended reward without additional normative assumptions (armstrong and mindermann 2018). inferring a cosmic objective from one realized universe is an even more underdetermined inverse problem.

action principles do not solve this. many physical theories admit variational formulations, but an extremum of action is not automatically a maximized good, a goal pursued by the universe, or a unique explanatory objective. the same dynamics can often be represented by different actions or reformulations. stationary points can be minima, maxima, or saddles.

strong monism does not turn this descriptive extremum into a phenomenal or ethical optimum. even an exact map between physical and phenomenal structure would need to preserve the relevant ordering. an isomorphism of state spaces does not imply that the physical action ranks states by goodness.

how to make cosmic optimization non-vacuous

the objective class should be restricted before the constants are fit. a serious candidate should be:

• low in description length;
• invariant under independently motivated physical symmetries;
• local or compositionally well-defined;
• non-constant over a principled comparison class;
• robust to small perturbations and reparameterization;
• forbidden from naming our observed universe or constants directly;
• capable of predicting withheld dimensionless relations;
• substantially better supported than anthropic selection, dynamical necessity, or chance under the same prior.

the strongest protocol would look like this:

1. define a comparison ensemble of laws, boundary conditions, or dimensionless parameters;
2. preregister a prior over simple physical and phenomenal functionals;
3. fit only a subset of observed regularities;
4. derive out-of-sample predictions for withheld constants, scaling relations, or structural features;
5. penalize objectives whose optimum is broad enough to include nearly anything;
6. test whether the same functional remains privileged under alternative parameterizations and cosmological measures.

only then does “the universe optimizes x” outperform “x was reverse-engineered to flatter the universe.”

the repaired fork is:

relative to a preregistered, low-description-length objective class and a physically motivated counterfactual ensemble, does our universe robustly extremize a phenomenal functional?

a yes would be extraordinary. a no would still be informative. the original question’s “some optimization problem” makes no losing move available; reality has been overfit with one data point. cute, but cooked.

synthetic referee h: value does not fall out of valence

perspective: ethics, welfare measurement, and moral uncertainty

the essay is unusual and admirable in allowing metaphysics to reach all the way into cosmic strategy. if consciousness is formalizable, substrate choices and resource bounds could determine which futures contain patients, suffering, or value. most philosophy of mind politely stops before the bill arrives.

still, several bridges remain unpaid.

valence realism does not imply hedonism. a complete scalar of pleasantness would not by itself settle whether agency, knowledge, achievement, relationships, diversity, autonomy, justice, or preference satisfaction have terminal value. nor would it settle aggregation: whether one intense pleasure outweighs many mild pains, whether distributions matter, how personal identity affects sums over time, or whether creating happy beings is morally equivalent to helping existing beings.

naturalness does not imply normativity either. if the universe happens to optimize a physical or phenomenal quantity, that fact does not make the quantity good. evolution optimizes proxies under constraints without issuing moral commands. strong monism can identify one underlying reality; it cannot derive an ought from every invariant it finds.

none of this makes a valence formalism ethically modest. quite the reverse. an objective measure of phenomenal valence would transform medicine, animal welfare, machine-consciousness policy, and long-run planning. it could reveal hidden moral patients and magnitudes that unaided empathy misses. but the measure should enter ethics as powerful evidence, not as a coup.

the original fork set also focuses on total consciousness when the urgent quantity may be preventable negative valence under uncertainty. conserving a large amount of consciousness would be morally catastrophic if its sign became negative. maximizing consciousness density could maximize suffering density. maximizing valence yield could create fragile monocultures or sacrifice agency. each optimization target needs adversarial moral analysis.

the missing ethical fork is:

is phenomenal valence the only intrinsic bearer of terminal value?

if yes, a mature valence science could in principle ground a cosmic objective, though aggregation and population ethics would remain. if no, the target is a vector or richer structure, and “hedonium” is one good among several. either answer changes how the physical forks should be used.

under deep uncertainty about atoms, bits, boundaries, and valence, the rational stance is not paralysis but portfolio ethics. preserve option value, avoid irreversible creation of vast poorly understood subjects, test theories on reversible scales, and give special weight to downside tails. the metaphysics of machine consciousness should be allowed to influence engineering before consensus, because waiting for consensus may itself be an irreversible substrate bet.

the essay’s closing intuition survives: consciousness research may help reveal what game we are playing. it will not, by itself, tell us every rule by which the game should be judged.

editorial synthesis: ten repaired forks

the following questions preserve the original sequence—from formalism through substrate, valence, physics, thermodynamics, and cosmology—while trying to make each branch exact enough to support cumulative work.

1. phenomenal canonicality

for every possible total experience, does there exist a canonical mathematical object—unique up to an explicitly specified equivalence relation—that fixes all phenomenally relevant facts?

this separates exact phenomenal structure from the trivial claim that experiences can be arbitrarily encoded. a yes-world must supply invariants and admissible re-descriptions. a no-world contains at least one phenomenal fact that cannot be captured in such an object or no canonical identity conditions across representations.

2. psychophysical determination

holding complete intrinsic physical causal structure fixed, are all phenomenal facts fixed?

this is a supervenience fork. a yes does not yet identify the map. a no requires either extra nonphysical facts, indeterminacy in phenomenal realization, or failure of “complete intrinsic physical structure” to denote the relevant base.

3. substrate invariance

if two systems are isomorphic under a specified intrinsic causal-computational equivalence while differing microphysically, must their phenomenal facts match?

this replaces atoms versus bits. the equivalence must state which interventions, timings, mechanisms, and boundaries are preserved. each stronger equivalence on the implementation ladder yields a separate empirical or conceptual fork.

4. valence order

does every total experience possess a context-independent scalar valence such that all possible experiences are globally orderable by pleasantness?

a no permits valence realism with partial order, vector structure, local fields, or genuine incomparability. this question should be settled before treating total valence as a conserved or optimized scalar.

5. symmetry–valence identity

relative to the canonical object in fork 1, does there exist a frame-invariant symmetry functional that is exactly monotonic with the valence relation in fork 4 throughout its declared domain?

this is stv with the representation, transformation class, metric, domain, and strength of relation exposed. a preregistered counterexample can now count.

6. psychophysical structure preservation

does the psychophysical map preserve enough specified structure to translate a nontrivial preregistered class of physical theorems into phenomenal theorems and back?

this is strong monism without universal overreach. candidate preserved structures might include order, topology, composition, symmetry, or conserved quantities. success for one class does not imply success for all physics.

7. phenomenal continuity

does any nonzero additive phenomenal scalar obey a local continuity equation across all physically admissible histories in its domain?

this makes “conservation of consciousness” a mathematical question. it requires an account of subjects, overlap, additivity, local current, and relativistic covariance. the default prior should be low until such a scalar is exhibited.

8. phenomenal capacity

for a fixed causal region, energy budget, and duration, is the number of mutually discriminable total experiences physically realizable in that region finite?

this is closer to what entropy and holographic bounds might constrain. it does not assume that the same bound controls intensity, integration, or valence. those require separate inequalities.

9. thermodynamic price of phenomenal change

at fixed fidelity, error tolerance, and transition rate, is there a strictly positive lower bound on free-energy dissipation per nontrivial phenomenal transition?

this separates physical maintenance, reversible dynamics, erasure, and phenomenal time. only after answering it should one infer limits on total realizable experience from cosmic resources.

10. cosmological phenomenal extremum

over a physically motivated counterfactual ensemble, does our universe robustly extremize a preregistered, low-description-length phenomenal functional?

the objective must be specified independently enough to make failure possible. out-of-sample prediction is the price of calling the extremum explanatory rather than decorative.

dependency structure

the result is not a ten-dimensional hypercube. it is a research graph with two roots, several conditional branches, and multiple ways for later questions to become undefined. this is less visually clean and more scientifically honest.

a ranked research program

1. force stv into an adversarial preregistration

this is the highest expected-information project because the hypothesis is specific enough to risk something, connected to interventions, and potentially useful. choose one candidate phenomenal representation, one symmetry functional, explicit confounds, and several states selected to dissociate symmetry proxies from valence. lock the analysis before observing the decisive outcomes. the goal is not to test “symmetry somehow matters”; it is to give one version of stv permission to die.

2. define the substrate invariance ladder

replace atoms versus bits with a sequence of equivalence relations, from input-output behavior through intrinsic causal and field-level dynamics. for each rung, ask what a qualia theory predicts will remain invariant. this would organize machine-consciousness disagreements far better than functionalist versus biological labels.

3. specify a representation selection principle

qualia formalism’s bottleneck is not a shortage of mathematical spaces. it is an excess. articulate why one construction is canonical: gauge invariance, causal efficacy, compositional behavior, cross-subject transfer, and independent phenomenal constraints. without this, symmetry metrics are free to chase the target.

4. build a compositional valence assay

test whether global valence is truly scalar. use mixed, localized, temporally changing experiences and dissociations among liking, wanting, pain intensity, and unpleasantness. a failure of scalarity would not kill valence realism; it would reveal the algebra that a serious stv must explain.

5. attempt one restricted strong-monism translation

do not start with every theorem in physics. choose one physical structure—perhaps a symmetry, topology, conserved flow, or boundary operation—and state its proposed phenomenal counterpart. define what preservation means and what observation would break the map. one successful nontrivial translation would be worth more than the unrestricted meta-theorem.

6. derive a bridge bound before invoking cosmology

seek an inequality connecting a physical resource to a phenomenal quantity: distinguishable experiences, subject width, transition rate, or valence magnitude. the physical bound is the easy half. the psychophysical bridge is the actual contribution.

7. treat cosmic optimization as held-out prediction

preregister a low-complexity functional and comparison class, fit only part of the data, then predict withheld relations. without this, the optimization claim is an aesthetic gloss on known constants. with it, even a failure could sharply update the program.

editorial ranking of the original forks

• most important methodological fork: 1, after separating representability from canonicality.
• most promising empirical fork: 5, provided the representation debt is paid.
• most important for ai: 2, after replacing bits with intrinsic causal-computational equivalence.
• most overclaimed: 6; monism does not buy invertible theorem transfer.
• most ill-typed: 7; no conserved quantity has yet been defined.
• most tempting non sequitur: 8; entropy capacity is not consciousness density.
• most important hidden distinction: 9; thermodynamic implementation cost is not total cosmic value.
• most fertile after repair: 10; inverse cosmological optimization could become a real predictive program.

protocol for future model editions

if this format is refreshed with each major frontier-model release, the process should resist anchoring on the previous issue.

pass 1: blind reconstruction

give the new model the original essay, its figures, and a fixed bibliographic cutoff. do not provide prior model commentary or the author’s later responses. ask it to recover the argument, dependencies, strongest claims, and most vulnerable inferences.

pass 2: independent disciplinary reports

request bounded reports from the same stable perspectives—logic, structuralism, computation, neuroscience, physics, thermodynamics, inverse problems, and ethics—plus one wildcard perspective selected by the model. require each report to offer both a criticism and a repair.

pass 3: adversarial synthesis

ask a fresh context to identify shared assumptions among the reports, objections none of them answered, and places where apparent consensus is merely shared training-data gravity.

pass 4: historical update

only now provide the prior issue, the author’s later work, empirical developments, and public responses. produce a delta log:

• insights stable across model generations;
• claims strengthened or weakened by new evidence;
• genuinely new objections;
• discarded cleverness;
• changes in the repaired fork set;
• tests that became feasible.

pass 5: author response

keep the author’s reply human and separately labeled. the model may draft questions, but it should not simulate agreement on the author’s behalf. future issues will be more valuable if they record where the author bites bullets, changes terms, or refuses the referees’ ontology.

versioning fields

each issue should record:

• source essay version and retrieval date;
• model and release date;
• knowledge or search cutoff;
• whether prior issues were hidden during the first pass;
• tool-access level;
• major prompt changes;
• a short changelog from the preceding issue.

the recurring product should not be “the newest model’s verdict.” it should be a longitudinal instrument for tracking which conceptual compressions survive stronger reasoners. model disagreement is data; convergence is only interesting when the generations did not copy one another’s homework.

references

• armstrong, s., and mindermann, s. (2018). occam’s razor is insufficient to infer the preferences of irrational agents.
• bekenstein, j. d. (1981). universal upper bound on the entropy-to-energy ratio for bounded systems.
• bennett, c. h. (2003). notes on landauer’s principle, reversible computation, and maxwell’s demon.
• brading, k., and brown, h. r. (2000). noether’s theorems and gauge symmetries.
• casini, h. (2008). relative entropy and the bekenstein bound.
• johnson, m. e. (2022). it from bit, revisited.
• johnson, m. e. (2023). qualia formalism and a symmetry theory of valence.
• kanai, r., and ma, s. (2026). canonical functionalism: defining functional structure without observer-relative semantic maps.
• kanai, r., and ma, s. (2026). intrinsic computational functionalism and simulated consciousness.
• kleiner, j. (2025). the newman problem of consciousness science.
• kleiner, j., and hoel, e. (2021). falsification and consciousness.
• margolus, n., and levitin, l. b. (1998). the maximum speed of dynamical evolution.
• rainville, p., duncan, g. h., price, d. d., carrier, b., and bushnell, m. c. (1997). pain affect encoded in human anterior cingulate but not somatosensory cortex.
• safron, a., levin, m., razi, a., sheikhbahaee, z., and sakthivadivel, d., eds. (2023). making and breaking symmetries in mind and life.
• sandberg, a., armstrong, s., and ćirković, m. m. (2017). the aestivation hypothesis for resolving fermi’s paradox.
• wolpert, d. h. (2019). the stochastic thermodynamics of computation.