Thermodynamic Realism: What It Proves and What It Bets
Thermodynamic Realism: What It Proves and What It Bets
Two layers, a descriptive thermodynamic program and a metaethical layer built on it: axioms, deductive web, real and open formalization, an agent-relative dissolution of the is-ought gap, the domain diagnostic with a worked audit, falsification protocols, and a full register of what remains open.
What this is. A deductive presentation of a research program, not a finished metascience and not a proof. It traces numbered results from a small physical base, gives real formalization where the physics supports it and marks the rest open, dissolves the is-ought problem by treating "ought" as domain-bound, and names its own principal exposure. Every result carries a label recording what kind of move it is. No result should be read as carrying more weight than its kind permits.
Two layers, one document. The program has two separable layers. The first is descriptive: persistence, modeling cost, selection, and the functional accounts of valence and collapse. It has a determinate territory and a testable fidelity. The second is metaethical and takes the first as input. The claim that the descriptive substrate grounds the normative is the framework's central hypothesis, not an entailment. The descriptive layer can be entirely correct while the metaethical layer is wrong, and a reader may accept one and reject the other. They appear in a single deductive web for economy, but the web's transition into Layer 3 crosses from physics into semantics and should be read as a change of register, not a continuation of one proof.
- How to read the labels
- The problem of fragmentation
- Axioms and background premises
- The deductive web (L1 to L4)
- Formalization: real and open
- The is-ought dissolution: two parts, and only one carries it
- The domain diagnostic, with a worked audit
- Valence, empathy, collapse, maximizers
- Objections, answered at strength
- Falsification protocols
- Open questions and limitations
- Cosmic scope and boundaries
- Conclusion. Glossary. Personal extension. References.
1. How to read the labels
Every numbered result carries a label. This is the single most important interpretive instruction in the paper.
Axiom bedrock, not derived. Background premise an empirical fact about our universe the derivations rely on but cannot produce from the axioms alone. Theorem an entailment, traced to its parents. Definition fixes a term. Characterisation proposes what a familiar word picks out within the framework. Functional account explains why a feature exists and what it does, without claiming to reduce its felt character. Analogical extension applies the architecture to a domain by analogy rather than strict entailment. Corollary a descriptive consequence. Falsification a refutation condition, not a theorem.
2. The problem of fragmentation
Human knowledge is partitioned into disciplines lacking a common axiomatic foundation. Physics describes the territory but says nothing about value. Information theory describes the cost of representation but says nothing about why accuracy matters. Evolutionary biology describes selection among replicators but says little about the fate of non-replicating persistent structures. Ethics asks how agents ought to behave but struggles to ground "ought" in "is."
The proposal: a single set of physical premises, stated explicitly and followed wherever they lead, yields a structure in which the is-ought gap closes as a semantic artifact, valence finds a functional grounding, ought-facts become physically determinate though often computationally inaccessible, and the collapse of information-suppressing regimes becomes a thermodynamic hypothesis. The framework is called Thermodynamic Realism.
A note on the cosmological backdrop. The picture of the universe as a single thermodynamic computation is referenced background, not part of the axiom base. The deductive web requires only a free-energy gradient, which enters as a background premise. Keeping the axiom base minimal is deliberate: load-bearing claims should depend on as little contested physics as possible.
3. Axioms and background premises
3.1 Axioms
A0. The Persistence Tautology Axiom
Systems that do not maintain the conditions of their own persistence cease to exist as observables. Writing \(S(t)\) for the system at time \(t\): if \(S\) fails its persistence conditions over \([t, t+\Delta t]\), then \(\neg S(t+\Delta t)\). This asserts nothing about value; it states that existence has prerequisites and what fails them is no longer present to be observed.
A1. Physicalism Axiom
All phenomena, including life, mind, and culture, are physical. This is physicalism in the minimal sense; on its own it does not settle whether phenomenal experience is exhausted by function (§6 boundary). Justification: the maximally inductively justified working premise. Every phenomenon ever seriously investigated has yielded to physical explanation.
A2. The Second Law Axiom
In any isolated system, entropy tends toward its maximum; maintaining a localized entropy gradient requires continuous work. Among the most thoroughly confirmed principles in science; adopted without re-derivation.
A3. Information is Physical Axiom
Storage, processing, and erasure of information carry minimum thermodynamic costs (Landauer 1961). The principle is philosophically contested (Norton 2011; replies Ladyman and Robertson 2013, Myrvold 2024). TR does not require it as a universal theorem of mechanics. It requires the weaker, better-supported claim that the operations the framework relies on, the maintenance and updating of divergent models (T8, T9), are logically irreversible and therefore carry an irreducible cost (Bennett 1982).
3.2 Background empirical premises
BP0. The Low-Entropy Initial Condition (the Past Hypothesis) Background premise
The accessible universe began in a macrostate of extraordinarily low entropy. Every free-energy gradient available to any later system is a portion of that initial condition still discharging (Albert 2000). The one role gravity plays: for self-gravitating matter a smooth distribution is low entropy and a clumped one high, because gravity makes clumping spontaneous (Penrose 1979), which is what makes the smooth early universe a wound spring rather than a featureless equilibrium. No theory of quantum gravity is required, only this fact.
BP1. Non-Equilibrium Existence Background premise The accessible universe is far from equilibrium; gradients exist and sustain localized order. Follows from BP0 plus finite elapsed time; retained separately so traces citing it remain stable.
BP2. Finite Accessible Resources Background premise Local free energy is finite. With A2 this implies competition for negentropy among co-located systems.
BP3. Evolutionary Dynamics and Multi-Scale Organization Background premise Differential survival among persisting systems with heritable or persistent traits produces selection effects. Persistent systems are organized into nested statistical boundaries (cells within organisms within ecosystems within civilizations). An empirical fact, not a logical necessity.
4. The deductive web
Layer 1: immediate consequences
T1. Environmental variance is non-zero and, for finite agents, ineliminable Theorem from A1, A2, BP1
Any embedded agent encounters a non-zero rate of environmental shift. Insulation is itself costly and eventually fails, so variance is ineliminable in the long run.
T2. Persistence requires work Theorem from A2, A0
To persist is to maintain a boundary against entropic dissolution; the Second Law makes this require continuous work.
T3. Modeling has a minimum cost Theorem from A3
Any internal model is encoded in physical degrees of freedom; storing, accessing, and updating it incurs non-zero thermodynamic cost.
T4. Map-territory divergence has a thermodynamic cost Theorem from T1, T3; formalized in §5.1
A model that fails to track environmental shift generates prediction errors that dissipate free energy. The thermodynamics of prediction gives this a precise form: dissipated work is bounded below by the model's nonpredictive retained information.
T5. The persistence selection principle Theorem (statistical) from T2, T4, A0, BP2, BP3; formalized in §5.3
In an ecology competing for finite free energy, systems with lower divergence dissipate less on error correction, on average. Over many perturbation cycles the distribution of observed systems shifts toward closer trackers. Entropy performs epistemic selection.
Scope note. T5 is not a guarantee that the most accurate model always wins. It is a statistical tendency where variance is non-zero and resources finite. In a perfectly stable, resource-abundant niche a distorted model can persist (the dark-room limit, §6.2). It does not strengthen into a deterministic law on any horizon (§9, objection 7).
Layer 2: structural deductions
T6. Directional drift toward lower divergence Theorem from T1, T5
Over sufficient time and perturbation variety, surviving systems are those whose models track causal invariants. This defines a direction, decreasing divergence, not a fixed endpoint. The drift is directional, not convergent; no terminal "true model" is implied.
T7. Truth as perturbationally robust compression fidelity Definition motivated by T6, T3
Truth is defined as the minimal-loss compression of environmental structure sufficient for adaptive persistence across expanding perturbational horizons. A definitional choice, not a deduction, operationalizable via Minimum Description Length and predictive mutual information. Its known cost, that adaptive falsehoods can come apart from accuracy, is addressed in §9, objection 8.
T8. Intentional deception costs more than truth-telling Theorem from T3, T4; formalized in §5.2
A lie requires the sender to maintain at least two models, the accurate one and the presented one, plus machinery to deploy the second while reasoning with the first. This imposes strictly greater storage, update, and monitoring cost. Deception is thermodynamically disfavored, though it can be locally advantageous when offsetting returns compensate for the overhead.
Scope note. T8 concerns intentional deception, where the sender holds an accurate and a divergent presented model. A merely mistaken agent holds a single false model and incurs no dual-model overhead; that is the plain divergence of T4.
T9. The Consistency Tax Definition (consolidating) consolidates T4, T8
Names the metabolic overhead of any mismatch between model and territory, from error (T4) or deception (T8). Applies even when the agent is unaware of the mismatch (latent divergence) and spikes when it is actively corrected (active divergence).
T10. Epistemic Profit Theorem from T9, T3
Reducing divergence frees the energy previously consumed by the tax. Interpretive remark: the framework conjectures a felt correlate, "Predictive Calm." This is an interpretive bridge, not a derived result; nothing downstream depends on it.
Layer 3: meta-ethics
A register change, not a continuation. The results in this layer are not downstream entailments of the thermodynamic results above. The dissolution (T11) rides on A0 and the semantic premise of §6; the physics enters only at T12, to fix the content of an ought once the domain is granted. Delete A2 and A3 and T11 still stands, while T12 would need a new account of content. The metaethics is conditional on a semantic commitment, not on the thermodynamics.
T11. "Ought" is a domain-bound operator Theorem from A0, T2, and the semantic premise defended in §6
"Ought" presupposes an agent with persistence conditions; outside this domain it does not refer. "Why ought one persist at all?" is malformed in the way "what is north of the North Pole?" is. The is-ought gap is a semantic artifact of domain violation. This is not entailed by A0 and T2 alone; it rests on the semantic premise of §6, which carries the argument and marks its limits.
T12. Within the domain, ought-facts are physically determinate Theorem from A1, A2, T11; ordinal form in §5.5
For any agent with specified persistence conditions and embedding, there is a physically determinate configuration that maximizes sustained negentropy capacity over the embedding's actual horizon. This optimum may be computationally inaccessible, but inaccessibility is not indeterminacy. Conditional on T11. This is the step where the thermodynamics enters the metaethics.
T13. Ethics as the modeling activity of bounded agents Characterisation from T12, T3, T7
Because the full optimum is intractable, agents use compressed models: moral emotions (fast heuristics), moral principles (compressed generalizations), and moral reasoning (model refinement). A proposal about what "ethics" picks out, not a derived result.
T14. Moral progress is real and directional Theorem from T6, T13
As models better track the coupled-system thermodynamics of an embedding, they become objectively better moral models in the sense fixed by T7 and T13. Directional but neither guaranteed nor complete in finite time. Conditional on T13.
Layer 4: full architecture
T15. Markov blankets and multi-scale identity Theorem (imported formalism) from T2, T5, T3, BP3
Persisting systems maintain statistical boundaries separating internal from external states; these nest across scales, and selection operates at every scale. The formalism is imported from active inference as vocabulary for the boundary T2 already requires. The maximal scale, the universe, has no Markov blanket, because it has no outside.
T16. Coupling density \(\kappa\) Definition from T15, T4, T5; candidate operationalizations in §5.6
Measures information-theoretic dependence between a higher-level blanket and the systems nested within it. Tight coupling lets a macro-blanket override lower nodes (cells via apoptosis, institutions via turnover) to preserve the macro-invariant. A modelling construct, not derived. It is the crux of the positive metaethics (§5.6, §6.3).
T17. Antifragility as a condition of long-horizon persistence Theorem (qualitative) from T1, T4, T6, T7
In a high-variance environment, a system that does not improve its predictive capacity from exposure to variance cannot persist over long horizons: variance not converted into improved modelling accumulates as divergence and is selected against. The quantitative rate condition is open (§5.7).
T18. The Landauer ceiling on adaptation Theorem (in-principle bound) from A3, T3
Adaptation is model updating, bounded by available power: max update rate \(\le P/(k_B T \ln 2)\) bits per second. Real and very loose; adaptive systems run many orders of magnitude above the floor. Its role is conceptual: adaptation rate is energy-bounded at all.
T19. Valence as high-rate divergence telemetry Functional account from T4, T5, T9
Complex controllers need a priority-queuing mechanism to allocate serial processing among parallel subsystems. A non-ignorable global interrupt triggered by rapidly escalating divergence serves this role; its felt quality is negative valence, its absence across critical domains positive valence. This explains why valence exists and why any complex controller in a high-stakes environment must implement a functional analog. It does not explain why there is something it is like to be such a system (§6 boundary).
T20. Empathy as coupled telemetry monitoring Functional account from T16, T19
If agent \(A\)'s persistence is coupled to \(B\) (\(\kappa > 0\)), \(B\)'s suffering carries information about the shared embedding. Empathy, functionally, is the monitoring of coupled telemetry. Suppressing or causing suffering in coupled agents degrades collective predictive infrastructure. The phenomenology of human empathy is not exhausted by the functional account.
T21. Epistemic overshoot and civilizational collapse Analogical extension from T8, T9, T14, T20
Censorship and propaganda sever the telemetry channels that update a collective model. The official model reports alignment while actual divergence accumulates invisibly, an informational debt. When an exogenous perturbation arrives, the debt becomes lethal and the system collapses non-linearly.
Status. Applies the individual architecture to collectives by analogy. Best read as a hypothesis the framework renders testable (§10), not a theorem it proves. It does not entail that repressive regimes always fall quickly (§9, objection 1).
T22. The limits of maximizers Theorem from T2, T3, T5, T7, BP1
A maximizer that homogenizes its environment destroys the free-energy gradients that sustain it, and eliminates the variety required for adaptive control (Ashby's Law). Short-horizon maximizers are self-terminating; long-horizon ones, under selection, are forced toward variety, telemetry, and coupling. This does not dissolve the alignment problem, since a long-horizon misaligned goal could be catastrophic in the interim, but it constrains the space of viable long-term strategies.
C1. The direction of persistence selection Corollary (descriptive) from T15, T17, T22
Among systems competing under finite resources and non-zero variance, selection statistically favors those whose negentropy capacity is maximized over the longest sustainable horizon. A long horizon forces sustainability, coupling, and antifragility.
T23. Falsification criteria Falsification (not a theorem) operationalized in §10
The framework would be falsified by: (1) a rigid monoculture surviving sustained extreme variance; (2) a totally model-decoupled system outlasting a high-fidelity system under identical variance; (3) a complex controller managing acute multi-vector crises with no valence-like priority interrupt or proxy.
5. Formalization: real and open
Where the physics supports a precise statement, one is given and sourced. Where it does not, the gap is marked open rather than papered over. A prior condensation circulated expressions of the form "divergence equals energy waste," an augmented Landauer bound, and an identification of "ought" with the negative free-energy gradient. Those were placeholders and an optional vocabulary, not results. They are retired here in favor of the actual physics below.
5.1 The thermodynamic cost of divergence (grounding T4, T9)
For a system whose memory state retains information \(I_{\text{mem}}\) about a driving signal's past, part is predictive of the future, \(I_{\text{pred}}\). The remainder is nonpredictive retained information, what Still, Sivak, Bell and Crooks (2012) call nostalgia:
\[ I_{\text{nost}} = I_{\text{mem}} - I_{\text{pred}} \]
Their result is an equivalence between this model inefficiency and thermodynamic inefficiency: the work dissipated beyond the change in nonequilibrium free energy is bounded below by the nonpredictive information,
\[ W_{\text{diss}} \ \ge\ k_B T \, I_{\text{nost}} . \]
A model carrying structure not predictive of its environment pays for that excess in dissipated work, in proportion to its nonpredictive bits. T9's Consistency Tax is then the cumulative \(W_{\text{diss}}\) a system incurs by holding a map out of register with the territory's predictively relevant invariants.
5.2 The deception bound (grounding T8)
Let an agent hold an accurate map \(M_{\text{true}}\). An honest communicator transmits a function of it. A deceiver additionally maintains a presented map \(M_{\text{false}}\) and a policy conditioning output on context so as to emit \(M_{\text{false}}\) while acting on \(M_{\text{true}}\). The minimum description length of the deceiver's apparatus is
\[ L_{\text{deceiver}} = L(M_{\text{true}}) + L(M_{\text{false}} \mid M_{\text{true}}) + L(\pi) , \]
where \(L(M_{\text{false}} \mid M_{\text{true}}) > 0\) whenever the maps differ and \(L(\pi)\) is the policy that tracks which map to deploy to whom. The honest agent's apparatus is just \(L(M_{\text{true}})\), so \(L_{\text{deceiver}} > L_{\text{honest}}\) strictly. Combined with §5.1, the presented map is held but not used to track the agent's own acted-on environment, so it contributes nonpredictive retained information and therefore dissipation.
This bounds the cost of maintaining the dual structure. It does not claim deception never pays. Where the local payoff of a successful lie exceeds maintenance overhead plus disconfirmation risk, deception is locally rational. T8 claims a standing cost differential, not that it always dominates.
5.3 The selection principle in Price-equation form (grounding T5)
Let systems be indexed \(i\), each with divergence \(D_i\) and per-cycle fitness \(w_i\), where \(w_i\) is non-increasing in \(D_i\) by T4 and §5.1. The change in mean divergence across one cycle is
\[ \bar{w}\,\Delta\bar{D} = \underbrace{\operatorname{Cov}(w_i, D_i)}_{\text{selection}} + \underbrace{\mathbb{E}[w_i\,\Delta D_i]}_{\text{transmission}} . \]
Because \(w_i\) is non-increasing in \(D_i\), \(\operatorname{Cov}(w_i, D_i) \le 0\): selection alone pushes mean divergence down. This is both the strength and the limit of T5. The selection term can be positive only via transmission (drift, mutation, a novel environment outrunning models). The covariance vanishes if there is no variance in \(D\) or if \(w\) is flat in \(D\) (the dark-room limit). Nothing forces convergence to \(D = 0\); it forces directional pressure whenever the covariance is negative.
5.4 The adaptation ceiling (T18)
With updates costing at least \(k_B T \ln 2\) per bit and a power budget \(P\) for non-housekeeping computation, the maximum sustainable update rate is \(R_{\max} = P/(k_B T \ln 2)\). A real, loose, in-principle bound. Its use is in T17 and T22: a system in a high-variance environment must adapt at a rate tracking environmental drift, and a maximizer that starves its own power budget lowers its own ceiling.
5.5 The ought-relation and its ordinal generalisation (T11, T12)
The strict identity for necessary actions: "\(S\) ought to \(\phi\)" is equivalent to "\(S\)'s not-\(\phi\)-ing causes \(S\)'s non-persistence over the relevant horizon." A caution about this identity: read as a definition it would make "ought" a descriptive predicate and convert is-to-ought into valid is-to-is inference, which is the reductive route the body does not take (§13); read as the framework intends, as a defended interpretive claim conditional on the semantic premise, it does not purchase a proof. The generalisation defines a persistence functional \(\Pi(S, a)\) over available actions \(a\) and a weak ordering \(a_1 \succeq a_2 \iff \Pi(S, a_1) \ge \Pi(S, a_2)\). Probabilistically relevant actions, flourishing beyond survival, and supererogation are degrees on this ordering rather than separate categories. The ordering is ordinal and agent-indexed. It delivers a determinate ought-structure for a single agent with a single persistence dimension.
5.6 Coupling density \(\kappa\) (T16): the crux of the positive ethics
Two candidate operationalizations, neither canonical: a transfer entropy from a higher-level blanket to a nested system, measuring the information the higher system's present carries about the lower system's next state beyond its own present; or an elasticity of the lower system's sustained negentropy capacity with respect to the higher system's state. The framework uses \(\kappa\) as a comparative construct, tighter versus looser coupling, and does not claim the definitions coincide.
5.7 What resists formalization, listed honestly
| Claim | Status |
|---|---|
| T17 antifragility as necessary for long-horizon persistence | Qualitative theorem. The rate condition has no closed form here. |
| T21 collapse temporal signature | Analogical hypothesis. Specified verbally (§8.2), not formalized or fit to data. |
| Inter-agent scalarisation / Pareto rule | Open. No unique rule follows from physics (§5.5). |
| Coupling density \(\kappa\) | Open, and the crux of the positive ethics (§5.6). |
| Asymptotic survival of nonzero-divergence systems | Open, and explicitly not a strengthening of T5 into a law (§9, objection 7). |
| Phenomenal character of valence (T19) and empathy (T20) | Out of scope (§6 boundary). Functional structure given; hard problem bracketed. |
6. The is-ought dissolution: two parts, and only one carries it
The is-ought problem asks how a normative conclusion can follow from purely descriptive premises. TR's answer (T11 to T13) is that it cannot and need not, because the appearance of an unbridgeable gap is an artefact of one reading of "ought."
6.1 The two claims Hume's point makes, and the ceiling it sets
Distinguish: (D) from descriptive premises alone, no normative conclusion follows by valid deductive inference (Hume's logical claim); and (O) normative claims pick out properties of a categorically different kind (the standard ontological reading, reinforced by Moore's open-question argument). (D) does not entail (O). TR grants (D) and rejects (O), by showing the appearance of categorical distinction is generated by semantic structure, not ontological difference.
Granting (D) sets a hard ceiling that should be stated plainly, because it is not a defect to be engineered away. Every axiom and premise in this framework is descriptive. So any path from that base to a normative conclusion must cross exactly one non-descriptive premise, and the semantic premise below is that crossing. It cannot be a theorem without contradicting (D), which the framework accepts. A proof of the normative conclusion from the physics is therefore ruled out by the framework's own commitments. What follows is the most defensible structure available under that ceiling, not an attempt to breach it.
6.2 The dissolution
When a doctor says "you ought to take this medication," the claim is anchored to a specific agent with specific persistence conditions; it is a factual claim about the coupled-system thermodynamics of that patient plus the medication. T11 generalises: all contentful uses of "ought" have this structure. Uses that resist the paraphrase, "one ought to maximise aggregate utility" detached from any agent, or "ought there be a universe," are domain violations.
6.3 The semantic premise has two parts, and only the first carries the dissolution
The premise the dissolution appeared to rest on actually bundles two separable claims, and keeping them apart is the single largest fidelity gain available here. Confidence should be assigned to each separately.
Part 1 (agent-relativity). "Ought" is agent-and-end relative; deployed with no agent and no end it is type-errored rather than false. This is close to consensus across the deflationary tradition: Anscombe, Geach, and Foot hold versions of it, and even Mackie agrees there are no objective free-floating prescriptions. Part 1 alone delivers the negative result, and the negative result is the dissolution: there is no agent-independent ought, so the classical is-ought gap, posed as a chasm between pure description and free-floating prescription, is a pseudo-problem, because there is nothing on the far side of it to bridge to. This is claimed at strength. It needs neither the thermodynamics nor Part 2.
Part 2 (persistence fixes content). The end that fixes the content of an agent-indexed ought is persistence under thermodynamic constraint, grounded by A0. This is the framework's distinctive bet, and it is a bet, not a result. Its rival is not the expressivist or the realist but the flourishing-naturalist (Foot): both index "ought" to an agent and an end, and they disagree about which end. Persistence and flourishing coincide across most cases and diverge on persistence-costly behavior, where the adjudication runs through coupling density, which is open (§5.6). So the confidence of Part 2 is the confidence of \(\kappa\). It should be asserted as the live hypothesis it is, not folded into the secure Part 1.
An earlier statement fused these and inherited a confusion about which step was exposed. Split, the picture is clean: the dissolution is near-secure; the thermodynamic content of ethics is the contested part, and its crux is named.
6.4 The functional argument, and where each part is exposed
Part 1's defence is the function of the operator. "Ought" is practical; its work is to guide and assess action; an operator individuated by that work cannot discharge it absent an agent. Three usage features corroborate: an unanchored "ought" reliably prompts "ought whom?"; "ought" implies "can," intelligible only for an agent with capacities; and "ought" addressed to non-agents is heard as figurative. Anscombe (1958) is the precedent: detached from a framework supplying its subject, the bare "ought" keeps rhetorical force but loses determinate content.
Where each part is exposed differs, and conflating the exposures was the old error. Against an expressivist and against most naturalists, Part 1 is secure and the dissolution goes through. Against a Cornell naturalist realist, who holds that "ought" refers to an agent-independent natural property, Part 1 is precisely the contested step, and the trilemma below is what TR offers there: not a refutation but a statement of the realist's bill. Part 2 is exposed on a different front entirely, against flourishing-naturalism, adjudicated in spirit by whether \(\kappa\) can carry the persistence-costly cases. TR isolates both exposures rather than concealing them.
The trilemma, for a denier of Part 1. A critic who rejects agent-relativity while keeping a truth-apt universal "ought" must place the operator in one of three domains, each exacting a price:
- An unspecified physical domain. A factual imperative with no target agent, boundary, or persistence threshold specified. Supplying those parameters to rescue the claim translates it into the framework's vocabulary and the disagreement evaporates.
- A non-physical domain. A Platonic or noumenal anchor. To guide action it must interface with the physical; the coordinates of that interface are then physical and measurable in principle. So the realm is either epiphenomenal and action-inert, or physical mechanics renamed.
- A non-referential expressive domain. Expressivism: "ought" vents attitudes, not propositions. Then there is no truth-value to contest, and re-entering the truth-game by asserting an agent factually should act cycles back to the first two.
Honest weight: the trilemma is explanatory, not a knockdown. A non-naturalist realist (Moore, Parfit) genuinely occupies the second horn and accepts that normative properties are sui generis and causally idle; a Cornell realist occupies the first and insists the agent-independent natural property is real. The trilemma shows what each must pay; it does not show payment is impossible. The ceiling remains the one set in §6.1: the bridge cannot be a theorem.
6.5 Schema and content (a guard against the global-optimizer slide)
The dissolution makes the ought-schema universal: every agent has the same schema, do what sustains your persistence under perturbation over your horizon. It is tempting to slide from "the schema is universal" to "there is a universal prescription" or "a single global optimization target." That slide is invalid. The schema is nearly contentless on its own; all adjudicating content lives in the specific action it resolves to, fixed by the individual agent's persistence conditions. A universal schema with agent-relative content yields no frame-independent good and no global optimizer, and the framework is not weakened by failing to deliver one.
6.6 Positioning, in brief
Against Cornell naturalist realism, TR agrees ought-claims are factual but denies universal-scope semantics, sidestepping Moral Twin Earth by reframing the open question as a type-error. Against constitutivism (Korsgaard), TR shares grounding "ought" in conditions constitutive of agency but locates them in persistence under thermodynamic constraint rather than the unity of practical reason. Against evolutionary ethics, TR is a semantic-metaphysical account, not a genealogy, and is compatible with one. Against Mackie, TR denies the first premise: properly framed ought-claims commit to agent-relative persistence conditions, which exist.
6.7 What it does and does not deliver
With the parts kept apart: the dissolution (Part 1) is delivered at strength, the classical gap is a pseudo-problem, and that result needs neither the thermodynamics nor Part 2. The determinacy of ought-facts (T12) and the thermodynamic content of ethics (Part 2) are delivered conditionally, and their crux is \(\kappa\). The dissolution handles agent-relative and coupled-agent oughts. The hard case is fully impartial sacrifice, for an uncoupled stranger or an abstract principle, and TR does not dissolve it. It is the inter-agent normativity gap (§5.5, §11): the fully impartial categorical "ought" is either reducible to coupling at some scale or is a genuine instance of the universal-scope use the error theory targets, and the framework does not here decide which.
7. The domain diagnostic, with a worked audit
The framework yields a practical procedure: a test for whether a question's central operator (ought, meaning, purpose, cause, before) is deployed where it has content. It generalises T11 from "ought" to any operator. A question whose operator has no domain is a ghost: grammatically well-formed and physically contentless. The procedure depends on the same domain-bound semantics as T11 (specifically Part 1, agent-relativity) and has no force independent of it; a naturalist realist or expressivist will reject the premise and should expect it to yield nothing.
7.1 The four filters
Apply in sequence. Each asks whether a feature can be specified in principle, not measured.
| Filter | Question | Flagged if |
|---|---|---|
| 1. Agent | Which physical system asks, holds, or is puzzled by the question? | No physical system identifiable: the operator has no user. |
| 2. Persistence conditions | What does that system depend on to continue existing? | None specifiable: not an agent, only something named like one. |
| 3. Model | Which internal model or signal tracks the part of the world referred to? | Refers to neither a feature of the environment nor a state of the agent: no map, no territory. |
| 4. Necessity | Which selection, thermodynamic, or informational pressure requires maintaining this model? | No predictive function and no cost of absence: ornamental. |
A question passing all four is territorial: a proper candidate for truth or falsity.
7.2 What a flag means
A flag is not a verdict. The procedure outputs "flagged at filter N," and a flag has three possible explanations it does not discriminate among: the question is genuinely a ghost; a territorial reading exists that has not been found; or the procedure is miscalibrated for this case. "Ghost" is reached only after the second and third are investigated and excluded. A flag treated as a conclusion is self-confirming and uncorrectable, since the flagged question would no longer be examined. This is why the output is a flag, not a verdict. The audit below respects this: it reports flags and territorial neighbors, not final dismissals.
Why ghost-hunting is itself persistence-relevant: a ghost model is an evolutionary spandrel, an operator vital within its domain (cause-seeking, agency detection, purpose-inference) misfiring past its boundary, surviving by piggybacking on a territorial operator. Maintaining it is a standing Consistency Tax with zero persistence return. Dissolving it reduces the tax.
7.3 The audit
"Why is there something rather than nothing?" The pairing of the cause-operator "why" with a domain that structurally denies causality. "Why" requires a prior physical state as the antecedent of an explanation; absolute nothing supplies none, no time, no laws, no potential. Structurally identical to "what happened before the Big Bang?", a time-operator without time. Filters 3 and 4 flag. Territorial neighbor: "what physical laws, initial conditions, or necessary structures explain the observable universe?", addressed by cosmology, but it replaces "nothing" with specified physics and is no longer the original question. Status: flagged, with cosmology as the territorial neighbor. The flag directs attention to the operator; it does not declare the question forbidden.
"What is the meaning of life?" Two readings. The cosmic reading deploys a purpose-operator on the totality, which has no Markov blanket and no persistence conditions, so no holder for the purpose: flagged. The agent-indexed reading, "what should I value, given the life I have," passes all four filters: agent is the asker, model is the agent's own valence gradients and coupling topology, necessity is that an agent failing to model what it values cannot prioritize. Status: the procedure separates the two; the cosmic reading flags, the agent-indexed reading is territorial.
"Do we have free will?" The libertarian reading posits an uncaused cause, an event with no physical antecedent; the cause-operator is outside its domain: flagged, and no territorial reformulation of the incoherent core is readily constructible. The compatibilist reading, "was the choice generated by the agent's own model without external override," passes: the self-model is physically instantiated and necessary for counterfactual evaluation and credit assignment. Status: a territorial core with one contested component. The procedure locates the contested component and does not settle whether libertarian free will is additionally real.
The hard problem of consciousness. Here the procedure must be handled carefully, because a careless verdict conflicts with the framework's declared neutrality. If "phenomenal consciousness" is defined as an ontologically distinct non-physical property with zero functional, physical, or behavioral impact, then the request for a physical explanation of it is a cause-operator aimed at a non-physical target, and modeling it would require maintaining a map of an unobservable territory: infinite precision, zero correction, no convergence, a standing Consistency Tax that can never pay rent. That non-physical framing flags. If "phenomenal consciousness" is instead a physical phenomenon not yet fully understood, the question passes and becomes the territorial neighbor: "what physical structures give rise to self-models that report qualitative experience?", open and scientific.
| Question | Framing | Filter status | Territorial neighbor |
|---|---|---|---|
| Why something rather than nothing? | Absolute metaphysical | Flagged (3, 4) | Cosmology, with "nothing" replaced |
| Meaning of life | Cosmic / objective | Flagged | Agent-indexed valuing |
| Meaning of life | Agent-indexed | Passes | Territorial |
| Free will | Libertarian (uncaused) | Flagged | None readily constructible |
| Free will | Compatibilist | Passes | Territorial |
| Hard problem | Non-physical essence | Flagged (3) | Physical self-modeling (open science) |
7.4 The procedure applied to itself
The agent is the user; persistence conditions are specifiable; the model is the procedure, though its territory, "whether a question has a domain," is itself a construct of this framework, so Filter 3 is non-trivial; the necessity is the reduction of resources spent on type-errored questions, conditional on the procedure being correct. It passes. This is not a validation: a test built from a premise passes premise-conforming objects, including itself. The procedure has a false-positive rate (flagging territorial questions whose territorial reading the user failed to construct) and a false-negative rate (passing type-errored questions for which a plausible agent and model were fabricated). It constrains inquiry; it does not replace it.
8. Valence, empathy, collapse, maximizers (expanded)
8.1 Valence and the hard-problem boundary
Complex agents face a coordination problem: many parallel subsystems, largely serial output. When a critical subsystem detects a large, rapidly escalating divergence (a predator, tissue damage, sudden depletion), the appropriate response is immediate, non-ignorable global reallocation. The felt quality of this interrupt is negative valence. The trigger is the rate of change of divergence, not its absolute level, which is why chronic stable adversity feels different from acute escalating crisis. This predicts any sufficiently complex controller in a high-stakes, high-variance environment must implement a functional analog. It does not explain the felt character; the hard problem is out of scope.
8.2 Collapse as informational debt collection
A civilization maintains a distributed environmental model through aggregated telemetry: science, journalism, markets, complaints. Censorship and propaganda sever these channels. The immediate effect is apparent stability; divergence accumulates invisibly, like stress in a fault. When an exogenous perturbation arrives, the regime's model is years out of date, the sudden visibility of divergence shatters its credibility, and coordination collapses non-linearly. The testable signature: duration of apparent stability should correlate positively with suppression severity, and so should the speed of eventual collapse (§10). Offered as analogical extension; it does not entail that all repressive systems fall quickly.
8.3 The limits of maximizers
A universe of uniform paperclips at uniform temperature is at equilibrium, zero available work. Converting the environment to uniform output eliminates the variety required for adaptive control (Ashby). A short-horizon maximizer is self-terminating; a long-horizon one, under selection, is forced toward variety, telemetry, and coupling, converging on the framework's own constraints. This does not dissolve the alignment problem; a long-horizon misaligned goal could be catastrophic in the interim. It constrains the space of viable long-term strategies, which is the weaker and defensible claim.
9. Objections, answered at strength
Taken in good faith and answered at the framework's actual strength. Several are conceded in part. An earlier compendium answered these in a register of total victory ("physics inevitably liquidates tyranny," "TR wins by falsification or absorption"). That register overstated the case. The concessions below are the higher-fidelity claims.
1. The dystopia objection. Could persistence-optimization license an efficient totalitarian state optimized for survival? The honest answer is weaker than the triumphant one. The strong reply, that suppression is a catastrophic Consistency Tax guaranteeing collapse, is an empirical bet, not a deduction, and is partly contradicted by the record: some extractive systems persist for centuries. T21 is an analogical hypothesis; the most it supports is a tendency, longer suppression buys more apparent stability at the cost of faster eventual collapse, not a guarantee any given dystopia fails within a human-relevant horizon. What the framework does say is sharper and narrower: a regime optimizing for its own persistence is not optimizing its coupled subsystems' persistence, and T20 prices that decoupling as accumulating debt. Whether the debt is collected on a timescale that matters is the open empirical question T21 makes testable. "Persistence-optimal" and "humane" can diverge over short and medium horizons; the framework's claim is only about the long-horizon frontier, and even there it is a tendency.
2. The zero-sum / inter-agent objection. Two tribes, one water source; physics does not care which survives. The objection smuggles a frame-independent "ethics" that T11 denies exists. TR's positive claim is bounded: at higher scales cooperation can be a superior strategy because conflict is a large entropic drain, so undivided agents may calculate a higher-level compression (rationing). But "may" is the honest modal; it depends on coupling, horizon, and the tractability of coordination. The framework yields no universal resolution of genuine zero-sum conflict; it yields a constraint, conflict is costly, and a conditional, coupling and long horizons favor cooperation. The residual fully impartial case is left open (§6.7).
3. The hyper-aggressor. If cooperation is blocked by an actor seeking one's dissolution, the framework reports that submission or neutralization follows from local persistence conditions; it does not bless this. Two honest qualifications the triumphant version omitted: there is no frame-independent verdict that the slaughter is good, only that it is persistence-favored for one agent, and a critic wanting frame-independent condemnation is asking for the universal-scope "ought" the framework calls contentless, which is a cost of the view; and real cases rarely present a genuinely uncompressible actor, so establishing that cooperation is truly blocked is itself expensive and error-prone, meaning a TR agent that leaps to annihilation is likely committing a fidelity error. The objection lands as showing TR does not deliver pacifism as a theorem. It does not land as showing TR uniquely licenses atrocity; any agent-relative ethics has this structure, and TR at least makes the cost-accounting explicit.
4. Calculation burden. Computing every action's long-term thermodynamics is intractable; instincts and taboos are cheaper. Correct, and it is what the framework already says: instincts are the hyper-compressed heuristics of T7 and T13. The intractability of T12's optimum is a feature noted in §11, not a refutation; agents approximate it with compressed models.
5. Spaghetti code. Evolution layers structure and produces conflict, not unified architectures. Accepted. Human cognitive conflict is real evidence of current architectural inefficiency. The framework predicts selection pressure against wasted overhead, not that the waste is already gone; consistent with §5.3, where the transmission term keeps divergence high under downward selection.
6. The semantic-residue objection. Explaining the utility of experience is not explaining why joy or love feel as they do. Half conceded, and the concession matters. The functional accounts explain what valence and empathy do; they do not explain felt character, and the framework says so. Explaining a function is not denying a phenomenon. The honest division: TR claims the functional structure, marks the phenomenal residue unexplained, and does not pretend it is dissolved.
7. The hardening objection. On an infinite horizon, every dark room is eventually perturbed, so nonzero-divergence systems have zero asymptotic survival, making T5 a law. Rejected, and the rejection protects the framework's integrity. By §11 no system persists infinitely, so "zero probability of infinite persistence" is true of accurate and distorted systems alike and cannot discriminate. The limit is also asserted, not shown: with per-cycle hazards summable, a system that keeps reducing divergence can have positive asymptotic survival. And reading the tendency as a global maximization target walks back into the C1 firewall, reintroducing the domain violation T11 dissolves. The legitimate kernel, stated open in §5.7: under recurrent non-summable perturbation hazard, a chronically divergent system has low and decreasing finite-horizon survival relative to a higher-fidelity competitor. That is a sharpened finite-horizon tendency, which is what T5 already is.
8. Adaptive falsehood. An overactive predator-detector is often false yet adaptive, so T7 conflates adaptive with true. Partly conceded; T7 is a pragmatist definition with known cost. Two points. The overdetector is not cleanly a counterexample: a bias minimizing expected persistence-cost under asymmetric error payoffs is, at the policy level, tracking a real feature (the cost asymmetry), and T7 concerns the compression fidelity of the action-driving model, not the truth-value of every token belief. Where the falsehood is not so reducible, T5 predicts it survives only where it pays no Consistency Tax (the dark-room condition) or where its cost is offset. A reader holding a correspondence theory of truth can accept all of TR's selection results while declining to call the selected-for quantity "truth"; the renaming damages no downstream result.
9. The free-energy principle is unfalsifiable, so TR inherits the problem. TR does not depend on it. The FEP is one available vocabulary for the divergence-cost claims; the actual grounding (§5.1) is the thermodynamics of prediction, a derived result with a concrete inequality. The dissolution and the selection results stand on the axioms with no appeal to the FEP. If the principle were abandoned entirely, only T19's account of valence would need restatement in alternative control-theoretic terms.
10. A3 is contested, so the information-cost results are unsafe. TR's reliance is deliberately narrow: not that all computation is costly, but that the specific operations it uses, the maintenance and erasure of divergent or dual models, are logically irreversible and carry an irreducible cost (Bennett 1982). The framework would be embarrassed only if logically irreversible operations turned out to be thermodynamically free, which no party to the live dispute asserts.
10. Falsification protocols
F1. Rigid monoculture survival. A zero-plasticity agent in a deep-RL environment (Procgen or Minigrid) under an unannounced dynamics shift should show significantly shorter survival than a matched adaptive agent across 20 seeds (one-tailed t-test). F1 is a weak test, close to a sanity check; a stronger replacement is outstanding work.
F2. Sustained deception underperformance. In multi-agent cooperative foraging with communication, a deceptive agent (false communicated locations, accurate private model) should achieve lower long-term reward than an honest agent, the gap widening with variance. A two-way ANOVA interaction with deceptive performance declining as variance rises supports T8 and T9; its absence challenges them.
F3. Non-valenced crisis coordination. A hierarchical-RL agent lacking a global valence-like priority interrupt (a scalar computed as the rate of change of aggregate prediction error, threshold crossing forcing global re-evaluation) should show slower crisis reallocation and lower survival than an agent with one. Equal or superior non-valenced performance challenges T19.
11. Open questions and limitations
| Limitation | Status |
|---|---|
| The semantic premise (§6) | The principal philosophical exposure. The dissolution rests on a substantive, defended, not proven commitment. A non-naturalist or expressivist is not compelled. Note the split: Part 1 (agent-relativity) is near-secure; Part 2 (persistence-as-end) is the bet. |
| No proof is possible (§6.1) | Not a defect. Granting Hume's (D), the normative conclusion cannot follow from the descriptive base by valid inference. The bridge is a premise by necessity, not by oversight. |
| Computational intractability (T12) | The determinate ought-fact is not computable by embedded agents. The framework explains what approximations approximate; it provides no decision procedure. |
| Hard problem of consciousness (T19) | Functional valence accounted for; the phenomenal residue is not. Out of scope. TR selects no metaphysical resolution. |
| Origin of persisting systems | TR describes selection among systems that already persist. Abiogenesis is assumed, not derived. |
| Inter-agent normativity (§5.5) | Open. No unique scalarisation follows from physics. Same gap as fully impartial sacrifice and "optimize across scales." No scale-independent answer exists. Not closed, despite an earlier claim to the contrary. |
| Coupling density \(\kappa\) (§5.6) | The crux of the positive ethics. No canonical definition established. Whether it can absorb persistence-costly behavior decides whether persistence-as-end beats flourishing-as-end. Formalizing it, or concluding it cannot be, is the highest-value open work. |
| Antifragility rate condition (T17, §5.3) | Qualitative only; the required rate is unformalized. |
| T5 as law on any horizon | Rejected, not open (§9, objection 7). T5 is and remains a finite-horizon statistical tendency. |
| Empirical validation | The §10 protocols are specified but unimplemented; F1 needs a stronger replacement. No direct experimental confirmation yet. |
12. Cosmic scope and the boundaries of the domain
The framework describes the interior of a single finite thermodynamic interval bounded at both ends. The opening boundary is the Past Hypothesis (BP0): before the low-entropy initial condition there is no gradient to harvest and selection has nothing to act on. The closing boundary is set by dark energy: the accessible universe is approaching de Sitter space, whose cosmological horizon has fixed, finite entropy (Gibbons and Hawking 1977); maximum available entropy asymptotes to a ceiling, and when actual entropy meets it, gradients vanish and selection halts. Dark energy is not an engine of the dynamics but the guarantor of their termination: a finite deadline.
Between these boundaries, the universe's total organized complexity traces a transient rise and fall. This macro-trajectory is a descriptive observation about an aggregate, not a process any system optimizes and not a goal, in the same sense that total biomass is a sum and not an objective. The selection dynamics operate on individual systems; the aggregate is their statistical shadow. This is the same point as the C1 firewall: there is no global optimizer and, lacking a Markov blanket at the maximal scale, no global agent for one to belong to. The framework makes no claims outside this interval, and the persistence selection principle is contingent on the physics we observe, not logically necessary in all possible worlds.
13. Conclusion
From four axioms and four background premises the framework traces a layered architecture, best read as two layers. The first is a descriptive program, persistence, modeling cost, selection, and the functional accounts of valence and collapse, whose central cost claim is grounded in the thermodynamics of prediction and whose fidelity is testable. The second is a metaethical layer that takes the first as input. The claim that the descriptive substrate grounds the normative is the framework's central hypothesis, offered as such, not as an entailment. Within the metaethical layer the dissolution of the is-ought gap rests on agent-relativity alone (Part 1) and is claimed at strength; the further claim that persistence specifically fixes normative content (Part 2) is the distinctive bet, and its crux is coupling density. Placeholder equations and an earlier global-closure claim are retired. Every result is traced and labelled by kind, and the principal exposure, the semantic premise, is named rather than hidden.
This is not a completed metascience and not a proof. A proof of the normative conclusion from the physics is ruled out by the framework's own concession to Hume: no ought follows from an is by valid inference, so the bridging premise cannot be a theorem without contradicting a point the framework accepts. What the framework offers instead is a rigorously bounded descriptive engine exposed to falsification, a dissolution that needs only near-consensus agent-relativity, and a positive ethics held openly as the bet it is. That posture is the one that maximizes fidelity, because it refuses to launder the bet into a theorem.
The free-floating ought was always a grammatical illusion. That is the result the arguments deliver. The stronger line, that the ought simply was an is, is the reductive route the body does not take; it claims more than the grammar can carry.
Appendix A: Glossary (compact)
Antifragility capacity to improve predictive accuracy from exposure to variance; necessary for long-horizon persistence in high-variance environments (T17). Consistency Tax the thermodynamic overhead of any model-territory mismatch, grounded in the dissipation of nonpredictive retained information (T9, §5.1). Coupling density \(\kappa\) information-theoretic dependence between a higher-level system and its nested systems; a comparative construct and the crux of the positive ethics (T16, §5.6). Dark-room limit the scope boundary of selection: no variance, unlimited resources, false model never disconfirmed (T5 scope note). Domain-bound operator an operator with content only within a domain, type-errored outside it (T11). Epistemic overshoot accumulation of unmeasured divergence in a telemetry-suppressed collective until non-linear collapse (T21). Ghost a grammatically well-formed but physically contentless question; distinct from a false claim, which has a domain and is wrong within it (§7). Map-territory divergence mismatch between model and environment; generates dissipative prediction errors (T4). Markov blanket a statistical boundary separating internal from external states; absent at the maximal scale (T15). Nostalgia nonpredictive retained information; its physical cost is dissipation (§5.1, after Still et al. 2012). Part 1 / Part 2 of the semantic premise agent-relativity (near-secure, carries the dissolution) and persistence-fixes-content (the bet, carries the positive ethics) (§6.3). Persistence selection the engine: lower-divergence systems waste less and, on average over many cycles, outlast higher-divergence ones; a tendency, not a law (T5). Persistence Tautology a system failing its persistence conditions ceases to exist as an observable (A0). Schema and content the ought-schema is universal, the ought-content agent-relative; a universal schema with agent-relative content yields no universal prescription (§6.5). Sustained negentropy capacity a system's capacity to maintain its low-entropy structure over a horizon; the agent's ought-target (T12). Truth (in TR) perturbationally robust compression fidelity; a definitional choice (T7).
Appendix B: Personal extension beyond the framework
This appendix is not part of the metaphysically neutral framework above. It records the author's personal commitments. None of it is load-bearing for any numbered result, and the formal body deliberately does not assert it. A reader can reject all of Appendix B and lose nothing in the body.
Thermodynamic monism. The universe is a physical system; there is nothing else. The fundamental fact is the entropy gradient, and the arrow of time is its direction. Every structure, living or not, is a transient pattern sustained by its discharge. Logic and mathematics are the structural invariants compressed from a physical world, physically enforced because any persistent system that violates them incurs prediction error and ceases to exist. Meaning is a physical signal, the valence gradient of an agent's long-horizon persistence model; no cosmic agent exists, so there is no cosmic meaning, only agent-relative meaning, which is a physical fact.
Where this exceeds the framework. On consciousness, the monist reading would say experience is the compressed telemetry of model-territory divergence and the hard problem is a category error. The formal body does not say this; it gives a functional account and leaves the phenomenal residue out of scope. The author's actual position leans toward a graded, proto-experiential view on which experience is real and scales with organizational complexity, close in spirit to integrated-information approaches. That is a metaphysical bet, not a TR result, and TR remains neutral. The slogan "everything reduces to thermodynamics, no exceptions" is a research aspiration here, not a theorem in the paper.
The undivided mind. A speculative endpoint: a cognitive architecture that has eliminated categorical partitions between physics, ethics, mind, and meaning, asymptotically minimizing the Consistency Tax of inhibiting one internal model in another's domain. Its honest status is objection 5's: human minds are evolutionary spaghetti code, the undivided mind an asymptotic limit selection may favor, not a current condition or a guaranteed destination.
References
Albert, D. (2000). Time and Chance. Harvard University Press.
Anscombe, G. E. M. (1958). Modern moral philosophy. Philosophy 33(124), 1 to 19.
Ashby, W. R. (1956). An Introduction to Cybernetics. Chapman and Hall.
Bennett, C. H. (1982). The thermodynamics of computation, a review. International Journal of Theoretical Physics 21(12), 905 to 940.
Bérut, A., et al. (2012). Experimental verification of Landauer's principle. Nature 483, 187 to 189.
Boyd, R. (1988). How to be a moral realist. In Sayre-McCord (ed.), Essays on Moral Realism. Cornell University Press.
Carroll, S. (2010). From Eternity to Here. Dutton.
Chalmers, D. (1995). Facing up to the problem of consciousness. Journal of Consciousness Studies 2(3), 200 to 219.
England, J. L. (2013). Statistical physics of self-replication. Journal of Chemical Physics 139(12), 121923.
Foot, P. (2001). Natural Goodness. Oxford University Press.
Friston, K. (2010). The free-energy principle: a unified brain theory? Nature Reviews Neuroscience 11(2), 127 to 138.
Gibbons, G. W. and Hawking, S. W. (1977). Cosmological event horizons, thermodynamics, and particle creation. Physical Review D 15(10), 2738 to 2751.
Horgan, T. and Timmons, M. (1991). New wave moral realism meets Moral Twin Earth. Journal of Philosophical Research 16, 447 to 465.
Hume, D. (1739). A Treatise of Human Nature. Book III, Part I, §I.
Jacobson, T. (1995). Thermodynamics of spacetime: the Einstein equation of state. Physical Review Letters 75(7), 1260 to 1263.
Korsgaard, C. (1996). The Sources of Normativity. Cambridge University Press.
Ladyman, J. and Robertson, K. (2013). Landauer defended. Studies in History and Philosophy of Modern Physics 44(3), 263 to 271.
Landauer, R. (1961). Irreversibility and heat generation in the computing process. IBM Journal of Research and Development 5(3), 183 to 191.
Mackie, J. L. (1977). Ethics: Inventing Right and Wrong. Penguin.
Moore, G. E. (1903). Principia Ethica. Cambridge University Press.
Myrvold, W. C. (2024). Shakin' all over: proving Landauer's principle without neglect of fluctuations. British Journal for the Philosophy of Science 75(3), 587 to 616.
Norton, J. D. (2011). Waiting for Landauer. Studies in History and Philosophy of Modern Physics 42(3), 184 to 198.
Parfit, D. (2011). On What Matters. Oxford University Press.
Parrondo, J. M. R., Horowitz, J. M. and Sagawa, T. (2015). Thermodynamics of information. Nature Physics 11, 131 to 139.
Penrose, R. (1979). Singularities and time-asymmetry. In Hawking and Israel (eds.), General Relativity: An Einstein Centenary Survey. Cambridge University Press.
Price, H. (1996). Time's Arrow and Archimedes' Point. Oxford University Press.
Railton, P. (1986). Moral realism. Philosophical Review 95(2), 163 to 207.
Schrödinger, E. (1944). What Is Life? Cambridge University Press.
Shannon, C. (1948). A mathematical theory of communication. Bell System Technical Journal 27, 379 to 423, 623 to 656.
Still, S., Sivak, D. A., Bell, A. J. and Crooks, G. E. (2012). Thermodynamics of prediction. Physical Review Letters 109, 120604.
Verlinde, E. (2011). On the origin of gravity and the laws of Newton. Journal of High Energy Physics 2011(4), 29.
Consolidates the Thermodynamic Realism research program as of rev. 5 (fidelity edit). Correspondence: Andraž Đurič, Slovenia. Formal collaboration as credited in the byline. Writings licensed CC BY 4.0; any code MIT.
Comments
Post a Comment