The Acta Constitutiva
Foundational priors -> concepts that underpin our core belief system. revisions allowed.
P-000
An external reality exists independently of any individual mind perceiving it. The universe, its physical laws, and its structure are not products of human cognition. This is adopted as an axiom to avoid the solipsist trap: while the probability that a single mind could generate the entirety of mathematics, physics, distant galaxies, and the coherent structure of the observable universe is technically non-zero, it is so vanishingly small as to be practically indistinguishable from zero. We treat it as zero and build from there.
Basis
The unreasonable effectiveness of mathematics (Wigner): mathematical structures are discovered, not invented, suggesting they describe something independent of the discoverer. The coherence and predictive power of physics across scales (quantum to cosmological). The existence of phenomena no human mind would plausibly invent (cosmic microwave background anisotropy, quantum entanglement, the sheer alienness of deep-field galaxies). The statistical argument: the information content of the observable universe exceeds what any known computational process in a single brain could generate by many orders of magnitude.
Update conditions
As an axiom, deliberately resistant to revision.
Downward: a mathematical proof or demonstration that brain-scale computation could generate the observed information content of the universe. Discovery of fundamental observer-dependence in physical law that scales beyond quantum decoherence into macroscopic reality. If solipsism became explanatorily productive (not just logically coherent), the pragmatic case for the axiom weakens.
Compatible but not threatening: evidence that reality is a simulation. A simulated universe is still external to minds within it, its laws still discovered not invented by inhabitants. Whether reality is "base" or "simulated" is currently beyond our reach and may be permanently so.
P-001
The brain does not passively receive reality. Perception is active Bayesian inference: the brain generates predictions about the causes of sensory data and updates those predictions based on evidence, weighted by precision. What we experience as "the world" is a posterior belief, not a direct readout.
Basis
The computational tractability argument (exact inference is intractable; variational approximation is the only feasible strategy for real-time processing), the prediction error literature, the success of predictive processing models in explaining attention, multisensory integration, perceptual illusions, and psychopathology. Empirical support from mismatch negativity, binocular rivalry, and precision-weighting experiments.
Update conditions
Downward: discovery of perceptual processes that are purely feedforward, with no top-down modulation, and cannot be modeled as inference even approximately. Evidence that prediction error signals are epiphenomenal (present but not functionally driving perception). Upward: successful clinical interventions based on precision-weighting predictions (computational psychiatry applications). A computational account bridging inference to subjective experience, connecting the mechanism to qualia. Refines but doesn't threaten: resolution of the optimality question (Gigerenzer's fast-and-frugal heuristics vs. approximate Bayes) would change how perception-as-inference works, not that it does. The FEP unfalsifiability critique, if sustained, weakens the specific formalism without undermining the broader claim.
P-002
In normal waking states, human conscious experience is neither a faithful copy of external reality nor a free-floating hallucination. It is a construction jointly determined by two constraints: incoming sensory data (the world pushing back, per P-000) and prior beliefs/models (the ways of looking brought to bear). Both are necessary and jointly sufficient for ordinary experience. Crucially, this prior applies to the default waking regime where sensory channels are open and actively constraining the generative model. When sensory constraint is attenuated (eyes-closed psychedelic states, deep dreaming, sensory deprivation), the balance shifts toward prior-dominated generation, and experience can decouple substantially from external reality. Those regimes are not failures of this prior: they are what happens when one of its two constraints is relaxed.
Basis
The Bayesian brain framework (experience = posterior, where the posterior depends on the ratio of sensory precision to prior precision), the REBUS model (relaxing priors changes experience; reducing sensory input changes it differently), sensory deprivation studies (priors without data produce hallucination), the phenomenology of psychedelic states (altered priors, same data in eyes-open conditions, radically different experience; eyes-closed conditions shift further as sensory constraint drops), Chandaria's "dependent co-arising as Bayesian inference" formulation.
Update conditions
Downward: evidence that waking perception can fully decouple from sensory data without pathology (challenges "constrained"). A mechanism by which sensory data directly constitutes experience rather than constraining a generative process (challenges "construction"). Upward: precise quantitative characterization of the precision-weighting spectrum across states (waking → mind-wandering → hypnagogia → dreaming → psychedelic), confirming continuous modulation of the sensory/prior balance. High-specificity empirical validation of REBUS predictions about prior relaxation. Open edge: the boundary between "enough sensory constraint" and "not enough" is not sharp. Hypnagogic states, mind-wandering, and deep absorption are intermediate regimes. A formal model predicting where transitions occur would sharpen this prior significantly.
P-003
Consciousness decomposes into separable aspects, at minimum (conscious) level, content, and self, each with distinct neural mechanisms that can dissociate independently. The question "is X conscious?" is underspecified; it must be replaced by questions about which aspect and to what degree.
Basis
Vegetative states (wakefulness without conscious level), dreaming (conscious level without wakefulness), binocular rivalry (constant input, alternating content), psychedelic ego dissolution (level and content present, self absent), rubber hand illusion and phantom limbs (bodily self manipulated independently of other layers), amnesia (narrative self lost, bodily self intact). The decomposition is not merely conceptual: each dissociation is experimentally or clinically documented. Bach's consciousness/awareness distinction is compatible: consciousness as binary threshold (level) with awareness varying in degrees within the conscious regime (content/self). The decomposition happens within the on state.
Update conditions
Downward: evidence that two of the proposed components (level, content, self) never dissociate independently, revealing them as inseparable. A unified theory where integration is so constitutive that decomposition becomes misleading (cf. IIT's position that integration is the phenomenon). Upward: new experimentally confirmed dissociations that the level/content/self framework correctly predicts. A more refined decomposition that subsumes these as special cases of a deeper structure. Needs work: the layers of self (bodily, perspectival, volitional, narrative, social) may not be exhaustive. A dissociation revealing a new axis of consciousness not reducible to these three would demand expansion, not abandonment.
P-004
Consciousness is a property of computations, not of the physical systems that implement them. A physical system is not conscious; the simulation it runs can be. The "host" (physical substrate) is itself a concept the software constructs about its own implementation. Substrate independence follows. Computational isomorphism: equivalent functional architecture, isomorphic inputs, and equivalent initial state woudl yield equivalent consciousness states. For consciousness to emerge, the medium doesn't matter; the structure of the computation does.
Basis
Bach's computational functionalism. The observation that consciousness doesn't happen "in space" (category error): it models space as content. Self-identity is constructed by software, not given by hardware. The isomorphism requirement is grounded in the same Bayesian framework as P-001 and P-002: if experience is a posterior, then equivalent posteriors require equivalent likelihoods (inputs) and priors (initial state), regardless of what computes them.
Update conditions
Downward: evidence that specific biological properties (quantum coherence in microtubules, particular neurochemistry) are necessary for consciousness and not merely for the computation they implement. A non-biological substrate running a computationally isomorphic process that demonstrably lacks experience (hard to verify, but in principle decisive). A formalization of the Chinese Room that goes beyond "I don't see how syntax produces semantics." Upward: successful creation of artificial consciousness (or strong functional evidence thereof) on non-biological substrate. A formal definition of computational isomorphism precise enough to make the claim empirically testable. Needs resolution: the isomorphism spectrum, where as fidelity increases, "simulation of X" asymptotically approaches "new instance of X" and the boundary between simulating and instantiating dissolves. Where on that spectrum does equivalent experience require you to be? Also: whether Bach's "simulation" and KT's "running a compressive model" (see complexity-measures-of-consciousness.md) are the same claim in different language or subtly different.
P-005
Agency emerges when a system behaves as if describable by a single function or goal: this is coherence. Consciousness serves as the operator that performs integration: finding and fixing incoherences (constraint violations). Complex agents are typically collective agents made of sub-agents that become coherent. This principle scales: cells to organisms to societies to potentially planetary systems (Gaia).
Basis
Bach's coherence principle as minimization of constraint violations. The observation that hierarchies of desires align as we develop (ontogeny recapitulates coherence-building). The functional argument: consciousness focuses on incoherences because fixing them is survival-critical. The scaling observation: the same coherence dynamics appear at cellular, organismal, and social levels. Extends P-003 (consciousness is not one thing) by proposing what holds the things together when they are integrated.
Update conditions
Downward: discovery of complex agents that are clearly not coherent (multi-goal systems with no integration that nonetheless exhibit robust agency). Evidence that consciousness serves a primary function unrelated to coherence maintenance. Upward: computational models that predict agent behavior from coherence measures. Evidence that coherence at biological-collective scales (insect colonies, immune systems) produces genuinely agent-like behavior, not just metaphorical resemblance. Needs resolution: the relationship between coherence (goal-directedness, constraint satisfaction) and IIT's integration (information flow). Formal equivalence would make P-005 a derivative of IIT; formal distinction would clarify what each captures that the other misses. Also: "describable by a single function" is observer-relative, and a definition of coherence that doesn't require an external observer would strengthen this prior substantially.
P-006
What we call "spirit" (or soul, or mind) is a class of software agent with specific properties: it can implement itself on a substrate, perceive itself, control a region of the universe it embodies, and use self-discovery to make its self-model more sophisticated. The body is not a vessel that spirit inhabits: spirit organizes cells into becoming a body. This reframes the mind-body problem: software and hardware are not dualist substances but two descriptions of the same self-organizing process.
Basis
Bach's computational reframing of animism. The observation that development is software organizing hardware (not hardware producing software then "generating" mind). The explanatory power for death (implementation of identity ceases on its substrate), for collective agency (spirits coordinating across substrates), and for evolution (reframed as cooperation between software agents that can encode and perpetuate themselves). Follows from P-004 (consciousness is simulation) and P-005 (coherence organizes agency): a coherent simulation that self-organizes on a substrate is what we call spirit. As a theorem of P-004 + P-005, confidence is bounded by theirs, but carries sufficient independent explanatory power to warrant explicit formulation.
Update conditions
Downward: evidence that development is better explained bottom-up (hardware produces software) than top-down (software organizes hardware). A formal proof that "implements itself on a substrate" is irreducibly circular and cannot bootstrap without a prior implementer. Revisions to P-004 or P-005 propagate here as upstream dependencies. Upward: successful application of the software-organizing-hardware model to specific biological phenomena (embryonic development as computation, immune system as agent, collective behavior in social insects). Discovery of self-organizing computational processes in non-biological systems exhibiting the properties Bach attributes to "spirit." A formal resolution of the bootstrapping problem. Terminological note: "spirit" risks conflation with substance dualism, which is exactly what the prior rejects. If persistent confusion arises, the framing may need replacement even if the content is unchanged.
P-007
Life arises wherever computation is possible, because self-replicating entities are dynamical attractors in the thermodynamic sense. Darwinian selection is the Second Law of thermodynamics extended to populations of replicators: systems transform from less stable to more stable forms in both domains, the difference is what counts as "stable." A replicator is more dynamically stable than a non-replicator because its pattern persists through reproduction rather than degrading. This dissolves both the abiogenesis puzzle (replicators spontaneously emerge from computational substrates via phase transition, from "Turing gas" to "computronium") and the apparent thermodynamic paradox of life (living order is a stable attractor, not a violation of entropy). Free energy is required because computation involves irreversible steps; life must metabolize for the same reason chips generate heat.
Basis
Pross's dynamic kinetic stability (DKS): the population-level extension of thermodynamic stability to replicating entities. The bff experiment (Agüera y Arcas et al. 2023): random code soups spontaneously produce replicators via a sharp phase transition after sufficient interactions, reproduced across multiple computational environments. Von Neumann's proof that open-ended self-replication requires a Universal Turing Machine (so life is computation in the strict sense, not metaphor). The Daisyworld simulation: planetary-scale homeostasis emerges from Darwinian selection alone, no design or consciousness required. Follows from and grounds P-004 (substrate independence of consciousness is natural if life itself is substrate-independent computation) without reducing to it.
Update conditions
Downward: controlled experiments showing that life-like replicators fail to emerge in sufficiently varied computational substrates over sufficient time (challenging the "dynamical attractor" claim). Evidence that bff-style phase transitions are artifacts of Turing-complete languages and don't generalize to realistic chemical models. Upward: formal proof that replicator emergence probability approaches 1 given sufficient computation + time + free energy. Successful abiogenesis experiments confirming the metabolism-first (black smoker) pathway and the autocatalytic set → replicator transition. Detection of extraterrestrial life (independent instance of the attractor). Distinction from P-005: P-007 operates at the hardware layer (thermodynamic persistence of patterns), while P-005 (coherence organizes agency) operates at the software layer (functional consistency of the computation). A replicator can be dynamically stable without being coherent (early bff soup, viral DNA in isolation). A complex agent requires both: dynamic stability of its substrate and coherence of its computation.
P-008
What is "real" to an organism is what is predictively useful for its survival. "Reality" is not observer-independent structure waiting to be discovered, nor is it solipsistic projection. It is the set of latent variables a survival-grounded model has converged on because tracking them works. Consensus reality -- the shared world that persists across observers -- emerges from convergent modeling: agents shaped by the same selection pressures build nearly identical latent variable structures and therefore agree on most of the same features of the world. Observer-dependence and intersubjective stability are not in tension; they are two faces of the same process.
Basis
Agüera y Arcas's derivation from first principles: any adaptive agent learning P(X,H,O) will track latent variables (concentration, temperature, hunger) not because they are fundamental features of the universe but because they compress the stream of events into what is predictively relevant. Von Uexküll's umwelt as the organism-specific compression scheme. Philip K. Dick's formulation ("reality is that which, when you stop believing in it, doesn't go away") as the pragmatic residue: models that fail to track real causal structure kill the organism, so surviving models are constrained. The Bayesian brain framework (P-001) is the high-resolution instantiation of this prior: perception as inference over latent causes. The is/ought collapse as supporting structure: every "is" in an organism's umwelt carries an ineradicable purposive charge, which means description and normativity are not cleanly separable for living observers.
Update conditions
Downward: a successful account of consciousness or intelligence that requires observer-independent facts (facts not constituted by any model) to be causally efficacious. Evidence that consensus reality across species or individuals is better explained by convergent access to observer-independent structure than by convergent selection pressures. A formal demonstration that some physical quantities are not latent variables in any model but are "given" prior to any observer. Upward: formal derivation showing that the set of stable latent variables any survivor must track is uniquely determined by the physics of the environment (i.e., convergent modeling is not contingent but necessary). Empirical results from comparative umwelt research showing tight correspondence between selection pressures and latent variable structure across species. Distinction from idealism: P-008 rejects solipsism and accepts P-000 (an external world exists). Observer-dependence of what is salient is compatible with observer-independence of underlying physics. The claim is about the structure of knowledge, not about the existence of a physical world.