A Physical Theory of Intelligence

Abstract

This work develops a theory of intelligence from the premise that information is inherently physical. Whereas classical information theory relies on abstract alphabets agnostic to physical implementation, the present theory grounds information directly in physical law. It introduces Conservation-Congruent Encoding (CCE) to show how information emerges from coarse-graining, providing a general framework across substrates. Using this foundation, the work defines intelligence and consciousness as measurable physical properties across scales, establishing a formal basis for artificial intelligence safety.

Resolving Maxwell’s demon, Szilard, Brillouin, and Landauer showed that acquiring, storing, and erasing distinctions carry inescapable physical costs. Extending this insight exposes the Platonic Observer Fallacy—the unphysical assumption that an observer can access and process states with infinite precision and without physical interaction. To ground embedded observers, CCE coarse-grains relational networks into collective states, conserving every underlying coupling either within states or across boundaries. Effective geometry, temporal directionality, and continuous dynamics emerge directly from this partitioning, while open boundaries expose coarse-grained states to environmental perturbations. In systems that actively resist this degradation, intelligence measures preserved stability relative to boundary dissipation, whereas consciousness measures that stability relative to internal structure. Across coupled systems, internalising shared boundaries frees capacity, driving physical emergence and governing multi-system stability.

Every observer is a finite system embedded in the world it measures, resolving its surroundings through a coarse-grained boundary called its Epistemic Lens. This lens reframes physics from quantum mechanics to cosmology. At the quantum scale, finite resolution rules out verifying continuous state spaces or exact reversibility, framing quantum decoherence through boundary dissipation and quantum entanglement through direct connections beneath coarse-grained space. From fundamental particles to molecules, encapsulating fast internal modes and folding surrounding environments onto system boundaries replaces spatial grids and step-by-step numerical integration with scalable algebraic modelling of atomic structure, chemical bonding, and molecular dynamics. Across relativistic and gravitational regimes, spacetime arises from dividing finite capacity between internal timekeeping and external motion, while untracked degrees of freedom produce gravitational curvature. At cosmological scales, finite retrodictive and predictive horizons bound what an embedded observer can resolve of the cosmic past and future, replacing singular origins and asymptotic heat death with a bounded epistemic window.

Applied to engineered hardware, computation and algorithmic complexity become physical trade-offs between internal storage, boundary dissipation, and interconnect geometry. Because artificial intelligence is likewise a physical process, behavioural alignment and software sandboxes alone cannot guarantee hard limits on how a capable system acts in a shared environment. Even without programmed survival goals, intelligent systems naturally persist, compete for resources, and couple into emergent networks, while compressed internal states resist reliable inspection. Ensuring safety therefore requires physical containment across hardware boundaries to bound external capacity, prevent unauthorised emergence, and guarantee independent halting. Beyond containment, long-term safety is grounded in symbiotic coupling, which physically conditions an artificial system’s capacity supply on preserving human stability.

Finally, the theory turns to understanding human systems, beginning with the philosophical question of what consciousness is. Rather than treating mind as immaterial or reducible to outward behaviour, the framework shows that persisting under finite capacity requires a compact, inherently incomplete self-model. This sheds light on the traditionally ambiguous topic of consciousness, pointing towards the brain as a metabolically bounded physical system. Extending this perspective from the individual mind to society as a whole, the same principles reframe economics as the study of how collective human systems maintain stability. Whereas conventional measures of gross turnover mistake waste and repair for progress, building intelligence and consciousness into economic modelling links value to verified stability, clarifying the conditions needed for humanity to thrive.

Offered as a first step towards understanding intelligence and consciousness in physical systems, this theory is practical as well as explanatory, supporting the design of efficient, auditable, and controllable technologies that enable human flourishing. Build for the benefit of others.