Interface Theory of Perception

The Interface Theory of Perception (Mark, Marion & Hoffman, 2010) starts from a standard evolutionary-game setup. Imagine a population of organisms whose perceptual strategies determine which actions they take in environments whose fitness payoffs are some function of the underlying objective state. Now ask: under what conditions does a perceptual strategy that tracks the truth of that state outperform one that tracks only the fitness payoff? The answer, formalized as the Fitness-Beats-Truth (FBT) theorem, is essentially never. In generic Monte Carlo evolutionary games, truth-tracking strategies are driven to extinction by fitness-tracking strategies with probability approaching one as the state space grows.

The intuition: truth-tracking is expensive — it requires representing structure that has no bearing on the next decision. Fitness-tracking compresses ruthlessly: it represents only what predicts payoff. Across millions of years, the compressed representations win. So our perceptions are interface icons — useful for action, systematically silent about underlying structure. The desktop icon for a file is not a smaller version of the file; it's a control surface that happens to be useful. Hoffman's claim is that 'reality' as we perceive it bears the same relation to whatever is actually there.

Hoffman and Chetan Prakash extended this in 2014 with the Conscious Agent formalism. A conscious agent is a measurable space of possible experiences plus three Markovian channels: perception (world → experience), decision (experience → action), and action (action → world). Agents compose: a network of conscious agents itself behaves as a conscious agent at a higher level. The world an individual agent 'perceives' is the marginal dynamics of the network from that agent's perspective. The formalism is austere — no spacetime, no objects, no observers in the classical sense — and the claim is that spacetime and objects emerge as interface renderings of a substrate that is, fundamentally, a network of interacting conscious agents.

The theory has detractors. Pizlo and others argue from veridicality research in vision that perception does track many objective structures (3D shape constancy, symmetry, etc.). Hoffman's reply: those are interface invariants useful for action; they do not entail that the icons resemble the substrate. Others (Cohen, Dennett) argue the formalism elides the problem of consciousness rather than solving it — that calling the substrate 'conscious agents' relabels the explanandum. The debate is live and methodologically interesting because it has both an empirical face (evolutionary simulations) and a philosophical face (what the formalism implies about ontology).

Whether or not you accept the full ontological move, the perceptual claim is well-supported on independent grounds (predictive coding, sensory substitution, change blindness, attentional gating). The practical takeaway is the same as in Life Is Perfect: take your perceptions seriously as renderings, not as the thing itself.