TH-COMP-005 — Active Inference Composition Theory

Statement

A composition is an evidence environment for an active perceiver. It shapes the initial hypotheses formed from a glance, the uncertainty that remains, the locations sampled next, the relationships verified, and the actions made available.

Mechanism

1. Goals, expectations, and learned conventions establish priors.
2. Coarse features support rapid, partly parallel scene and object hypotheses.
3. Grouping constructs provisional perceptual units.
4. Salience, task relevance, history, and uncertainty establish priorities.
5. Attention and action systems select the next evidence sample.
6. Eye, head, body, or interface actions change the available input.
7. Recurrent processing integrates evidence and revises the current model.
8. The model stabilizes enough for action or remains open for more sampling.

Boundary Conditions

• Some rapid judgments require only one brief exposure.
• Not all prediction effects alter phenomenology.
• Not all recurrent processing is conscious.
• The theory does not require a unique predictive-coding neural implementation.
• The theory predicts task performance, not aesthetic preference by itself.

Falsifiable Predictions

• Brief or masked exposure preserves coarse hierarchy better than relation and state understanding.
• Context benefits ambiguous targets more than high-reliability targets and can reverse for clear anomalies.
• Task changes alter fixation sequences even when the visual stimulus is fixed.
• Peripheral fine or crowded targets cause additional saccades, longer search, and more errors.
• Stable visual correspondence across state changes reduces reorientation cost.

Evidence

Supported by EV-COMP-051 through EV-COMP-060 and constrained by contradictory evidence on expectation suppression, unconscious integration, and context-congruency effects.

Related Records

• Parent package: RP-COMP-005
• Review framework: DF-COMP-002
• Tests: EX-COMP-011 and EX-COMP-012