ArXiv 2610.12845 posts formal creativity postulates using newness, value, surprise validated on chess and paintings

The paper supplies three postulates drawn from Boden's criteria and converts them into a single formula that scores outputs by deviation from prior distributions, utility within task constraints, and statistical unexpectedness. Validation runs compare the formula against expert-rated chess positions from grandmaster games and feature-vector distances in Picasso, Renoir, and Braque canvases; both domains show rank correlation above 0.7 with human judgments.

Existing computational-creativity work either stays informal or applies untested metrics; this submission closes that gap by requiring domain-specific measurement functions while preserving the same three-term structure. The approach therefore scales across discrete search spaces such as games and continuous feature spaces such as images without rewriting the core definition.

Operational consequence is immediate: reinforcement-learning reward models and generative-art pipelines can now insert the formula as an auxiliary objective, replacing proxy novelty bonuses that ignore value. Follow-on experiments can test transfer to molecular design or theorem proving once domain value functions are supplied.

Next required step is release of the exact scoring code and the labeled chess-art datasets so independent groups can replicate the reported correlations within six months.