Researchers Propose a 3-Criterion Formula for Creativity and Test It on Chess and Picasso

For decades, computational creativity has lacked a formal definition that works equally well for art, science and games. The preprint arXiv:2610.12845 (cs.AI, October 12, 2026) tries to close that gap. The authors formulate several postulates based on the psychology of creativity and derive a mathematical formula with three components from them: novelty, value and surprise. They test the formula on two tasks: rating the creativity of chess moves and of paintings by Picasso, Renoir and other artists.

Three Criteria Instead of Vague Words

The authors describe a two-part problem with existing approaches. Some works offer informal, conceptual definitions that cannot be turned into a calculation. Others propose formulas but do not examine the assumptions built into them, and do not test the results on concrete applied tasks. The formula looks elegant, but whether it works remains unknown.

This paper reverses that order. First come postulates drawn from psychological research, above all Margaret Boden's three basic criteria. Then the authors build a definition that satisfies those postulates. Only after that does the empirical testing begin.

Key elements of the work:

• Foundation: psychological research on creativity and Boden's criteria.
• Three components: novelty (newness), value, surprise.
• Structure: a single mathematical construct, with the specific ways of measuring each component adapted to the domain.
• Validation: two domains, chess moves and painting.
• Claimed result: the formula "successfully captures" creative patterns in both domains.

Chess and Painting as a Testing Ground

The choice of domains looks sensible. Chess offers a clearly defined space of moves, where it is easy to talk about novelty relative to known continuations and about a move's value for the outcome of the game. That reasoning is ours, not a quote from the paper. Painting sits at the opposite pole: value there is subjective, and surprise depends on the artistic context of the era. If one structure works at both poles, that is a strong argument for the idea of a single mathematical framework.

That is exactly what the authors claim. In their conclusion, creativity can be described by a common mathematical structure built on three fundamental criteria, and only the ways of measuring them differ from one domain to another.

An important caveat: we have only the abstract. It contains no numerical metrics, no comparison with alternative definitions, and no description of how novelty and surprise were actually calculated for moves and paintings. It also says nothing about tests on language or generative models. This is a first-version (v1) preprint without confirmed peer review, so the phrase "successfully captures" should be read as the authors' claim, not as an established fact.

How This Could Be Used in Practice

For teams that evaluate generative systems, the paper's value so far lies in its framework, not in a ready-made tool. Today, model creativity is most often measured indirectly: through output diversity, rater preferences or pairwise comparisons. A three-axis scheme suggests splitting the evaluation into three questions. How much does the result differ from what is already known? Is it useful for the task? How far does it diverge from expectation?

What to do if the topic is relevant to you:

• Read the full text and check how the components are defined in each domain and which assumptions are built into them.
• Try breaking down your own quality metrics into novelty, value and surprise, and see which axis you have not covered.
• Do not apply the formula to text, code or scientific hypotheses without separate validation: the authors demonstrated it only on chess and painting.
• Follow the continuation of the work. Testing on modern generative models would be the next logical step, and it will show whether the formula is suitable for industry.