I went to the AI Forum meeting today, here at EECS/UCF. The topic is about creativity on computers. As the computers are not really perceived as creative, such discussions always end up steering towards creativity in art, advertising and propping the door open (*).

But, then, surprise, surprise: a much more technical definition arises from Ken Stanley : creative is what impresses **me**. That is, a solution is creative if it makes me say: “Gee, that would have never occurred to me”.

This sounds relativistic and philosophical, but in fact it is easy to formalize. Let us assume that we have a search problem and a search algorithm H after time deltaT did not find a solution. Then, the algorithm is presented with a solution c, which we assume that the algorithms can verify.

Questio: how can the algorithm establish whether it would have ever found that solution of if it would have taken a very long time? I have a feeling that in the extreme this problem might prove to be just as complex as the search itself. But one can imagine some simple approximation algorithms: a solution is creative if it can not be reached with the existing operators, if it is found on paths which run against the heuristics, if it has steps which are very far from gradient descent, and so on. This way, you might even determine which steps in the solution are creative and so on.

So there are no creative solutions in abstract. The creativity is always relative to a particular heuristic. There are no creative solutions, only bad heuristics. In the land of supermen, Bicycle Repairman is the creative one.

(*) At UCF, propping the door open is almost exclusively done with the garbage can; not much creativity there.