A Self-Defining Game for One Player: On the Nature of Creativity and the Possibility of Creative Computer Programs

Abstract

The AARON program has been generating original art-works for almost 30 years, but is denied by its own author to be creative. The author characterizes creativity as a directed movement towards an ill-defined but strongly felt end-state for the individual's work as a whole, not as a characteristic of any single work and profoundly knowledge-based in the sense of externalizing the individual's internal world-model and system of belief. He suggests that a creative program would be one that was able to modify the belief-based criteria that inform the rule-base in which expert knowledge is represented, not one that is able simply to modify the rule-base itself.

I once made a joke at a cocktail party to the effect that I would be the first artist in history to have a posthumous exhibition of new work. I should be more careful of what I say at cocktail parties. The joke has been quoted-though not, need I say, actually discussed-much more frequently than anything serious I have ever said.

I was referring, of course, to the fact that my computer program, AARON, is currently capable of generating about a quarter of a million unique, original images every year from now to eternity and, with computing power increasing over time, could soon be providing several new, original paintings a year for everyone on the planet if it were, in fact, generating all those images on paper, which fortunately it is not. It uses a mechanical painting machine to generate output in the real world, and I don't see its real world output ever getting much beyond one large painting per day. A good thing, too.

Note that I used the term "new, original images," not "creative." I use the word "creative," on those rare occasions when I use it at all, to refer to the ability of the individual-human right now, program potentially-to move forward, to develop, to introduce new material. To put it more precisely, I believe the word properly attaches to continuous change, not to single events. There is no question that AARON has moved forward and developed over its 30-year existence, but the agency of change and development has been me, not AARON. Unless it can pick up from where I leave off, developing new knowledge and new levels of capability for itself, it will go on generating images that are original and different one from the other-in the sense that any two faces in the human population are different from each other-but nevertheless they are all the same in the sense that there will have been no further development, no new material introduced.

This leaves an open question then: whether we will ever be able to claim that a computer program is creative. What would be involved in giving a program that capacity? What is creativity actually like?

If we survey the work of any major artist, we get the distinct impression of someone who knew exactly what he or she was doing and knew exactly where he or she was going. Mozart always sounds like Mozart. Matisse did not produce a Picasso on Monday and a Miró on Thursday, he produced Matisses every day. No doubt his admirers anticipated that whatever he did would be unmistakable as a Matisse, but they could never predict by looking at what he did one day what he would do the next day or next year, when some unexpected new element or quality would eventually appear that did not follow so obviously from what went before that we-or, we may suspect, the artist-could have predicted it.

Michelangelo, according to some accounts, said that the figure was inside the block of stone and that all he had to do was to remove the superfluous material. Picasso said, "I don't seek, I find." There's a curious similarity to these attitudes in the popularized view of the scientist's "search space" as a place where all possibilities pre-exist and...