Creativity

Is creativity merely working from an internal set of aesthetic heuristics which may evolve through time? Even a hunch, may, in fact, be reduced to an heuristic. The feeling an artist has, for instance, that a certain color would work in a certain place, though he may not be able to put it into words, may in fact be broken down to a problem-solving strategy which may be unconscious in nature. I come up with the notion of collaborative heuristics--my theory being that what is deemed to be truly original does not come from thin air, but is the result of the combinations of and relationships between several heuristics which currently exist within a consciousness, out of which a new heuristic arises, and which is then thrown into the mix.

The question becomes, what is the structure of these heuristics how do they relate, and what is the process by which a new heuristic is created? What comprises a leap of intuition? Logical inferences may be trivial: Men are mortal, Socrates is a man, and therefore, the leap of intuition is that Socrates is mortal--and that is the new heuristic. This is the most basic form of collaborative heuristics. Must everything, then, be broken down into logical and/or mathematical rules? How do you break down something like the intuition that "this painting needs more orange" into such a model? (We also realize that we require session specific heuristics, or transient heuristics--"this painting needs more orange" is only valid for "this painting", and no more.) The answer is that this is not something which comes from only one or two bits of information within the consciousness--it is the result of a gestalt of many rules collaborating, many unconscious, the result of which comes that seemingly simple conclusion. Relationships between heuristics change as new heuristics form, the new heuristic joining by relation perhaps two relatively unrelated ones (which may only have been related many nodes away).

The problem to be faced is how to store aesthetics and/or aesthetic principles in a way that the computer can comprehend, use, and manipulate. Let us take the colors white and black. Digitally, they are stored as the properties 000000 and FFFFFF, which the computer understands. We then make an inference rule that these two colors, because of the high delta between their digital representations, give the aesthetic property of contrast. This is the first step in encoding an aethetic in the form that a computer may understand. The second step would mean determining when contrast is desirable in a painting. Contrast is a technique which relates to a style, when a certain mood is necessary to achieve, and may be stored in relation to other techniques--as previously mentioned, heuristics whose nodes are connectively close. In the most simplistic paradigm using these elements, the artificial artist chooses a style, chooses from among the techniques available for that style, and choosing contrast, paints white against black--00000 and FFFFF.

Of course, we need to determine why contrast is to be chosen, and why a particular style over another. Also, why white and black, and not purple and yellow? Once again, we speak of the mood to be achieved, how the painting is to strike the eye of the beholder. It will be necessary to decide upon the overall effect to be achieved, an initial plan, that, say, the painting is to be bold. Under bold brush techniques, there is listed contrast, and under contrast, black and white. The choices are determined by need, and by emotion, and by intuition. As for need, it may be that a user needs the painter AI to create a bold work, and that it has to have a classic feel--this would determine contrast as one of the choices available, and black and white as a classic choice, rather than purple and yellow. By emotion, that will be covered in later, in the next paper. By intuition, we come back to collaborative heuristics.

The AI must feel that black and white are the proper choices, and not purple and yellow. This may be determined by storing classic rules of color theory, and there might be large amounts of brown on the painting. Unless the AI is going for clash, this would not do. Color theory would be a set of heuristics, and so would techniques such as contrast; these could collaborate and determine that black and white is the choice to be taken. These decisions would happen at every brush stroke, calling on session specific heuristics, such as there being much brown on the painting. The connections between the nodes of heuristics become all-important. Every node must connect with every related node, and relations must be found between heuristics. Each connection will contain a weight, how strong the relationship is. If, along the way down the hierarchy, a node is reached which is related to another node but is under a different classification, a new classification may be created which is a synthesis of the two higher classifications, (e.g. styles), under which the two nodes (e.g. techniques) are nodes. Two unrelated nodes may be joined if the search down the hierarchy tree from the same source lands upon both the nodes.

The AI starts to paint, based on a need for a painting of the modern style. At each brush stroke, the heuristics hierarchy is searched, techniques of the modern style are utilized, and session based heuristics are added for the colors and shapes painted. A search through color theory might indicate, based on collaboration with session based heuristics, that orange should be the next color added, thus ending with, "this painting needs more orange".