/ aaron

Machine seeks Creativity.

An interesting idea was presented to me in a paper I was reading about artificial intelligence and creativity, which was that the computer may not have trouble creating the idea, the programming challenge may be more in allowing a system to evaluate the quality of material that it (or anyone else) has generated. Which is to say -- it's easy enough to mash random ideas together, but you'd hardly call throwing random statements into a particular order very interesting or creative, necessarily. This has been my long standing problem with the story generator I've been playing around with, certainly I can throw random words and sentences into an order but nine times out of ten they aren't particularly novel or even sensical. A system that can evaluate whether an idea is good or not is probably closer to creative.

Two weeks ago I talked about the two different aspects of creativity, the conception and the execution, and really the evaluation of the effectiveness of a creative spark can occur in either -- you can either have a novel concept or execute something more mundane in a novel way, though it's difficult either way to tell if this is good novel or arbitrary nonsense novel.

A major problem is generally in this evaluation process is that even people can't always tell with anything even approaching regularity or objectivity what a great creative idea is. Some come closer to universal that others, but there is some grey area. Ask the opinion of any group of people as to the artistic merits of highly abstract modern art and you'll probably get jarringly divergent opinions on drawings of very obviously oblique squares in bright colours.

I've been trying to find a program that currently exists that is set to identify the so called "creative" idea, and there isn't yet much in this direction. Among the more interesting has been software that we use to try to detect humour. Culminating in this paper on a system designed to identify whether or not the tired joke "That's What She Said" can be pulled out of a given sentence[1]. Yes, I'm serious. I haven't found much in the way of a general creativity detector.

I've been thinking awhile about how I identify something as creative, and I admit it isn't that easy. If I read a book that takes some adventures with structure and language and I would call that nearly as creative as one that contains an interesting storyline. I appreciate that piecing together random things is presumably how humans become creative but we are somehow (slightly) better at picking out ideas which are novel and interesting from the ones that are mere nonsense. My instinct is that it has something to do with the way ideas are connected rather than the actual ideas themselves, which is to say, certainly two thoughts are random but if they piece together right then we have created a new idea.

We can take a quick interlude to talk about a major way in which we have creative programs right now which is somewhat "template-based" systems like the JAPE joke system[2], and AARON the illustration machine[3], and even the awesome Emmy who composes classical music[4]. Generally speaking these systems have a very constrained domain and then experiment with the variables, but what if we want a system that works counter even to the structure without looping back into nonsense?


  1. That's what she said. BOOM. ↩︎

  2. Kim Binsted, Graeme Ritchie (1994) "A symbolic description of punning riddles and its computer implementation." Research Paper 688, University of Edinburgh, Edinburgh, Scotland, 1994, reported at the International Conference on Humor and Laughter, Luxembourg, 1993 ↩︎

  3. http://www.stanford.edu/group/SHR/4-2/text/cohen.html ↩︎

  4. http://www.psmag.com/culture/triumph-of-the-cyborg-composer-8507/ ↩︎