"Are Computers Approaching Human-Level Creativity?"

Seminar #5 of 5: Games, Music, Humor: Can Human Creativity Be Simulated By Machines?


This was sort of the overview (and I didn't attend seminars 1 through 3, one of which was specifically on music). With no Steve Martin to attract the crowds, there were only about 60 people there.

Earlier seminars demonstrated software that could:

  1. imitate a work
  2. imitate a style of work

Human-level understanding is harder, requiring a knowledge of what people expect and the ability to go beyond it. Humans can be "meta-creative"; they can combine higher-level concepts to create meta-styles.

[Q & A]

Assertion from audience: Spielberg has a model of the audience in his head, which continually informs his creative process.

Hofstadter: Agrees that he (Hofstadter) has some sort of model when he writes, although he doesn't know who his audience is. But he's also writing just for himself.

(audience member) Mentions the category of computationally-expenseive computer-science problems knows as "NP-complete" problems, and suggests a category of problems to be called "AI-complete." These would be problems hard enough that solving one in software is sufficient to demonstrate artificial intelligence.

Hofstadter asserts: There is a great deal of intelligence inside a piece of music.

Demo tape: "GenJam" by John Bil, RIT. Generates random music; user tells it "good / bad", and then eventually the user plays along with the software.

Audience member mentions a distinction made in the IBM Speech Recognition group, between "complexity" and "perplexity." Perplexity puts constraints on the combinatorial explosion, corresponding to a model of human expectation (limiting the problem scope enough to make it computable in real time?).

Creative Math vs. "Junk I just made up". Is the "junk" creative? Hofstadter says No, it's not. The correctness adds an additional level of complexity that we can appreciate.

Hofstadter insisted that work "has to have some germ of 'rightness' and eventual value to be called creative." (Some disagreement in audience.)

(audience) "How wrong do you have to be before you're no longer creative? Even a very bad episode of 'Three's Company' is still very creative."

Hofstadter does not have any clue what "Three's Company" is, and nobody tries to explain it to him (although some people are trying not to laugh too loud while considering whether the assertion is true, or just how redundant the phrase "...a very bad episode of Three's Company" is).

Hofstadter: "If it appeals to nobody then it's junk, and it's not creative." Eventually he clarifies that what he meant all along by "rightness" is that it appeals to some audience, rather than being somehow objectively correct.

[ Page last revised 20-Jul-2001. ]

[ www.lyons42.com | email Dave ]