To produce “dynamic artistic practice” AARON would have to create AARONs. It would have to do what Harold Cohen does; set immediate and long term goals, create systems, evaluate results, reflect on its achievements and work this meta-knowledge into its goals and systems.
This sounds like the sort of goalpost-moving that has always plagued AI. If a computer can play chess then chess playing must not be a sign of intelligence… But “dynamic artistic practice” would be relatively simple to achieve given a system of axiomatic graphical microdomains and a Lisp-style function composition system.
The microdomain approach is the one that Douglas Hofstadter and FARG took: find a simple task that is representative of an aspect of creative behaviour and then analyze it in depth.
AARON, like any painter, just pushes pigments around in 2D. It’s a historically sufficient aesthetic domain in which to create and evaluate constraints or axioms.
It may even generate historically novel techical practice, Margaret Boden’s “h-creativity” from The Creative Mind. But this wouldn’t be necessary to satisfy the requirements of “dynamic artistic production”, as Damien Hirst’s appropriations show.
So I’m not making a ridiculous demand of an AARON-like system. This behaviour could be shown by a fairly simple system (like Copycat), although to be aesthetically as well as conceptually satisfying it would need to be more complex.
(From an answer to a comment by Yaxu below.)
Technorati Tags: [art computing](http://www.technorati.com/tag/art computing), [generative art](http://www.technorati.com/tag/generative art)