AI in games

Sat in on this extremely interesting talk today by Alex Whittaker.. a more detailed version of this one. Alex works for Eidos, which is a largish game company and publisher (Tomb Raider, Commandos, Deus X, Thief among many others). The first thing, I was under the impression that all the big guns used simple rule based systems for their bots. Not so, according to Alex. He spent a thoroughly enjoyable hour describing first how rulesets don't scale for larger, more complex problems ... and then walking us through his solution... The fun part here was that each time he described an extension to the basic idea (Augmented Transition Networks, first widely used in NLP), he'd bring up a game where this particular extension was first deployed. Lots of interesting snippets about Republic, for instance. The neat thing was that he described a complete development environment, based around the ATN, which included an interactive debugger for game designers, check points and save positions... reused, extended and refined over a score of titles.

What is it with people wanting to get into the gaming industry ? Are things like the EA scandal not enough to dissuade them ? Apparently not. Game programming is more popular than ever, and even OReilly has gotten in on the act. Unfortunately, truly adaptive AI in the Skynet sense of the word, seems to be a long way off. Like much of the AI field in general, a series of shortcuts (Eliza ?) perform close enough to the ideal to satisfy most needs.