"If [game AI] can remember and tell you, ‘I shot you because I saw you bug the ambassador,’ then that’s starting to be a conversation between the human and AI player that feels fair and natural."
- SpyParty developer Chris Hecker suggests AI designers give their creations a way to tell players why they act.
Designing artificial intelligences for games seems especially challenging because you can't just create an AI that plays a game well -- you also have to try and make one that can play like a human, with believably human errors.
This is a topic Gamasutra explored at length last fall with a number of AI experts, and today a New Scientist column about the state of game AI design continues that line of inquiry (now that a Google AI has beaten a top Go player at his own game) by seeking input from game developers Chris Hecker (SpyParty) and James Ryan (Talk of the Town.)
It's an interesting read, especially for anyone with an interest in how contemporary game makers are devising solutions to the tricky problem of making a game AI that can believably do things like make mistakes or misremember facts.
"This is actually one of the biggest problems with AI for games...it’s hugely important to make them fallible," Hecker told the New Scientist. "It’s relatively trivial to make an AI that can kill a player every time, but making it feel like a worthy competitor and, more importantly, fun and interesting to play, is hard."
Hecker has spoken about this subject at length in the past, with Gamasutra and other outlets; but it's interesting to see his input here alongside that of Ryan, who's reportedly designing Talk of the Town (with colleagues at UC Santa Cruz's Center for Games and Playable Media) to be a game where AI characters can share knowledge, misremember that knowledge, and even lie about it.
"Lying is the most challenging aspect to model because lying is a very complex and nuanced human phenomenon,” Ryan said. “People lie about all sorts of things for all sorts of reasons."
You can read further comments from both developers, as well as a deeper inquiry into the state of game AI design, over at New Scientist.