GDC 11: Turing Tantrums (AI Developers Sound Off)
AI developers, stereotypes would suggest, must be of a nebbishy sort, almost mole-like. Nothing, it turns out, could be further from the truth. The “Turing Tantrums: AI Developers Rant Back” panel, on the inaugural day of GDC 2011, featured some of the day’s best humor, and its most engaging and confident speakers.
Richard Evans began proceedings in an endearing lilt. As an AI programmer for classics like The Sims 3 and Black & White, Evans was furnished with both wit and experience, and his PowerPoint presentation set the tone early on with a number of hilarious slides. The topic of his speech was “Rolling Your Own” AI code, and there were mighty chuckles when Evans switched to a slide of a sweaty, dirty hippy with a rolling paper pressed to his tongue. The appearance of Gregory Peck as Ahab later in the presentation was also hilarious. The British developer’s point was basically this: writing your own AI code is not as hard as people think it is, and it will lead to more unique and more interesting games.
His colleague Kevin Dill, an employee at Lockheed Martin, argued the converse. People are too worried about being “unique,” he claimed. Design people re-use Direct X all the time, so why shouldn’t AI people borrow bits of pre-existing code? There are certain behaviors that are common to all games within the same genre. These similarities, Dill pointed out, can help developers. Though he didn’t use humor as effectively as Evans, the man from the Military Industrial Complex was just as convincing, leaving the AI devs in attendance with a difficult choice: whom to believe?
Blizzard’s Brian Schwab was next, and he cut quite the figure. Built like an NFL linebacker, with a long ponytail gathered by four or five rubber bands, Schwab looked like he could have done motion capture work for any number of his company’s hulking NPC’s. Despite his musclebound appearance, however, his brain was clearly his biggest asset. The theme of his talk was the calibration of the AI to maximize player enjoyment — too hard, and the gamer gets frustrated. Too easy, and the gamer feels let down — this is, as Schwab points out, just as annoying as too hard (I’m looking at you, Assassin’s Creed). The burly blonde summed up his speech with a perfect analogy: AI should let the gamer win like a gracious father competing against his young son.
Panel leader Dave Mark finished off the event, and the personable Nebraskan had a great metaphor of his own. AI, he said, is the simulation of life. It follows, therefore, that good AI programmers should be careful observers of human behavior. He quickly undermined his point, however, with a number of amusing but counterproductive examples of real-life behavior that resembled bad game AI — the manager of his local supermarket, to name one, gives him a perfunctory hello every time they pass in aisles, even if that happens four times in one shopping trip. There were many of these, but they seemed to belie Mark’s point. If bad AI behavior happens in real life, doesn’t that mean that sometimes, bad AI behavior is actually good, because it resembles real-life behavior?
Regardless of your answer, the Turing Tantrums panel proved one thing: don’t underestimate the ability of AI programmers to tell jokes. It also undoubtedly provided a lot of insight into that particular aspect of game design.