If you’re not worried about the oncoming struggle between humanity and artificial intelligence, you’re not paying close enough attention. Scores of experts both within the AI field and in other disciplines have issued dire warnings about the dangers of creating super-intelligent machines which can act and make decisions autonomously and whose existence does not depend on feeble, senescent meat sacks. Still, AI researchers continue to find ways to make these artificial abominations more intelligent and human-like. At this point, you have to wonder: how can this not end badly? I mean, do AI researchers not watch or read any science fiction? There are only two ways this could end: termination or enslavement. Then again, maybe the machines already have enslaved us…
Much of the research into how artificial intelligence systems learn centers on teaching machines how to play games – or letting them learn on their own. Some of the most high-powered AI constructs currently dominate humans at some of the most sophisticated human games like poker, chess, and Go, not to mention absolutely annihilating us at deathmatch video games. Forget Korean prodigies in smoke-filled internet cafés – the gaming world is about to be overrun with AI systems passing as human players.
It turns out, though, that the somewhat rigid constraints of video games or tabletop games might mean a very different type of game is better suited to training AI: open-ended role-playing games like Dungeons & Dragons. That’s according to Beth Singler, a research associate at the Faraday Institute for Science and Religion at the University of Cambridge. Singler recently penned (keyboarded?) an essay asking if role-playing tasks might be just the thing to lead to more human-like AI:
Do we need a new test for intelligence, where the goal is not simply about success, but storytelling? What would it mean for an AI to ‘pass’ as human in a game of D&D? Instead of the Turing test, perhaps we need an elf ranger test?
Singler argues that since Dungeons & Dragons and similar role-playing games task the player with switching between roles, cooperating with other unpredictable players, and improvising according to changing game conditions, they might be able to train AI which can better function alongside humans who constantly and unconsciously do the same. “Instead of beating adversaries in games,” Singler writes, “we might learn more about intelligence if we tried to teach artificial agents to play together as we do: as paladins and elf rangers.”
Is more human-like AI what the world really needs though? With all of the threats facing humanity, why add a new potential adversary to the mix? Unless it’s a healer, because our party really needs one of those for our upcoming raid on that necromancer’s dungeon.