Trust Me–I’m a Robot

The Issue: Hollywood androids are dysfunctional human wannabes. Real AI researchers are devising something else

Here’s something that engineers designing the next generation of
unmanned combat air vehicles may not have considered: A well-placed lightning strike could rewire the plane’s artificially intelligent (AI) brain, transforming the craft into an enemy of the state bent on destroying a major city. Absurd? Maybe. Good plot for a thriller? Evidently: It’s the source of the action in the new movie Stealth, in which an AI-controlled fighter jet turns evil.

Artificial intelligence has returned to the big screen in a big way in recent years, but Hollywood hasn’t offered much new thinking on the subject. Artificial Intelligence: AI (2001) was based on a 30-year-old short story by Brian Aldiss. Bicentennial Man (1999) and I, Robot (2004) grew out of old Isaac Asimov tales. And the AI in Stealth is essentially HAL 9000 from 2001: A Space Odyssey–but with less personality and more missiles at its disposal. Martha Pollack, an AI researcher at the University of Michigan, says this type of story owes more to Godzilla than to real-world research: “An intelligent creature that we can’t understand goes crazy and does evil things.”

I certainly understand that the man-versus-machine bit makes for good thrillers. And that moviegoers embrace stories about Anthropomorphic Robots Who Can Do Everything . . . But Love. The little tear dribbling out of the robotic eye–I get it. My gripe isn’t with the lack of realism so much as the way these films inflate expectations, making actual advances seem mundane. Case in point: Last year’s DARPA Grand Challenge robot-vehicle race. The event took place three years after HAL 9000 was supposed to have been killing astronauts on the way to Jupiter (not to mention 36 years after the movie came out), so I expected more than just a bunch of vehicles eking out a few off-road miles before stalling out or self-destructing. Not only did I want to see the winner cross the finish line, I envisioned it standing up on its rear wheels, taking a bow, and saying something to the effect of: “I’d like to thank all the wonderful humans at Carnegie Mellon. Without them, I wouldn’t be here . . .”

In the past, wild expectations have, when dashed, slowed real-world research. The boom in AI research in the
1980s was fueled in part by science-fiction visions in books and movies, in part by the grand prognostications of early pioneers, plus a little cross-Pacific competition with the Japanese. When it failed to produce anything approximating Robby the Robot, funding dried up, and a period now known in the field as the AI winter set in. It’s unclear whether the chill actually delayed progress–or whether it allowed researchers to step out of the limelight and refocus their efforts to make real advances. Either way, with AI now running our junk-mail filters and gaming consoles, the technology is back.

But not in the form of the do-everything brains beloved by Hollywood. Today’s AI is task-oriented–what Pollack calls “niche intelligences.” One of her projects is designed to track whether an elderly person is taking his medicine, and to prompt him to pop a pill if he forgets. This doesn’t measure up to the
vegetable-dicing, dog-walking NS-5s in I, Robot, but it’s still pretty cool.

Not everyone is convinced that such narrow utility is the ultimate horizon for AI, of course. Pollack says many researchers still dream of the humanlike robot. And if we ever do get to that point, the movies have provided us with a few helpful warnings. For example, fathers will need to keep their eyes peeled for robotic suitors. In Bicentennial Man, unsuspecting parents purchase a household robot for domestic chores, only to have their daughter fall for it. This is an unsettling prospect, but Gigolo Joe, the robotic male prostitute played by Jude Law in AI, is even more frightening. Beyond his superior physical endowments, he’s programmed to always say the right things. That’s hard to compete with. Another rule is to always check your AI for homicidal tendencies. The voice is a particularly good giveaway. If it sounds anything like HAL, it’s either low on batteries, high or evil. And if Stealth is any indication, it’s probably not a good idea to give our AIs too many weapons, either.

My advice: Quit wondering when you’ll have an affordable, brilliant robot slave to run out and get your coffee, and start thanking your spam filter and its more homely intelligence. It won’t answer back today, but my research at the movies convinces me that someday it will.