Hal Et Al.: How Smart Is Artificial Intelligence?

HOMICIDAL MANIACS don't usually have posthumous birthday parties. But next month, tech types and movie buffs will head to the University of Illinois at Urbana-Champaign to celebrate the inception of HAL, the mutinous computer from Stanley Kubrick's 30-year-old masterpiece ""2001: A Space Odyssey.'' (In Arthur C. Clarke's novel based on the film, HAL says he was born in Urbana in 1997.) The celebration, dubbed Cyberfest, will be a Nerdstock of sorts, featuring sci-fi film screenings hosted by movie wonk Roger Ebert, lectures from the likes of MIT's Marvin Minsky and a future-tech show and tell. Scientists have long admired ""2001'' for getting the details right--from the soundlessness of space to the hamsterlike practicalities of astronaut exercise. But in at least one respect, the movie jumped the gun: today HAL is nowhere to be found.

The reason, say the experts, is that for machines to truly think, they must simulate the operations of our own brains, which we don't fully understand. ""Artificial intelligence has turned out to be a profoundly hard problem, on the magnitude of understanding the meaning of life,'' says David Stork, editor of a new book, ""HAL's Legacy'' (MIT Press). As a result, current researchers have turned away from the goal of building consciousness into computers, and are instead working toward intelligent machines that will serve humanity by offloading our difficult calculations, sorting through our information gluts and generally going where our gray cells fear to tread.

While HAL-lovers might be disappointed, AI's more modest goals still hold interesting promise for the next millennium. Witness the work of the Beckman Institute, an interdisciplinary research center on the same campus as HAL's party. The engineers, scientists and mathematicians working there aren't even sure that we need a HAL. ""We already have a pretty good way of replicating humans,'' jokes Caroline Hayes, a professor of AI. Instead, the team is focusing on building programs like ""intelligence assistants,'' software that perches over the shoulder of professionals such as pilots to pick up errors and offer suggestions. The programs rely on ""heuristics,'' or case-based reasoning, using rules or past experiences stored in a database to make intelligent assumptions.

In the Beckman Institute's robot lab, a small, ceiling-mounted camera attached to a robotic arm overlooks an old air-hockey table. The robot, nicknamed ""Gretzky,'' actually plays, mounting a mean defense and knocking shots back. Soon, say Gretzky's creators, it will hit harder, learn from mistakes and even find weaknesses in its opponent. The point is not to win barroom bets but to study how robots perform in unpredictable environments, like landing an airplane in the fog or deactivating a bomb. ""We're looking at the robot's learning strategies to see what other tasks we can apply it to,'' says Prof. Gerald DeJong.

Not everyone, though, has given up on building a HAL. In Austin, Texas, former Stanford professor Doug Lenat has spent 13 years entering millions of everyday truths (for example: a father is older than his son) into a computer called CYC. The hope is to create an artificial intellect the same way we raise our kids: by exposing it to a range of facts, and then letting it gather information and make inferences by itself. And at MIT, Prof. Rodney Brooks has built a small, humanlike robot called COG and let it roam, with the idea that by learning from sensory experiences--like bumping into a wall--it can develop intelligence.

Even skeptical AI researchers concede that a computer like HAL may be possible, perhaps in a century. But do we really want machines that might be smarter than we are? Not even author Clarke seems sure. His new book, ""3001: The Final Odyssey,'' paints a fascinating picture of our future: cities atop needlelike towers that extend into space, the colonization of Venus, the pacification of humanity and the abolition of religion. (Warner owns the movie rights.) Sentient machines, however, are noticeably absent. From his home in Sri Lanka, Clarke e-mails, ""As a member of the species that will be superseded, I have mixed feelings on the matter.'' If the computers of the next millennium turn out to be anything like HAL, we all should think twice.