AI's Next Brain Wave

Palo Alto Research Center: AI Gets Attention
Computer scientists at the Palo Alto Research Center also are trying to bring user interfaces to life by replacing raw information with material that selects itself based on what the computer thinks the user wants to know. PARC's user-interface group published a paper in January describing new software called ScentHighlights that helps users skim information by extracting key sentences from an electronic book, relevant to keywords a user types in or clicks on in the text. The system, based on a PARC theory called information scent, is part of an emerging class of user interfaces that react to what gets a user's attention, says Stuart Card, manager for user-interface research and a senior research fellow at PARC.

Card, who's been with PARC since 1974 and performed an early user-interface study that showed the utility of the computer mouse, says PARC's user-interface research lies along a technology continuum, with classical AI techniques on one end and Bayesian statistical analysis of the occurrence of words in a text at the other. In classic artificial intelligence, software tried to form logical groups, called ontologies, to infer relations among concepts. Robins and sparrows both fit into the larger category of birds, for example. Rather than use this approach, PARC's E-book interface uses new AI approaches, but "we're working with our linguistics and AI groups to push us further along," Card says.

ScentHighlights helps E-books make inferences about their own content. In experiments with an electronic version of the book Biohazard, about the author Ken Alibek's experiences working on biological weapons in the former Soviet Union, users skimmed passages of the text to effectively complete tasks such as finding or comparing facts. Exact keyword matches to terms users typed or highlighted appeared in pastel colors, and passages that were related but not exact keyword matches appeared in gray. Says PARC research staff member Ed Chi, "When I type in a few words, [the software] seems to know what sentence I ought to read. The AI is part of the pattern recognition."

Artificial-intelligence research dates back to 1956, when pioneers at Stanford and MIT held a conference at Dartmouth College to explore the technology. If critics say it has fallen short of expectations, perhaps that's because its practitioners have aimed so high. And if today's AI researchers still draw fire 50 years hence, that could be one measure of their success.

Illustration by Brian Stauffer