Like a good gambler, Daphne Koller, a researcher at Stanford whose work has led to advances in artificial intelligence, sees the world as a web of probabilities.
There is, however, nothing uncertain about her impact.
A mathematical theoretician, she has made contributions in areas like robotics and biology. Her biggest accomplishment — and at age 39, she is expected to make more — is creating a set of computational tools for artificial intelligence that can be used by scientists and engineers to do things like predict traffic jams, improve machine vision and understand the way cancer spreads.
Koller’s work, building on an 18th-century theorem about probability, has already had an important commercial impact, and her colleagues say that will grow in the coming decade. Her techniques have been used to improve computer vision systems and in understanding natural language, and in the future they are expected to lead to an improved generation of Web search.
“She’s on the bleeding edge of the leading edge,” said Gary Bradski, a machine vision researcher at Willow Garage, a robotics startup firm in Menlo Park, California.
Koller was honored last week with a new computer sciences award sponsored by the Association for Computing Machinery and the Infosys Foundation, the philanthropic arm of the Indian computer-services firm Infosys.
The award to Koller, with a prize of US$150,000, is viewed by scientists and industry executives as validating her research, which has helped transform artificial intelligence from science fiction and speculation into an engineering discipline that is creating an array of intelligent machines and systems. It is not the first such recognition: In 2004, Koller received a prestigious US$500,000 MacArthur Fellowship.
Koller is part of a revival of interest in artificial intelligence. After three decades of disappointments, artificial intelligence researchers are making progress. Recent developments made possible spam filters, Microsoft’s new ClearFlow traffic maps and the driverless robotic cars that Stanford teams have built for competitions sponsored by the Defense Advanced Research Projects Agency.
Since arriving at Stanford as a professor in 1995, Koller has led a group of researchers who have reinvented the discipline of artificial intelligence. Pioneered during the 1960s, the field was originally dominated by efforts to build reasoning systems from logic and rules. Judea Pearl, a computer scientist at the University of California, Los Angeles, had a decade earlier advanced statistical techniques that relied on repeated measurements of real-world phenomena.
Called the Bayesian approach, it centers on a formula for updating the probabilities of events based on repeated observations.
The Bayes rule, named for the 18th-century mathematician Thomas Bayes, describes how to transform a current assumption about an event into a revised, more accurate assumption after observing further evidence.
Koller has led research that has greatly increased the scope of existing Bayesian-related software.
“When I started in the mid to late 1980s, there was a sense that numbers didn’t belong in AI [artificial intelligence],” she said in a recent interview. “People didn’t think in numbers, so why should computers use numbers?”
Koller is beginning to apply her algorithms more generally to help scientists discern patterns in vast collections of data.