B.S., psychology, University of Otago, Dunedin, New Zealand, 1984; Ph.D., computer science, Carnegie Mellon University, 1995.
In the 1950s, computer science pioneers such as Herb Simon and Allen Newell were trying to build a thinking machine that would reason and solve problems using the same logical process as a human. By the 1990s, this traditional artificial intelligence research had been overshadowed by statistical machine learning. Taking advantage of extreme scale and powerful processors, statistical techniques have solved many problems in speech recognition, language translation and understanding images.
Yet although a computer such as IBM’s Watson can now answer fairly complicated questions, “it doesn’t really understand the stuff that it’s processing,” says Michael Witbrock, vice president of research at Texas-based Cycorp. “We’ve got machines that can do very shallow reasoning at scale, and that’s proved to be useful. Now it’s time to pay attention to cognition—the sort of things that provide intelligence worthy of the name.”
Cycorp is developing Cyc, a project first launched by Doug Lenat in 1984 to create an AI that learns and reasons about problems using first-order logic—the holy grail that has tantalized researchers since the dawn of the computer age. But Witbrock says the rule-based AI systems of the 1950s and ’60s were hampered by the state of the art in computing. The Internet has now enabled collection of seemingly endless amounts of data. Multi-terabyte storage systems are inexpensive, as are reliable commodity processors.
“It’s an exciting time to be in AI, because computers are fast enough now that we can do fairly deep automated reasoning fairly quickly,” Witbrock says. “In retrospect, it was kind of ludicrous to think that you could build a reasoning machine out of the technology that was available. And even though we’ve got vastly more powerful computers today—more storage, huge data sources—automated reasoning is still a very difficult problem.” And modern AI research is very much informed by statistical machine learning, he says. “It’s not an ‘either-or’ situation,” Witbrock says. “There is an enormous potential for synergy.”
There’s also commercial potential in AI development, he says. “Why is speech recognition taking off? Because in the 1990s, there wasn’t the ecosystem in which it could flourish. Now, everyone has speech recognition on their smartphones. I believe we’re just beginning that process with general automated reasoning.”
Cyc is currently reading vast databases of literature to understand and develop theories about the biology of cancer. “It’s an area that needs AI, because the amount of information is just enormous,” Witbrock says. “The number of possible mutations that a particular person with cancer may have is just of incredible complexity, beyond the ability of any group of human beings to grasp—we simply can’t communicate with each other quickly enough. But the biology is also heterogeneous enough that we also can’t simply write a program to process all of the data.”
British physicist Stephen Hawking recently expressed his concern that AIs might become sentient, malevolent and harmful. “It’s not an illegitimate concern,” Witbrock says. “There are several technologies that it could be possible to lose control of in a bad way. One is naontechnology, another is synthetic biology, and a third is artificial intelligence. People for a long time have had sort of glib answers with respect to these threats, and why they may not be serious threats.”
But over the last few years, with the rapid development of computing power, researchers are giving serious thought to the problem of keeping future AIs in check, Witbrock says. “Looking at the behavior of large organizations such as corporations and universities—constructed intelligences built from people—we can learn a lot about how to mitigate the failure of artificially intelligent systems.”
Outside of the lab, Witbrock is active as a board member of Startout.org, a national organization that encourages entrepreneurship in the LGBTQ community. “All communities and affinity groups, as they develop, need to shore up their position in society,” he says, “legally, yes, but also including financially.”
Jason Togyer | 412-268-8721 | email@example.com