R&D Magazine, October 1998.
"Computer scientists search for ties to biological intelligence."
By Tim Studt

The low cost and high performance of today's microprocessors are impressive, yet simple sea slugs still outperform them in the efficiency of computing operations. The fact of the matter is that computers and biological beings process information in dramatically different ways. And while much effort is put into developing and understanding digital-based computer operating systems, very little is known about how even simple biological organisms process information, let alone how complex vertebrate organisms do it.

Leading computer scientists and biologists met this past summer to explore the intelligent behavior of humans and animals and discuss how that knowledge might be applied to computer science. The scientists were attendees at Microsoft Research's (Redmond, Wash.) Summer Institute on Intelligent Systems: Biological and Computational Perspectives. It was co-sponsored by the Univ. of Washington (UW), Seattle.

The meeting was put together by Eric Horvitz, a senior researcher and group manager of the Decision Theory and Adaptive Systems Group at MSR; Dennis Willows, director of Friday Harbor Labs and a professor of zoology at UW; and Chris Diorio, assistant professor of computer science and engineering at UW.

"We're interested in increasing our computer-based theories and applications by examining clues from neurobiology about the functioning of systems that have been designed and 'tested' by natural selection," says Horvitz.

More than 40 of the top researchers in the world in evolutionary science, biology, neurology, computer science artificial intelligence, and mathematics were invited to the session to cross fertilize each other's concepts of intelligence.

"It hadn't occurred to me that the concept of intelligence in a brain would be directly comparable to intelligence in a machine," says Willows. "But the more I thought about it, the more I saw that it was a no-brainer."

"Computer engineers tend to ignore all the lessons that biology has spent billions of years perfecting," says Diorio. "Brain neurons don't work like computer chips, they aren't good at math, but they do things that are hard for a computer to do, like understand speech," he says. "If we can figure out how the neurons in our brains represent information and learn, then perhaps we can copy biology and build new types of computer chips with broader capabilities."

"On the flip side," says Horvitz, "computer scientists have developed theoretical models of learning and decision-making that may be useful for guiding neurobiology research." One of the items that came out of the meeting was the realization that biological beings function well by limiting the information flow they receive. "Biological nervous systems operate best and most efficiently by ignoring nonrelevant information," says Horvitz. Conventional computer systems, on the other hand, operate on systems that process all available information--at lighting-fast speeds, but with little regard for the specific relevancy of the individual information.

Nervous systems also operate on the basis of threshold values of excitatory stimuli to each other. "Understanding these processes could affect development of modern control theory--something like making use of the "less is more" theory," says Horvitz.

Digital computers also operate through software by mapping logic statements. Biological systems don't do that, they operate through a system of adaptation.

As the session progressed, the biologists also picked up information useful in their experimental studies. "We have lots of mathematical and analytical tools to study the behavior and communication of biological systems," says Diorio. Some scientists found new ways to directly embed microprocessors in their biological specimens. These will gather and analyze information on how organisms process information by mapping their neural circuitry.

Sessions at the Summer Institute covered topics on languages; actions under uncertainty; control, stability, and robustness; and selection, evolution, and environments. Attendees were also divided into seven working groups on intelligent systems that focused on its computational basis, reliability and robustness, evolution, sensory integration, architectures, learning, and synergies between the neurobiological and computer sciences.

While the participants of this year's Summer Institute were enthusiastic about the results of their discussions, they uniformly believed that it will be some time before any solid results will be made to understand the workings of biological neural systems and longer to apply them to computer-based systems. Most estimated that a practical understanding would not occur for another 10 to 20 years.

This year's MSR/UW Summer Institute follows a similar session in 1997 on datamining. The success of both sessions assures followup sessions on additional topical computer-based subjects.