By Rebecca Vesely, January 1999 Issue, Business2.0

Watch a jellyfish gingerly navigate its way past wooden pier pilings, a gelatinous mass of tentacles hanging from a pulsating bell. Could this 700 million-year-old creature teach us something about making software and networks smarter? Researchers at Microsoft think so.

"Theories about how to take optimal action with limited resources have already been figured out by biology over billions of years," says Eric Horvitz, senior researcher and manager of Microsoft Research's Decision Theory and Adaptive Systems group. "Now we are trying to figure out how we can make good use of this in computers."

Horvitz believes that studying the nervous systems of jellyfish and other simple organisms could reveal insights that could be integrated into artificially intelligent systems. His goal is to unravel the puzzle of how elemental creatures make effective, flexible decisions in situations of uncertainty.

The jellyfish, for instance, may reveal insights as to how to build more complex networks that would mimic the human brain. "Physiologically [the jellyfish] is almost identical to you and me," says Peter Anderson, professor of physiology, neuroscience, and zoology, and director of the University of Florida's Whitney Laboratory at St. Augustine, who is working with Microsoft Research on neuroscience applications. "There are far less neurons interacting, of course, but basically they are the same."

Insights revealed in studying this simplified version of our own nervous system could go a long way in helping neuroscientists create a reliable artificial intelligence (AI) engine that could apply principles of biology to make computers work smarter.

As one of the leading researchers in applying natural characteristics to technology, Horvitz already has a few AI laurels on his mantle. He led the efforts to create the help technology behind several Microsoft software programs, including Office 97. "I'd like to crack the code of how neurons talk to each other and, more specifically, how information is encoded," he says.

Other simple creatures may also provide valuable AI insights. Dennis Willows, a zoology professor at the University of Washington and director of the Friday Harbor Laboratories in the San Juan Islands north of Seattle, studies the nervous system of sea slugs in hopes of gaining insights into navigation.

"In simple organisms, the nervous systems are incredibly accessible," Willows says. "We wouldn't learn much from the human brain because there are too many cells. They are small and hard to reach, and human subjects are hard to come by."

Sea slugs, on the other hand, have brain cells so large that they can be seen with the naked eye. At between half a millimeter and one millimeter, they are easily manipulated and dissectable. Willows and others have identified individual brain cells that control locomotion behavior—how a sea slug navigates through its world—and discovered that the species uses geomagnetic homing (reading the Earth's magnetic fields) to orient itself. Even in a lab, a sea slug will crawl toward shore.

Willows hopes that Microsoft Research will help him make new discoveries about biological networks in much the same way that the Microsoft researchers hope Willows' work will aid them. "Horvitz's work and that of other AI researchers reveals principles of neuro-network design, so that we can look at computer designs and see if they apply to nervous systems in animals," says Willows.

Getting back to nature

Artificial intelligence research typically focuses on creating a computer that can mimic the instinctive_decision-making and information prioritization that occurs in the wild. Cambridge University zoology professor Malcolm Burrows has shown that grasshoppers free up brain space (or "bandwidth") to react to surprises. A grasshopper is essentially able to put its legs on autopilot to increase sensory functions that look out for predators. "Any sensation that is not valuable information is somehow suppressed so more important information can be readily detectable," Horvitz explains.

Horvitz and other Microsoft researchers are applying basic principles of decision science, such as prioritization, to software, though Horvitz says thorough application of neuroscience could be decades away. In Microsoft Intelligent User Interface projects, computers are taught to sense and react to information by learning what is important to people—their goals and priorities—beyond simply storing data.

An example of this technology is Look Out, a Microsoft program that schedules appointments by reading email and observing how the user decides what to schedule. Paying attention to words in an email like "lunch tomorrow" or "confirmation" or to dates and times, the program could ask if you want it to schedule the appointment, based on probability computations. Even more, Horvitz says, the program would adapt to your timing, knowing when to ask if you need help, without being intrusive.

The next version of Microsoft Outlook Express, due this year, will include an anti-spam program developed by Horvitz's group that analyzes banners, phrases, and percentage of capitalized letters to determine whether email is important or trash.

Computers could also use artificial intelligence to communicate with one another to help us with basic functions like transferring data. As part of Microsoft's Millennium Project, a distributed, network-based operating system, Horvitz is working on principles of organization and communication in biological systems. They may make the network more user-friendly and allow users to move data around with as little effort as possible, or, like the grasshopper, instinctively.

Transition plan

Microsoft Research, founded in 1991, is growing at lightning speed. Some 300 researchers work on Microsoft's Redmond, Wash., campus and the company hires on average one researcher a week. By 2000, Microsoft plans to have doubled the Research staff to 600. Meanwhile, the software giant has kept pace with spending, doling out $2.5 billion to all research and development in 1998, up from $1.43 billion in 1996.

While Microsoft's research department is by no means the largest institute of its kind (Bell Labs, for example, has 24,000 researchers, and IBM has some 2,700 R&D employees), it has a higher proportion of Ph.D.s than some competing research facilities, with many hailing from the top learning institutions in the world. A 1997 Business Week survey on the top U.S. research facilities ranked the five-year-old Microsoft at number 11, behind Stanford University (number one), MIT labs, and AT&T Labs, but ahead of the University of Michigan and the Santa Fe Institute.

"Creating a research arm is part of the classic transition from a hugely aggressive, massively growing company to a much more mature company," says Ted Schadler of Forrester Research, who tracks Microsoft Research and other research facilities. "Microsoft doesn't worry about revenues for the next quarter, but it does worry about revenues in the next decade."

Unlike some other blue sky research centers, forays into the neurosystems of jellyfish must translate into better software applications on store shelves—or Microsoft will quickly lose interest. "Microsoft's product strategy is clear: selling software," says Schadler. "They have to do research in the context of Microsoft products."

Even when that context means diving into the ocean. Ultimately, there may be no greater teacher than Mother Nature.