| Follow Us: Facebook Twitter



Features

Natural Intelligence








One of the most dynamic trends in science has been more than 50 years in the making, largely at the University of Michigan and research labs in the Ann Arbor area. It's as simple as bird and ant behavior and as complex as the human genome.

"Bio-inspired" computer algorithms  - so-called "genetic algorithms" and "swarm intelligence," among others  - were named one of Scientific American magazine's 10 "World Changing Ideas 2010" in its December edition.

John Holland, a professor of psychology, electrical engineering, and computer science at the U-M, was wandering in the mathematics library one day in 1955, when he spotted a book on genetics. As world changing ideas go, that was his "ah-ha" moment.

"I didn't know you could do mathematics in biology. I thought, there are things I already knew about computers…. Boy if I could put these two things together. It all sort of folded out of that." He calls it an "adaptive plan," which one of his students called "genetic algorithm."

Origin and evolution

At the time, theories about artificial intelligence -- the use of computer programs to simulate human reasoning -- were in their earliest development. But Holland was skeptical. "People thought that you could hack it... that you could write a clever program and do anything intelligent." For example, you could model an analytical diagnostic technique, but it would be static. "It didn't adapt, it didn't learn, it just was just a smart program," notes Holland. He felt a program that would adapt and learn, as in human evolution, would be more effective.

Holland went on to literally write the book on genetic algorithms in 1975. Adaptation in Natural and Artificial Systems was just the first of three books he's authored on the subject. In a July 1992 Scientific American article Holland wrote, "Living organisms are consummate problem solvers. They exhibit a versatility that puts the best computer programs to shame. ...Pragmatic researchers see evolution's remarkable power as something to be emulated rather than envied." In the mid-1960s, he formalized his concept of a genetic algorithm that would act like the human genetic process, and created a genetic code that could be used by computers.

Evoking "evolution" in that article proved threatening to Holland and his colleague Rick Riolo,
computer lab director at the U-M's Center for the Study of Complex Systems (CSCS). Riolo, whose article, "Survival of the Fittest Bits," was published in the same issue of Scientific American, wrote, "In Darwinian terms, life is a struggle in which only the fittest survive to reproduce. What has proved successful to life is also useful to problem solving on a computer."

At the time, both researchers received death threats as a result of their writing.

Acceptance, commercialization

Holland and Riolo were early members of CSCS, which annually convenes an interdisciplinary group of researchers' work in the field. In October, the Center will host the Fifth International Conference on Self-Adaptive and Self-Organizing Systems in Ann Arbor.

Today, genetic algorithms may not be new to the international research community working in the realm of "complexity," but the commercial sector is just beginning to put them to work.

Ted Belding, president of Belding Consulting, set up a consulting business that focuses on applied genetic algorithms for business solutions. "One thing genetic algorithms are really good at is instead of having to search the whole [universe] of possible combinations, genetic algorithms allow you to test a small subset of that. That's a big advantage, especially testing something that's risky. You don't want to put a bad website up."  For example, they can be used to test aspects of a website layout. "You can apply genetic algorithms to just about anything," he says. Like any form of research, this can be a costly process. But Belding adds, "There's a cost in putting up a lousy website."

Among the Ann Arbor-area companies using bio-inspired algorithms are Everist Genomics (formerly Genetics Squared), a medical prognostics company; SimuQuest, an engineering company working in the automotive, aerospace, and alternative energy industries, and Vector Research Center, which provides research services for the U.S. Dept of Defense, the intelligence community, and manufacturing.

Van Parunak, chief scientist at Vector Research and a U-M grad, has been working with swarm intelligence, a research method drawn from observations of animal behavior. Vector Research conducts studies for manufacturing, defense, and intelligence community clients. In the mid 1980s, while working at the Industrial Technology Institute in Ann Arbor, he was struck by a flock of geese forming a "beautiful V-shape" in flight.

"I had recently taken on the responsibility of getting manufacturing machines to coordinate with one another. I looked at those geese and wondered, 'How are they doing that?'" He collected behavioral information about a variety of animals. "What we found, over and over, is animals, with no central plan, were able to interact with each other through a shared environment with very simple individual rules, but the behavior that comes out is very complex."

"Swarming," he says, is "a lot of little things simply interacting with each other and something interesting comes out of it." This differs from artificial intelligence, in which a logic model is created to solve a problem. With swarming, digital "ants" or "agents" are set in motion through algorithms and naturally create solutions to a problem.

Although these bio-inspired researchers are working in a disruptive realm that challenges what has been known as "artificial" intelligence, Parunak says that the two can be complementary. Clients who need to make risky decisions with the intelligence derived from swarming don't get the kind of rationalization that comes through logic models. "If I try to solve the problem with swarming, and asked how did you get this path? I'd say, I don't know. I turned the ants loose and that's the path that came out," notes Parunak.

Vector Research collaborates with Soar Technology, an Ann Arbor firm that models human reasoning. Vector sets swarming in motion to solve the problem and Soar rationalizes it through its analytics. "If it passes their test, then I have the explanation I can give to the CEO why it's good," says Parunak.

The seemingly simple behaviors of birds and ants are anything but simple. They are complex actions conducted by creatures with little intellectual capacity. So what is it about human beings with advanced intellect and computer capability who stop in their evolution to not only examine animals, but learn from them?

"In a way, it's going back to nature," says Belding. On the other hand, it's the result of advances in basic science, mathematical theory, and computer capability. "We are really bad about thinking about  complicated things in our head. The good thing about computers is that if you have a theory about how something works, you can simulate it on a computer and see if it actually works. ...It's a way of double-checking our theories of the way things work. In a fair number of cases it shows that things are far more complicated than we thought."

Holland is working on the next generation of bio-inspired computing to create simulated consciousness.  Human beings, he says, are good at doing things instinctively, "but we don't know how it really happens.  ...Consciousness is high level stuff that we do as humans. We don't have any machine models of that. Little by little, people are trying to build computer models that will go from the neural network that approaches consciousness. We're in the early stages. This is a difficult problem."

For those who worry that artificial intelligence -- drawn from animal behavior or simulated human reasoning -- will eventually replace human thinking, Holland offers reassurance in the form of a different caution: "I worry a lot more about deciding our babies genetically than I worry about a model like that. That is really worrisome. We don't have any controls, yet. I think humans are so diverse in intelligence. Even if I build a very good computer model, it's just going to be like another smart human. ...There's plenty of room for intelligence."

For information on the International Conference on Self-Adaptive and Self-Organizing Systems visit www.saso-conference.org.


Dennis Archambault a freelance journalist and regular contributor to Metromode and freelance writer. His previous article was Manufacturing The Future

All photos by Doug Coombe

Pointing out the queen bee with a green dot painted on her thorax

82 years young John Holland at West Hall on the U of M campus

John Holland at West Hall

Ted Belding getting some work done at Sweetwaters in downtown Ann Arbor

Van Parunak at Vector Research in Ann Arbor

Sven Brueckner, John Sauter and Van Parunak at Vector Research

Ted Belding at Sweetwaters
Signup for Email Alerts
Signup for Email Alerts