What’s the difference between a dolphin and a smart computer program? Not as much as you might think, according to Kenneth De Jong ’66.
“We can get dolphins at Sea World to do rather complex things with a reward/punishment feedback scheme,” De Jong said. “That’s basically the same kind of feedback we use with computers to get them to do complex things.”
Like play a really good game of chess. Set multiple chess-playing computer programs to play each other. The losing programs “die off” while the winning programs are rewarded with the chance to play each other—and to join with other winners to produce “offspring” in a new and gradually improving “community” of chess-playing programs.
This is survival of the fittest as Darwin never dreamed it.
De Jong is a pioneer in a field he helped to name: evolutionary computation. As the director of the Evolutionary Computation Laboratory in the Volgenau School of Information Technology and Engineering at George Mason University (Va.), he’s taken inspiration from how species adapt in the natural world and applied it to computers, to the complex computer programs that drive robots, for example. “We’re trying to make robots whose behavior evolves or adapts over time by making the engineering more like nature,” De Jong explained.
Take the robot that is the Mars Rover. “Right now we preprogram the Rover without knowing exactly what it’s going to encounter on Mars. So, when it gets into trouble, the engineer back on Earth has to reprogram it and upload the fixes. We’re trying to get the Rover to say, ‘Hmm, I seem to be stuck here. What can I do about it?’ Given the unexpected, the Rover could do its own adapting and learning.”
Get ready to meet such adaptable robots a lot more in daily life, De Jong said: in health care settings, for example, or in any situation risky for soft-fleshed creatures. Imagine a robot as the lead “agent” on a SWAT team sent into a hostage situation or into a hazardous waste spill.
In an interesting turn, evolutionary computational techniques that were inspired by nature are now helping us understand natural phenomenon, De Jong said. At the Krasnow Institute for Advanced Science of George Mason University, where he is associate director, De Jong and a team modeled the clinical data related to the treatment given to postal workers who died in 2001 after inhaling anthrax powder sent through the mail.
“We built what is the first attempt at a computer model of the effects of inhalation anthrax on the human body,” he said. In the obvious absence of human volunteers, doctors can try out various “what-if” treatment scenarios for anthrax on the model instead, an example of in silico science.
According to De Jong, computer models can help us understand complex systems with long-range effects; these models allow us to explore questions like: What happens if a particular invasive species is introduced to an ecosystem? What happens if a certain incentive is subtracted from our economy or added to Afghanistan’s?
In fact, De Jong, whose interests have always been interdisciplinary, works to get scientists from across a broad range of disciplines talking with the computer scientists who make models.
A computer model can’t predict any outcome exactly or explain why it happened. The actually evolving universe is too complex and mysterious for the best evolutionary computation. But it can, De Jong said, “provide a probabilistic envelope of likelihoods on which you can act. That’s better than sitting with your feet on the desk scratching your head.”
Giving to Calvin
Majors & Minors
People at Calvin