"Robots Are Learning, But No 'Terminators' Are About to Appear"

Investor's Business Daily (12/31/04) P. A4; Isaac, David

"Machine learning--the ability for a machine to expand its intelligence through experience--is making great strides, but a "Terminator" scenario in which the machines take over will not arrive anytime soon, according to researchers. Machine learning is a much more efficient and practical alternative to coding every circumstance a robot might have to contend with via programming. Tom Mitchell with Carnegie Mellon University's Center for Automated Learning and Discovery is using machine learning in an attempt to determine what kinds of neural activity take place when a person is reading, and he believes such research could delve into how the human brain organizes conceptual categories; at the same time, Mitchell cautions that machine brains cannot equal human brains, noting that the human mind's architecture is much more distributed than that of a computer. "We are frequently assured by the popular press that we are on the verge of having household robots that will do all of the housework, baby-sit our children, amuse us with lively conversation, and take care of us in our old age," explains philosopher John Searle, who dismisses such visions as "nonsense." Still, parallels can be drawn between machine learning and human learning: One such case is reinforcement learning, in which a reward impulse encourages proper behavior. Mitchell says that in much the same way the chemical dopamine "rewards" the brain with a pleasurable sensation, so does pressing a green button indicating good behavior reward a robot for proper responses or actions. Applied Perception research scientist Mark Ollis says the biggest challenge for robots is interpreting the things they see, and he is developing a self-navigating robot for the Defense Advanced Research Projects Agency (DARPA). IRobot, meanwhile, is working with DARPA on swarm, a project involving the development of hardware and software for robot communities; swarm technology would enable robots to communicate and learn from one another as they move, and be an ideal application for pinpointing an object's location."