Immobots Take Control

It's interesting that programmers are able to develop systems that think of things that the programmer can't ("Immobots Take Control," TR December 2002/January 2003). Or, more precisely, things that the programmer didn't. And here lies the fundamental barrier to intelligent computing: our reality must be distilled into terms the system can understand, and the frame of reference for that understanding is defined by the model. It's time, however, to put our models on the shelf and do some real thinking.

Michael Schupan
Ottawa, Canada

This is a good article that touches on a deep subject. Clearly, immobots can fail. How far would you trust an immobot driver of your car? Would you fall asleep and let it drive? What if one of the sensors fails and this leads the immobot to incorrectly deduce that the sensor reading is seeing a very low value? Would the immobot incorrectly invoke an emergency procedure, amplifying a minor problem into a major one? Can we countermeasure by adding a second immobot to watch over the first immobot? This quickly leads to a Rube Goldberg tower.

Donald J. Christian
Fremont, CA