Wednesday, February 11, 2009

Robots that Evolve and Improve (Relentlessly?)

Roboticists at Aberdeen University, under team leader Christopher MacLeod, have been experimenting with robots that they program to fulfill one objective. The robots are designed around The Incremental Evolutionary Algorithm (IEA) using a neural network. They have learned to perform a task such as walking and then re-learned and improve this task when the research team added knees to a system that already had hips and legs. The system is able to "decide" to add more neurons in order to accommodate additional features. Eventually the robot may even learn to instruct the human engineers to add additional limbs or components in order to fulfill its task more effectively.

In reporting on this technology an article in the online The Daily Galaxy alludes to melodramatic prospects that such systems might be relentless in their evolution, and may request machine guns or other weaponry to fulfill their goal.

No comments: