Computer Scientists Systematically ‘Kill’ Robots To Hasten Evolutionary Process

It might sound odd, but what if the only way to rapidly improve something was to kill off most of what we have? We're talking something so grim, I almost hate to write it: mass extinction. The idea is that if a large part of the population is purged, it'll inevitably speed up the evolutionary process of surviving persons. Fortunately, there are no planned tests of such a theory outside of a computer simulation, such as one some scientists at the University of Texas in Austin developed.

Through much testing, Risto Miikkulainen and Joel Lehman discovered that when most of the learned information in a walking simulation was destroyed, the remaining AI began to evolve at a much quicker pace. As with real evolution, different mutations would be created over time with this simulation and result in a large number of different niches. To come to the conclusion that mass extinction can be a good thing, the scientists "killed" robots in 90% of the niches. Evil, but it's progress at stake.

Walking Simulation

After killing off all of these robots, there was proof that the surviving robots were the most evolvable and similarly had an improved chance of developing new behaviors, such as more effective ways of walking.

Where on Earth does such experimentation benefit us as humans? A couple of good examples are given, including the development of robots that could handle obstacles better, or even making a robot exploring Mars more nimble. Says Mr. Lehman, "This is a good example of how evolution produces great things in indirect, meandering ways".

It sure seems that way, but I wouldn't blame you if you found this entire subject a hard one to wrap your head around.


Show comments blog comments powered by Disqus