April 22nd, 2007

Anhinga, snakebird

Digital evolution: some recent studies

I've read some recent studies on Darwinist evolution of the computer code, and what is required to make such evolution computationally feasible. Basically, there are two main problems: almost all mutations of a typical piece of code are lethal, and the rewards landscape is too sparse (only goals we care about are rewarded), so the evolutionary process cannot find a "ladder" to climb to the goals. A couple of recent papers present case studies of systems where these problems were sufficiently rectified to enable nontrivial evolution.Collapse )