Ken Stanley (University of Central Florida)
Abstract. The process of natural evolution on Earth has discovered over its history a vast diversity of complex (sometimes astronomically so) and ingenious capabilities that often provide inspiration for human inventors, such as photosynthesis, flight, and even human intelligence. While much is known about evolution in nature, attempts to replicate its prodigious creativity through evolutionary algorithms on computers have so far fallen far short, especially in the pursuit of high-level complexity, raising the intriguing question of what is missing from our understanding of its mechanisms. Answering this question would not only impact our ability to harness evolutionary search in the service of human aims such as artificial intelligence and robot design, but more fundamentally could enhance our understanding of all processes of creativity and discovery, both in nature and human society. Taking a step in this direction, this talk will share counter-intuitive results from computational experiments in evolution that show that divergence rather than adaptation or optimization is often the fuel for high-level discovery, and consequently that search processes of all types are often most effective when liberated from seeking any specific objective. This insight alters our perspective not only on the development of new evolution-inspired search algorithms, but also on social endeavors such as the funding of science and the improvement of education.