Researchers at the University of Freiburg in Germany have developed an algorithm that plays eight Atari games and advances within them using two methods.
In one method, the algorithm dispenses with trying to win and instead baits an enemy into killing itself before committing self-destruction as well, scoring just enough points to advance. The other strategy exploits bugs that enable the algorithm to cheat.
Google DeepMind's algorithms were taught to learn gameplay by observing the pixels on screen, while the Freiburg team applied "evolution strategy." The process begins with an initial approach for playing the game, then at each point in time randomly alters or mutates the strategy to produce new ones. The algorithm assesses which of these "offspring" strategies realizes the highest score, then further mutates the best-performing ones at the next time-step.
Over time, evolution guarantees the best strategies, or at least the most successful ones, are dominant.
From New Scientist
View Full Article - May Require Paid Subscription
Abstracts Copyright © 2018 Information Inc., Bethesda, Maryland, USA