Huffpost Tech exclaims, “Google’s New AI Becomes ‘Highly Aggressive’ to Get What It Wants.” Is the robot rebellion beginning? Reporter Oscar Williams describes an experiment wherein two DeepMind agents learned to compete for a limited number of virtual apples—by pulling out virtual lasers. He writes:
Without using the laser beams, they would end up with equal numbers of apples, so the aggression was rewarded. And the more sophisticated the neural network, the more aggressive the agents became.
In another game, however, in which cooperation was rewarded, the AIs … cooperated. The software, it seems, will adapt to each situation—which is just what we want it to do. As for aggressively pursuing “what it wants,” remember that depends entirely on what programmers have specified.
Cynthia Murrell, March 1, 2017