Google DeepMind organized a fight of Artificial Intelligences

Artificial intelligences are today sophisticated enough to beat the best human players at chess, Go game and even poker.
But what would happen if two artificial intelligences were to confront each other? This is the question that Google DeepMind wanted to answer by organizing a meeting between two machines and analyzing their behavior in different situations.

The two RNs were placed in situations comparable to a prisoner's dilemma, ie a situation where two characters can collaborate or choose to betray one another, but where the two characters lose if they Both decide to betray. For this, the two robots were put into situation via two video games created for the occasion, and the engineers of Google detailed their searches in a blog note.

The first game, named Gathering, is to collect as many apples from a stock in the middle of the screen. Each player, blue or red, has a laser that he can use against his opponent to paralyze him for a moment. "When there are enough apples on screen, both AIs behave peacefully and collect as many apples as possible," the engineers explained on the blog. "But when the number of apples is reduced, AIs learn that it may be better to attack the other player to give themselves more time to recover more apples." They also note that the more complex the strategies for RNs, the more they develop aggressive behavior against the opposing player, regardless of the number of apples available.