Science

Google is building artificial intelligence

The time when an artificial intelligence will be able to learn, to think, even “feel” without the intervention of man, is not as far as you imagine. Already gaining people to gaming world and now comes a new study from Google reveals that the AI ​​is capable of both cooperative behavior, and for competitive.

Google’s scientists put two relatively simple scenarios DeepMind. In the first case the “Gathering” name, the DeepMind parted in two, in Blue and Red, with the purpose to collect green “apples” from a confined space. Indeed, the Blue and the Red armed with lasers that could invalidate a short time the opponent. So, the choices were two. Either to collect all the apples for themselves or allow each other an approximately equal distribution.

After thousands of script applications, Google concluded that the DeepMind acted very peacefully and cooperatively when there were enough apples. But there were fewer, the better were the chances to turn against one another, a behavior that we find everywhere in the animal kingdom, including humans. Also the smaller and less “smart” neural networks were more cooperative . Larger and complex networks tended to betray and be selfish.

In the second scenario called “Wolfpack”, the Red and Blue had to pursue a vague form of prey. They could try to catch separately or together. In smaller networks, the results were mixed, larger networks but quickly realized that cooperation is more important than competition in this situation.