Experts at the OpenAI laboratory specializing in the study of artificial intelligence have managed to train the neural network to play hide and seek. Representatives of the company demonstrated how, using simple machine learning methods, it is possible to obtain complex models of AI behavior in various situations.
The game is designed in such a way that players who need to find the hidden get points as soon as the search object falls into their field of vision. Initially, those hiding are given little time to search for a shelter plan, which may be objects on the playing field with the possibility of their movement.
Artificial intelligence played about 481 million games, after which it began to develop its own strategies and counter-strategies. Instead of a chaotic run, the characters who were looking for other participants began to coordinate and more efficiently "comb the terrain." Those who were hiding began to build shelters, preventing the seeker from climbing onto the wall or hiding behind game blocks.
Using a simple machine learning technique, the researchers managed to create a rather complex algorithm. The incentive for AI in the search for effective hide and seek strategies was the reward system for the characters found. In the future, the study of the behavior model of artificial intelligence will help in the development of computer games of various genres.