OpenAI, a San Francisco AI research company founded by Elon Musk, has taught an AI to play hide-and-seek. Having played millions of game, the algorithm learned how to use blocks and ramps in their virtual environment to win the game.
The idea of developing this AI came to OpenAI Five bots became the first AI who beat a human team of professional players in Dota 2. Fans of this game are often impressed with how canny and quirky bots can be. Therefore, the research team was wondering if they could use team dynamics of OpenAI Five bots to get the skills that could potentially help humans in the future.
How does it work?
The game includes two teams: hiders that are marked with blue color and seekers, they are red. They are in a virtual environment with walls, boxes, and ramps and left to play AI hide-and-seek. Hiders didn’t cooperate with each other in the beginning, as well as seekers. At first, hiders just ran away from seekers and, of course, they were caught. After several games, they learned that they can use boxes to block the entrances between the walls. Watch this video created by OpenAI to learn more about other glitches that bots figured out.
In their turn, seekers figured out that with the help of ramps they can jump over the walls. But their joy lasted not so long. Hiders learned that they can hide ramps in their rooms before seekers get them. After some time hiders understood that cooperation is a key to success. They started to work as a team and use common efforts.
The main goal of creating this AI is to show that AI can be as smart as a human and maybe in some cases even find the solutions that are not visible to a human. This simple experiment shows the parallel between humans and AI. Basically, this game is a reflection of an evolution process. Here we see how bots adjusted their skills and knowledge to the environmental obstacles. As well as humans who have come a long way to the level of intelligence and knowledge that we have now.
Scientists from OpenAI don’t stand for replacing humans with AI. Instead, they insist on the ability of AI to see the way outs that humans sometimes can’t find. Ai is just faster and more mentally stable than humans.