Because of the active development of machine learning technologies, artificial intelligence and robotics, many experts express their concerns about the possibility of so-called machine revolt. Last but not least, this is due to the fact that in numerous fantastic stories and films, robots subjugated humanity or completely destroyed people for one reason or another. To prevent this, the two largest developers in the field of artificial intelligence decided to combine their efforts. We are talking about companies DeepMind, owned by Google, and Open AI, founded by Ilon Mask - the creator of SpaceX and Tesla Motors.

DeepMind and Open AI published a research article describing a new method of machine learning. Its essence lies in receiving signals from people to study new tasks. Experts believe that this should be safer than allowing artificial intelligence to solve this or that task independently, which can lead to unexpected consequences.

According to researchers, the main problem is that artificial intelligence uses the most effective method for achieving maximum benefits. This is comparable to shoveling all the scattered things from the floor into the closet and calling this room clean. Purely technically the room is really clean, but this is not the result that is required. Machines are able to find such workarounds and use them to solve any problem.

The urgent problem is the reward system. It was in this direction that both companies focused their efforts.

Instead of creating an overly complex reward system that machines can bypass, researchers used human input to reward artificial intelligence. When the car solves the problem, as the coaches wanted it, it gets a positive response. With this method, the system was able to learn how to play simple video games.

DeepMind and Open AI have just started to work in this direction, but companies believe that human control will help prevent unwanted consequences of decisions taken by the machines.