Research by the OpenAI organization has made it possible for a group of bots to create their own language to communicate with each other.
Machines need to communicate with each other. This is becoming increasingly evident at a basic level. Hence, companies devote resources to making their programs understood with third-party products. When we talk about artificial intelligence, communication can have greater implications. And this aspect facilitates the interaction of autonomous systems , so that they coordinate among themselves.
The work carried out by the scientists of OpenAI, which is dedicated to the research of non-profit artificial intelligence, has achieved that a group of bots create their own language to communicate with each other. They did it for sheer survival. so to speak. They had to invent something to accomplish a task and it could not be anything but a language. It is the drift that nature normally takes: the spontaneous generation of forms of communication.
The group of OpenAI bots was placed in a scenario as simple as possible: a white plane. While each of the bots, which were points of a different color, were given a goal, such as moving to other color points in the plane. From there they alone they had to travel to reach their goals. And to do so it was necessary to understand each other.
The learning was based on trial and error. It is different from the typical training of an artificial intelligence algorithm, in which thousands of images of dogs are introduced to recognize this animal. In this case the bots had to create their own terms, to define the objects or actions. That’s how they ended up naming concepts like ‘going towards’ or ‘looking towards’.
The way to channel learning was through positive reinforcement. When bots changed their goals, they remembered what had worked in the past and applied it. In this way, every time improved their communication skills and efficiency at the time of completing the goals.
OpenAI researchers believe that the complexity of the language used by bots could be increased. That is, these algorithms could going beyond the basics of nouns and verbs they have developed in a blank plane.
Image: Luigi Mengato
