The new algorithm will teach robots to help each other.

Researchers at the Massachusetts Institute of Technology have created a mathematical model with which robots will learn some social skills. Scientists are confident that this will improve the relationship between humans and machines, as well as help with the recruitment of robots in the future. The press service of the university told about it.

During the experiment, scientists conducted a simulation on a two-dimensional grid. In it, robots had to fulfill physical and social goals. Finding a tree on a grid and watering it was a physical goal, and a social goal to figure out what another robot was going to do. Depending on this, he should have been helped, or, conversely, he had to stop the device. If, for example, a robot wanted to water a tree, then his partner sought to help him get to it faster.

According to scientists, by teaching robots basic social skills - to help each other, people will be able to expand the area of interaction between them and machines. The social skills of machines will help teach robots to help each other and them quickly assimilate with society, making robots full-fledged workers. In addition, the new model will help to quantitatively analyze social relations. Also, the data obtained can later be used to study autism or the actions of antidepressants.

As a result, the researchers identified three types of robots. Zero-level devices serve only physical purposes. Robots in the first category can think both physically and socially, but they think that all other robots act out of physical motives. Let's call them ideal workers for complex industrial areas, as well as ideal candidates as wage workers in hazardous areas of production. Tier 2 machines, on the other hand, imply that other robots are also capable of social relationships. These are potentially robot leaders. Those who will be able to organize the work of first-order robots, optimizing their actions, and building them into the desired algorithm. The scientists then gave a few people a look at recordings of how robots acted in a given situation, and then asked them to rate the social behavior of the machines. So scientists found out that the criteria by which the model and people determine social relations, in most cases, coincide. And this gives humanity hope that, after acquiring robots in the future, people will receive not just separate units, but robots will be able to unite with each other in a metaverse, where they will interact at different levels, improving the productivity of their own work for humanity.

The researchers plan to equip robots with a neural network to accelerate the accumulation of social experience - teach robots to help each other. In addition, they are working on a 3D sensor system that will allow robots to independently perform more complex operations, such as controlling household appliances.

Share with friends:

Write and read comments can only authorized users