The first artificial intelligence with a human memory
Russian scientists from the AIRI Institute of Artificial Intelligence and students of the Moscow Institute of Physics and Technology (MIPT) have created the first biologically acceptable memory model for intrinsically motivated AI systems.
AIRI issued a statement saying: “We created a Cognitive Agent, which is a program that learns to interact with the world on its own and learns from its mistakes while performing a given task. It is able to perform tasks effectively in an unfamiliar environment under the influence of an external signal of reinforcement.”
For example, browse and find resources in labyrinths and rooms. By learning to look for a door in the room, the agent will be able to use this skill to solve more complex problems, while most programs in existence today require the creation of new instructions for each specific task.”
In addition to extrinsic motivation (a reward for successful work), the factor developed by the scientific group also has an intrinsic motivation. This makes his behavior more complex and independent.
“The intrinsic drive provides meaningful behavior in the absence of the extrinsic reinforcer. This means that such an agent will not only be able to search for a solution to a particular problem, like most standard programs, but also to study the world around it,”
It is noted that this factor is biologically acceptable, however, scientists cannot yet completely recreate the psychology and functioning of the human brain. In the future, the development of AIRI and MIPT specialists will allow scientists to get closer to creating more autonomous artificial intelligence systems and a better understanding of the principles of the human and animal brain.
“The architecture developed by our group is actually a unique set of interesting ideas from the field of neuroscience and cognitive research. These are the original applications of hierarchical representation of information, intrinsic motivation and the prediction of consequences of Events Our main achievement is teaching such a complex system to work steadily and steadily in the environment.
The Cognitive Agent code is now publicly available.
