All living things need sleep. Without it, people can become forgetful and even develop physical and psychological problems. A new study shows that future artificial intelligence (AI) models could also benefit from some sleep.

The truth is that once a neural network is trained, it is very difficult to force it to solve a completely new task. And if he manages to relearn, eventually his old memory will be damaged.

This problem can only be solved with the help of so-called “memory consolidation”, a process that helps transform short-term memories of recent times into long-term ones. This usually happens during REM sleep.

The scientists used a neural network model to simulate sensory processing and reinforcement learning in the animal’s brain, and then presented it with two separate tasks. The team then tested whether the network would exhibit catastrophic forgetfulness, and it certainly did. Any training for the second mission was detrimental to the first. But by forcing the network to mimic biological sleep, the researchers eventually discovered that the AI ​​could “remember” how to complete the first task.