The American neuroscientist Eric Hoel proposed a new hypothesis to explain the functional purpose of dreams. According to the scientist, the weirdness and hallucinatory nature of dreams is not a neurophysiological glitch caused by errors in neural networks, but an important feature that allows the brain to learn by simulating situations that go beyond everyday experience. The research results are published in the journal Patterns.
The question of why a person dreams is traditionally controversial in the scientific community. Inspired by artificial intelligence techniques used to train deep neural networks, Erik Hoel, an assistant professor of neuroscience at Tufts University, has proposed a hypothesis that dreams help our brains generalize everyday experiences.
"The original inspiration for deep neural networks was the brain," Hoel said in a Cell Press release.
One of the problems with machine learning is that the machine always thinks that the variety of situations it can face is limited by a set of learning patterns. To avoid this, AI developers introduce certain glitches and chaos into templates. For example, there is a data dropout technique that forces the machine to make decisions in the face of incomplete information.
So, during testing of unmanned vehicles, developers specifically remove certain areas of the internal screen image so that the navigation system predicts what may be hidden behind the black squares.
"There are an incredible number of theories about why we dream, but I would like to draw attention to the hypothesis that it is the experience of the experience in a dream that causes dreams," - says the scientist.
This hypothesis suggests that dreams make our understanding of the world less simplistic and more comprehensive.
"It is the unusualness of dreams, their difference from the experience of waking, which gives them a biological function," the author of the article writes.
In support of his point of view, Hoel cites the fact that situations often arise in a dream that are repeated many times in real life. The scientist claims that in this case, the state of overfitting is triggered, and the brain tries to generalize the task, creating dreams.
In part, according to the researcher, the function of "artificial dreams" is performed by feature films and novels - they act as a substitute for dreams. Virtual reality technologies play the same role.
The difference between the brain and artificial neural networks is that it is impossible to disable brain learning, and if there is not enough new experience in life for this, dreams are turned on. Thus, according to the author, the brain trains its perceptual and cognitive abilities and maintains its performance.