Why do we dream? Scientists still do not know the answer. But they have some ideas. One idea says the brain makes a story from random signals.

Another says dreams help us practice escaping from danger. Some scientists think dreams help us sort memories or calm strong feelings. A new idea says strange dreams help the brain become more flexible, like training for AI. Maybe dreams have more than one job.

Why do we dream? Science has not found a final answer, but there are several serious hypotheses. In 1977, two Harvard researchers suggested that during REM sleep the brain receives semi-random signals and tries to turn them into a story. In 2000, a Finnish psychologist proposed that frightening dreams are a kind of training, helping us practice dealing with danger.

Other researchers believe dreams reflect the brain sorting memories. In a 2010 study, people who dreamed about a maze task during a nap improved much more on the task later, although only four people reported such dreams. A 2021 hypothesis compares strange dreams to the noisy data used to stop AI from learning too narrowly.

In a 2010 experiment, volunteers practiced a maze task and then some of them took a nap. The four who reported dreaming about the maze improved roughly ten times more than nappers who did not, although four dreamers make a very thin basis for any theory. That study is one of several attempts to explain why we dream.

The activation-synthesis hypothesis, put forward at Harvard in 1977, holds that during REM sleep the brainstem sends semi-random signals to the cortex, which weaves them into a story. On this view, dreams may be a by-product with no purpose of their own. The threat simulation theory, proposed by a Finnish psychologist in 2000, takes the opposite position. It argues that common dreams of being chased or falling evolved as rehearsals for danger.

A 2021 proposal borrows an idea from machine learning. Bizarre dreams, it says, work like noise added to training data and stop the brain from overfitting to daily experience. No single idea has been decisively tested, and the hypotheses may not be mutually exclusive. One more complication comes from a 2017 study, which found that people often report dreams even when they are woken from non-REM sleep.