Why do we dream?

The question

Every night, with the eyes closed and the body almost still, the brain creates a world: places, people, conversations, fear and joy. We accept it completely while it lasts, and in the morning most of it is gone. People have asked what dreams mean for as long as there are records; the ancient world treated them as messages, and Freud saw them as disguised wishes.

The scientific question is simpler and still open: do dreams do something, or are they a side effect of a sleeping brain? And how can a brain cut off from the senses produce experience at all?

What we know

In 1953 Eugene Aserinsky and Nathaniel Kleitman discovered REM sleep, with rapid eye movements and an active brain. People woken from it usually report vivid dreams, and for a while dreaming and REM sleep were treated as the same thing.
They are not. People woken from non-REM sleep also often report dreams, usually shorter and less vivid. And patients with damage to certain forebrain areas stop dreaming although their REM sleep is intact.
In 2017 a team measuring brain activity in people woken many times a night found a "hot zone" at the back of the cortex: lower slow-wave activity there predicted dreaming, whether in REM or non-REM sleep.
Dreamers can communicate. In lucid dreams, when the sleeper knows they are dreaming, they can signal with agreed eye movements. In 2021 four laboratories held two-way exchanges with dreamers: they asked simple maths questions, and the sleepers answered correctly from inside the dream.
Dreams are linked to learning. People who dreamed about a task they had just practised did better on it after sleep.

Main hypotheses

A side effect

According to the activation-synthesis hypothesis of the 1970s, the brainstem sends random signals during REM sleep, and the cortex tries to weave them into a story. The dream itself would then have no function, as the sound of a working engine doesn't.

Rehearsing threats

Dreams are often about danger, chases and failures. Perhaps they evolved as a safe simulator where the brain practises responses to threats without risk.

Processing memories and emotions

Dreams could be the conscious side of how the brain sorts fresh memories, links them to old ones and softens emotional experiences.

Protection from overfitting

A newer idea borrows from machine learning: a brain that learns only from daily experience would become too narrowly tuned to it. Strange, distorted dreams would work like noise added during training, helping it generalise.

Where understanding ends

The main difficulty is that dreams can only be studied through reports given after waking, and memory of dreams is fragile and easily rewritten. We know how often people report dreams, but not how often they actually have them: someone who remembers nothing may have been dreaming all night. The question is close to the problem of anaesthesia: lack of memory and lack of experience are not the same.

Another gap: the brain processes memories during sleep whether or not we dream. So even if dreams accompany learning, it isn't clear whether the experience itself does something or is only a glow above work that would happen anyway.

And the deepest question comes back to consciousness. A dream is experience with no input from the senses, generated by the brain entirely by itself. If we understood how that happens, we might understand how the brain creates experience while awake.

What would settle it

A way to detect dreaming from brain activity reliably, without depending on memories after waking.
Experiments that change dream content and show an effect on learning or mood that doesn't appear when the same sleep processes occur without dreaming.
Real-time communication with dreamers developed far enough to study dreams while they are happening.