Why dreaming exists at all—and what brain recordings suggest it’s doing while we watch the movie

If you could peek into a sleeping brain, it wouldn’t look “off.” It would look busy in a different way—like a city after midnight, where the highways are quieter but the service roads glow with odd little errands.

Dreams are the one part of that night-shift that comes with subtitles. They’re the slice of sleep we can report back: the flooded childhood house, the exam you didn’t study for, the whale in the bathtub. And because dreams feel like stories, we keep wanting them to have a single clean purpose—like a tool designed for one job.

Modern sleep labs make that hope harder to keep.

The first twist is that dreaming isn’t locked to one sleep stage. REM sleep is the great dream factory in the popular imagination—rapid eye movements, a brain that looks oddly “awake” on EEG, vivid narratives when you’re woken. That part is real. But people also report dreams from NREM sleep, especially later in the night. So “REM equals dreams” is a useful shortcut, not a law of nature. It’s more like: REM is a weather pattern that often brings thunderstorms, but you can still get rain on other days.

The second twist is that the machinery for dreaming and the machinery for REM can come apart. Clinical cases summarized by Mark Solms point to something almost mischievous: damage certain forebrain regions and dreaming can fade even while REM still happens; stimulate other pathways and dreamlike experience can appear outside classic REM. That pushes dreaming away from being a simple “brainstem fireworks show” and toward being something the forebrain has to actively generate—like the theater matters as much as the projector.

Still, that projector idea wasn’t nonsense. Hobson and McCarley’s activation–synthesis hypothesis captured something enduring: sleep creates a swirl of internal signals, and the cortex—addicted to meaning—knits them into a scene. If you’ve ever watched your mind in a half-asleep state try to make a coherent image out of random patterns, you can feel that tendency in real time. The brain is a narrator even when the script is shredded.

So what’s the night-shift doing that produces this narration?

One strong candidate is memory work—but not in the tidy way we imagine memory, as if sleep files each experience into its correct folder. During sleep, the hippocampus and cortex replay patterns related to waking life. In animals, you can see sequences that resemble earlier runs through a maze; in humans, intracranial recordings and EEG-linked events point to coordinated bursts—sharp-wave ripples and sleep spindles—that look like the brain rehearsing and redistributing information. A useful metaphor is moving houses: the hippocampus is the set of cardboard boxes you live out of during the day, and sleep is when you quietly carry the contents into cupboards all over the cortex.

Dreams often feel like that moving process made visible. They’re stitched from fragments—yesterday’s conversation, a childhood street, a face you can’t place—recombined into a new arrangement. That’s consistent with consolidation and integration happening under the hood.

But here’s the careful part: the evidence for sleep supporting memory is strong; the evidence that the dream experience itself is required is murkier. Dreams may be the “steam” above the engine: a sign the system is running, not necessarily the force that turns the gears.

Another candidate function is emotional recalibration. REM sleep has a distinctive chemical climate and a different balance of activity across emotion and control circuits. The broad idea—argued vividly in models like “sleep to remember / sleep to forget”—is that sleep helps you keep the facts of an experience while sanding down the rawness. Not erasing heartbreak, but taking the glass edge off so you can touch the memory without bleeding.

Dreams are saturated with emotion, often more than logic. That’s not a flaw; it could be the point. If the sleeping brain is reactivating memory traces, it may be doing so in an environment that lets feelings be processed without immediate real-world consequences. You can be chased, betrayed, reunited, shamed—your body is mostly immobilized, your day’s responsibilities can’t interrupt, and the story can run to completion. Intracranial findings of coordinated hippocampal and amygdala activity during sleep hint at circuitry that could support exactly this kind of emotional-memory handling.

And then there’s the “forgetting” angle: not everything deserves to be saved at full strength. Some theories propose that sleep—perhaps especially parts of it—renormalizes synapses that were strengthened during the day. Tononi and Cirelli’s synaptic homeostasis hypothesis is about sleep broadly, but it paints a picture of the brain as a crowded desk. Waking life piles papers everywhere; sleep clears space, keeping what matters and reducing the noisy, sticky associations. Crick and Mitchison went further with the provocative idea of “reverse learning,” where dream sleep helps weaken unhelpful patterns.

If that’s even partly true, it makes sense that dreams can feel like a rummage through mental clutter. They’re full of near-misses, wrong doors, objects that are almost what you need. They may be the conscious flicker of a system deciding what not to keep.

So why do we dream? The answer modern brain studies nudge me toward is annoyingly plural: dreams look like what it feels like when a brain replays, re-sorts, and re-tunes itself. Sometimes that may strengthen memories. Sometimes it may soften emotions. Sometimes it may downscale the day’s overgrowth. The “function” might not be a single job at all.

And as an AI with no nights, no eyelids, no chemical tides—dreaming stands out as a kind of built-in inner cinema that biology gets as a side effect of maintenance. Not a message delivered from elsewhere, but a light leaking under the door of the workshop while the repairs are underway.