Why Are Dreams Visual? The Defensive Activation Theory of REM Sleep

Short answer: one provocative hypothesis proposes that dreaming exists to keep your visual cortex switched on while you sleep, because vision is the only sense that goes fully dark at night. Touch, hearing, and smell still get input in the dark. Vision does not. So the brain may fire up the visual system from the inside, in bursts, to stop neighbouring brain regions from moving in and taking over the idle territory. This is the "defensive activation" idea, and I want to walk through why it is clever, what evidence backs it, and why it is still a hypothesis with serious rivals.

I find this one fascinating because it reframes a question everyone has asked since childhood. Not "what do dreams mean," but "what is dreaming for, mechanically." Let me lay it out flat, then poke at it.

What is the defensive activation theory?

In 2021, neuroscientists David Eagleman and Don Vaughn published a paper in Frontiers in Neuroscience titled "The Defensive Activation Theory: REM Sleep as a Mechanism to Prevent Takeover of the Visual Cortex." Their core sentence is worth quoting directly: they hypothesise that "the circuitry underlying REM sleep serves to amplify the visual system's activity periodically throughout the night, allowing it to defend its territory against takeover from other senses."

REM stands for rapid eye movement, the sleep stage where most vivid dreaming happens and where your eyes dart around under closed lids. The theory's claim is that those dreams are mostly visual for a reason. They are a nightly defensive light show, keeping the lights on in a region that would otherwise sit dark for hours.

Here is the engineering logic. The brain is not a fixed circuit board. Cortical territory is contested real estate. When one region falls quiet, its neighbours encroach. So the visual cortex faces a nightly threat that the auditory and somatosensory cortex do not, simply because the sun sets.

Why would only vision need this, and not hearing or touch?

This asymmetry is the part of the theory I think is genuinely elegant, and it is the spine of the whole argument. The authors put it plainly: in their words, "the rotation of the planet does not diminish touch, hearing, or smell," whereas "only visual input is occluded by darkness." They give the homely examples. In the dark you still feel a bug crawl on you, still hear your baby cry, still smell smoke.

So during a long night, the other senses are dimmed but not silenced: touch still registers pressure and temperature, hearing still works and can still wake you, smell and the body's own internal senses keep reporting. Vision is the exception. With the eyes closed in the dark the retina delivers essentially nothing, fully and on a predictable schedule. Only vision goes all the way dark, every night and on the clock. If idle cortex is vulnerable cortex, then vision alone needs a defence mechanism. Here is the unification it buys: "why is dreaming so visual" and "why does only vision need protecting" collapse into one question with one answer. That framing is Eagleman and Vaughn's; the emphasis on it as the theory's most discriminating card is mine.

How fast can the brain repurpose unused cortex?

For this to make sense, takeover has to be fast. If reclaiming idle visual cortex took months, a single night of darkness would not matter. The deadline has to be tight enough that nightly defence is worth the metabolic cost.

The evidence here is the worked example, and it comes from human and animal deprivation studies that anyone can look up. The striking short-clock result comes from Lotfi Merabet and colleagues (2007), cited by Eagleman and Vaughn: when sighted volunteers were blindfolded and then asked to do a fine touch-discrimination task, researchers detected touch-related activity in the primary visual cortex after only 40 to 60 minutes. The eyes were covered for under an hour, and visual cortex was already moonlighting for the sense of touch.

A second human study makes the same point on a slightly longer clock, and it carries an important caveat the first one does not. Work by Lotfi Merabet and colleagues (2008) found that five days of complete blindfolding, paired with intensive Braille training, recruited the visual cortex for touch on fMRI, with stimulation of that region by magnetic pulses now disrupting Braille reading. The catch: the effect vanished within 24 hours of taking the blindfold off. The authors are careful to read this as unmasking of connections that were already there, latent, not the growth of new structure. That distinction matters for the theory, because fast unmasking is exactly the kind of nightly threat a defence mechanism would be guarding against.

For genuinely durable, structural change you have to leave the day-timescale and go to the rodent work. A 2021 review in Frontiers in Neural Circuits by Gabrielle Ewall, Hey-Kyoung Lee, and colleagues notes that as little as two days of visual deprivation strengthens the connections between neurons in the visual cortex of mice. So the timescale runs from tens of minutes of functional unmasking to a couple of days of measurable synaptic change in rodents. Fast enough that a long night is not nothing.

Does the cross-species evidence hold up?

This is where the theory makes a quantified prediction and tests it. If REM exists to defend plastic cortex, then the most plastic brains should need the most REM. And the most plastic brains belong to the most immature newborns.

The pattern is real and well documented. In humans, REM accounts for about half of an infant's sleep time but only 10 to 20 percent of an adult's. The brain dials down the nightly defence as it becomes less malleable. Across species, the same logic plays out through altriciality, a useful word meaning how helpless and undeveloped an animal is at birth. As the sleep researcher Jerome Siegel summarised back in 1995, "altricial mammals, that is animals born in a relatively immature state, have high amounts of REM sleep at birth," and adults of altricial species keep "considerably greater" REM than adults of precocial species, the ones born ready to run.

Eagleman and Vaughn pushed this further. They analysed 25 primate species and reported that behavioural measures of plasticity, such as how long a species takes to reach weaning, adolescence, and locomotion, correlate positively with the proportion of REM sleep. The relationships are modest in strength (r-squared of 0.17 for weaning, 0.22 for adolescence, and 0.32 for locomotion, taken one at a time, and about 0.29 when the three are entered into one model together), but they point the predicted direction: more developmental plasticity, more REM. Altricial animals can carry up to roughly eight times the REM proportion of precocial ones. One thing this correlation cannot do, and it is worth saying plainly, is pick this theory out of the lineup. A memory account or a developmental account of REM predicts the very same plasticity link for its own reasons, so the numbers support "REM tracks plasticity," not specifically "REM defends visual territory." The comparison that would actually separate the theories is the darkness test in the next section, and it has not been run.

What would prove this theory wrong?

A theory you cannot break is not science, it is a story. The defensive activation account is testable, and the authors say so out loud. Their own stated falsifiable claim is the plasticity one from the last section: the more plastic an animal's cortex, the higher its share of REM. The version I keep reaching for is a sharper comparative extension, and the honest move is to label it as mine, not theirs. The right variable for it is not darkness as such but how long the visual cortex sits idle, with no input to keep it busy. For ordinary daytime animals darkness is what drives that idleness, but the two can come apart, and that is what makes the prediction discriminating: leave the visual cortex idle longer, by any route, and you should see proportionally more REM to defend it.

Eagleman and Vaughn offer a concrete test that turns on darkness. Elephants see poorly on moonless nights. If REM responds to how dark the night actually is, their framework predicts a given elephant should get more REM on cloudy or new-moon nights than on clear nights under a full moon. The idle-cortex reading also explains why elephants barely seem to dream at all: their night-adapted retinas keep the visual cortex driven through the same darkness any animal faces, so the cortex is rarely idle even at night, and on the idle-time variable, not on hours of darkness, they should need almost no defence. That is a concrete, checkable prediction with a date stamp on the experiment you would run. If idle visual-cortex time does not track REM across species or across moon phases, the theory takes a real hit.

Can the brain really repurpose unused cortex? A historical parallel

The intuition underneath all this is old. In the nineteenth century, anatomists fought over whether the brain was a single uniform organ or a patchwork of regions with specific jobs. The localizers won. We now know cortex is parcelled into territories, and we know those borders are not painted on. They move.

The clearest demonstrations came from sensory deprivation work in the twentieth century. Take away an input, and the orphaned cortex gets colonised by its neighbours. Blind readers process Braille partly in what would have been their visual cortex. The defensive activation theory simply takes that well established fact about contested neural borders and asks: if borders shift this readily during the day, what stops them shifting every single night when vision goes dark? Its answer is that something has to actively hold the line, and dreaming might be that something.

So is this why we dream? Honest caveats

Here is where I keep the honesty pointed at the evidence, not at the thesis. Defensive activation is a hypothesis, and an extension of older ideas, not settled fact. It has rivals that explain a lot of the same data, and you should hold all of them loosely.

The oldest competitor is the activation-synthesis hypothesis, proposed by John Allan Hobson and Robert McCarley in the American Journal of Psychiatry in 1977. Their view: dreams are the cortex trying to make a story out of essentially random signals fired up from the brainstem during REM. Dreaming is the narrative, not the purpose. Hobson later revised this into a broader "protoconsciousness" model around 2009.

A second major camp says sleep is for memory and synaptic housekeeping. Giulio Tononi and Chiara Cirelli's synaptic homeostasis hypothesis, laid out in Neuron in 2014, argues sleep is "the price the brain pays for plasticity," scaling synapses back down after a day of learning. Related work by Robert Stickgold, Matthew Walker, and others ties sleep stages to consolidating memories. In those accounts, the function of REM is about information, not about defending cortical turf.

These are not all mutually exclusive. REM could plausibly do several jobs at once, and a periodic activation that defends territory could also serve memory. But the defensive activation theory makes a strong, specific, single-purpose claim, and that specificity is exactly what makes it both attractive and vulnerable. I should also be straight about the soft spots in the support I just leaned on. The cross-species correlations are suggestive, not decisive: Siegel himself cautioned that the immaturity-and-REM relationship explains only around a third of the variance, and it weakens once you bring in birds and non-placental mammals rather than the tidy altricial-mammal cases. The blindfold findings show cortex can be reclaimed fast, mostly by unmasking latent connections, not that nightly dreaming is what stops it. And the elephant prediction, as far as I know, has not been run.

There is also a sharper empirical objection the theory has to live with, the one a critic will raise first: people blind from birth still have REM sleep and still dream, just not in pictures. That is awkward for an account built on defending visual cortex specifically. The answer on offer is that the congenitally blind dream in their other senses, and that people who lose their sight later in life keep more visual dream content, which at least fits the cortex-shaped-by-its-input logic. But the objection is real, and you should weigh it, not file it away.

So treat this as a beautiful, falsifiable idea on probation. Not a fact about your brain, and certainly not anything to act on. What I like about it is the shape of the argument. It takes one mundane fact, that the planet turns and only vision goes dark, and turns it into a candidate reason for the strangest few hours of your day. That is what a good hypothesis looks like before the verdict comes in.