What sleep is for
Animals from jellyfish to humans sleep and too little sleep harms health, yet no one has shown for certain what sleep does for the brain.
Open in the interactive tree →Sleep takes about a third of a human life and leaves an animal defenceless, so it must serve something essential. Leading ideas are that it resets the strength of connections between brain cells, clears waste, repairs cell damage and files memories. Each idea has supporting and conflicting experiments, and several may be partly true.
As of now
In January 2026 Nature Communications reported that jellyfish (Cassiopea) and sea anemones sleep, that neurons gather DNA damage during waking and sleep deprivation and lose it during sleep, and the authors propose sleep began as a way to limit cell damage. The synapse-reset idea rests on a 2017 Science study in which synapses in mouse cortex were about 18 percent smaller after sleep, while the waste-clearance (glymphatic) idea is disputed: a 2024 Nature Neuroscience study in mice found clearance slower in sleep and anaesthesia than in waking, and a rebuttal was published in March 2025. Stanford's SleepFM, trained on about 585,000 hours of overnight recordings from 65,000 people, predicted future risk for 130 of more than 1,000 disease categories with reasonable accuracy, including dementia at a C-index of 0.85, in Nature Medicine on 6 January 2026.
What is missing
- A trusted way to measure waste clearance and fluid flow in the sleeping human brain
- Experiments that switch off one proposed function of sleep without disturbing the others
- Tests across many species and brain areas to show which functions are shared
- A measure of sleep need, to tell when sleep has been enough
- An explanation for why people differ so much in how much sleep they need
Becomes possible once solved
- Treatments aimed at the function that fails in insomnia and sleep disorders
- Early warning of dementia and other disease from sleep recordings
- Evidence-based advice on how much and when to sleep
- Shift-work and travel schedules that protect health
Open steps
- Measure brain clearance in sleep Medium AI leverageDevelop a trusted, non-invasive way to measure waste and fluid movement in the sleeping brain, to settle whether sleep speeds clearance.
- Test the synapse-reset idea Medium AI leverageMeasure synapse size and number after sleep and waking in more brain areas and species, to see whether sleep resets connections everywhere.
- Is sleep a repair shift? Low AI leverageTest in mammals whether waking raises neuronal DNA damage and sleep lowers it, as reported in jellyfish and sea anemones in 2026.
- A measure of sleep need High AI leverageFind a signal in EEG, wearables or blood that shows how much sleep a person needs and how well a night met that need.
- Switch off one function at a time Medium AI leverageUse closed-loop stimulation and genetic tools to disturb one proposed sleep function, such as memory filing or repair, and see what fails.
Where AI could help
Medium AI leverage. AI can read huge sleep and brain datasets, but deciding what sleep is for needs causal experiments in animals and people.
- Models trained on overnight recordings that link sleep patterns to health outcomes
- Automated analysis of brain images and electron microscopy of synapses
- Simulation of fluid flow through brain tissue
Shown so far
- Stanford's SleepFM, trained on about 585,000 hours of overnight recordings from 65,000 people, predicted future risk for 130 of more than 1,000 disease categories from one night with reasonable accuracy, including dementia at a C-index of 0.85 (Nature Medicine, 6 January 2026). source
Prerequisites
- EEG1924
- REM sleep discovered1953Sleep stages found in 1953 frame every theory of what sleep is for
- Large-scale neural recording1990–2017
- Optogenetics2005
- Brain wiring maps2024