Sleep can feel like a pause button. Your body slows down, your eyes close, and the outside world fades away for a few hours.
For a long time, scientists have explained sleep as a time for rest or a chance for the brain to clean itself up after a busy day.
Helping the brain stay flexible
New data suggests that sleep has a bigger purpose outside of simple downtime.
It may help the brain stay flexible enough to recover from stress, adapt to change, and keep learning over a lifetime.
That ability, the authors argue, could explain why sleep has survived through hundreds of millions of years of evolution despite leaving animals vulnerable to predators.
“We wanted to move past the idea that sleep is simply a battery recharging overnight,” said Professor Xiaohui Wang of the Changchun Institute of Applied Chemistry, Chinese Academy of Sciences.
“When you look at the brain as a complex dynamic network, sleep starts to resemble something a careful engineer would design on purpose, a scheduled window for the system to repair and reorganize itself.”
Beyond staying stable
The researchers say many discussions about sleep mix together three different ideas, stability, robustness, and resilience. Although they sound similar, they describe different abilities.
Stability means the brain can keep working despite small disturbances. Robustness means it can continue functioning even when faced with noise, gradual changes, or partial damage.
Resilience goes further. It describes the brain’s ability to absorb a setback, reorganize itself, and regain healthy function over time.
The authors argue that sleep plays its biggest role in protecting this resilience across a network of roughly 86 billion neurons.
Two stages, two jobs
The paper describes the two main stages of sleep as partners with different responsibilities.
During non-rapid eye movement, or NREM sleep, especially deep slow-wave sleep, brain activity slows to rhythms below one hertz.
During this period, connections between neurons that have grown stronger through learning are adjusted so the brain does not become overloaded.
Rapid eye movement, or REM sleep, takes a different approach. Brain activity becomes more active and less synchronized.
The researchers suggest REM sleep helps the brain explore different patterns of activity and loosen connections that have become too rigid.
An important transition during sleep
The transition between these two stages may be especially important. According to the authors, the brain reaches its highest level of metastability during these moments.
This causes it to be more ready to shift between different activity patterns. Deep NREM sleep also supports another important process.
During this stage, the brain’s glymphatic system expands the spaces between cells and clears away metabolic waste, including amyloid-beta, a protein linked to Alzheimer’s disease.
While brain connections are being adjusted, waste products are removed at the same time.
“A network that only optimizes can paint itself into a corner. What sleep appears to protect is the ability to get back out,” said Longwei Yang, the study’s lead author.
Lessons for AI
The authors believe these ideas could reach beyond biology. Modern artificial intelligence systems often struggle with catastrophic forgetting, overfitting, and network saturation.
When trained on new tasks, they can lose knowledge gained from earlier ones. The paper points to several studies showing that sleep-inspired techniques can help.
Sleep Replay Consolidation allowed artificial neural networks to hold on to old skills while learning new ones.
Other researchers reported similar improvements in spiking neural networks after introducing sleep-like offline activity.
Taking a break to reorganize
Another study found that continuously trained networks developed unhealthy activity until oscillating noise, designed to resemble slow-wave sleep, restored stable behavior.
“The striking thing is the convergence,” said Professor Haohong Li of Zhejiang University School of Medicine, whose research centers on sleep homeostasis and neural oscillations.
“Replay, renormalization, pruning, and noise injection turn up as solutions in both brains and machines.”
“We are not claiming that a server farm sleeps. We are saying that any learning system running in a changing world may need structured offline phases to remain resilient.”
The authors suggest that future AI systems could alternate between NREM-like periods that strengthen existing knowledge and REM-like periods that reorganize information.
Instead of judging performance by accuracy alone, they argue that researchers should also measure how well a system recovers after disruption.
Implications for human health
The idea may also help explain why disrupted sleep often appears alongside conditions such as Alzheimer’s disease, schizophrenia, and epilepsy.
According to the review, these disorders are associated with fragile brain networks and disturbed sleep patterns.
The authors suggest that treatments designed to strengthen slow-wave sleep could eventually improve the brain’s ability to recover rather than simply provide a better night’s sleep.
Quiet hours support resilience
The researchers also stress that this remains a perspective rather than proof.
Their proposal combines existing findings into a framework that can be tested through future experiments.
Changing sleep-like phases should improve recovery after disruption, alter measures of brain activity in specific ways, and direct replay toward the most vulnerable parts of a network.
Instead of viewing sleep as a period when the brain shuts down, this perspective argues that those quiet hours may be when the brain does some of its most important work.
During sleep, the brain is preserving its ability to adapt, recover, and keep functioning in a changing world.
The full study was published in the journal Brain Medicine.
—–
Like what you read? Subscribe to our newsletter for engaging articles, exclusive content, and the latest updates.
Check us out on EarthSnap, a free app brought to you by Eric Ralls and Earth.com.
—–