N.V. Kharitonov
Have You Ever Wondered Why We Spend a Third of Our Lives Asleep?
From an evolutionary perspective, sleep is madness. Imagine you’re an animal in the wild. For a third of your life, you shut down consciousness and lose the ability to respond to threats. Any predator could sneak up. Any danger could catch you off guard.
Such an organism should have disappeared under selective pressure. But it didn’t. What’s more: all complex animals sleep. Birds sleep. Dolphins sleep (one hemisphere at a time). Even fruit flies show signs of sleep.
This means sleep must have a function so important that it outweighs the risk of being eaten.
What is that function?
The Brain Is Not Just a Computer That Gets “Tired”
For a long time, people thought sleep was simply rest—like turning off a computer to let it cool down. But that’s not true. During sleep, the brain is actually more active than when you’re awake. EEG shows intense activity. Dreams are complex visual and emotional experiences.
So what’s really going on?
Imagine your brain as an incredibly complex ecosystem. 86 billion neurons. Quadrillions of connections between them. Every thought, every action, every memory is a physical change to this system.
And this system has three problems that accumulate during wakefulness.
Problem One: Garbage
Neurons work. They consume energy and produce metabolic waste. In particular, beta-amyloid accumulates—a protein associated with Alzheimer’s disease.
In 2012, scientists discovered the brain’s own “sewage system”—the glymphatic system [1]. And they found something surprising: this system works only during sleep. During sleep, the intercellular space in the brain expands by 60%, and cerebrospinal fluid literally washes the brain tissue, flushing out toxins.
Try staying awake for just one night—beta-amyloid will already accumulate in your brain. Chronic sleep deprivation is a direct path to neurodegeneration [2].
The first function of sleep: cleaning.
Problem Two: Noise
When you learn something new, synapses—connections between neurons—strengthen. That’s good: you remember, you acquire skills. But there’s a side effect.
All connections that were active strengthen. Both important ones and random ones. Over the course of a day, “noise” accumulates—many weak, random correlations. The signal-to-noise ratio drops. More and more energy goes toward maintaining useless connections.
What to do? You need to “prune” the excessive connections.
That’s exactly what happens during slow-wave sleep (NREM). In 2003, neuroscientists Giulio Tononi and Chiara Cirelli proposed the synaptic homeostasis hypothesis: during sleep, the brain weakens synapses that haven’t been sufficiently reinforced [3]. Experiments have confirmed: after sleep, evoked potential amplitude decreases by 15–25% [4]. The brain “prunes the garden,” keeping only what truly matters.
The second function of sleep: pruning.
Problem Three: Dead Ends
Sometimes you struggle with a problem for hours. You try options. Nothing works. You go to sleep. And in the morning—eureka! The solution comes on its own.
This isn’t mysticism. This is the third function of sleep, and it happens during REM sleep, when we dream.
Imagine your brain searching for a solution in a vast space of possibilities. It’s like searching for the deepest valley in a mountain range. You walk downhill. But you can get stuck in a local pit—it seems like you can’t go any lower, but actually there’s a deeper valley, separated by a ridge.
To get there, you first need to go uphill—increase “entropy,” temporarily enter a more chaotic state. In mathematics, this is called “simulated annealing.”
REM sleep is exactly that kind of “annealing” for the brain. Connections loosen, associations become more chaotic, unexpected combinations are generated [5], [6]. And when you wake up—the solution that was hidden behind the barrier becomes obvious.
The third function of sleep: recombination.
Dissipation Is Not the Goal—It’s the Cost
Here’s the most important part.
In physics, there’s a concept called dissipation—the dispersal of energy. We’re used to thinking of dissipation as loss, leakage. But in the case of intelligence, it’s the opposite.
When you solve a problem, create something new, you expend energy. Part of that energy goes toward “useful work”—a new neural connection, a new structure. But part inevitably goes toward heat, noise, excessive connections. That’s dissipation.
Dissipation is not the goal. It’s the cost of doing work. Like friction in an engine: you don’t build an engine so it can heat up, but it inevitably does. And if you don’t remove the heat—the engine burns out.
Sleep is the brain’s cooling system. It doesn’t make you smarter. It removes the consequences of what you’ve already done. It quiets the noise, hauls away the garbage, recombines what you’ve accumulated, so you can keep moving forward.
Sleep is meta-dissipation: dissipation that enables future dissipation.
What Does This Mean for Us?
First. Getting enough sleep isn’t a luxury. It’s a condition for preserving intelligence. Chronic sleep deprivation isn’t just fatigue. It’s the accumulation of noise, garbage, and getting stuck in local dead ends.
Second. Understanding sleep as a necessary dissipation phase gives us a new way to look at artificial intelligence.
Modern neural networks face the same problems: overfitting (an analog of noise accumulation), catastrophic forgetting (an analog of being unable to learn new things without losing old ones).
And they have their own “sleep analogs”:
- Regularization (L1, L2, dropout) — an analog of synaptic downscaling [7].
- Generative adversarial networks — an analog of REM recombination [8].
- Experience replay — an analog of memory consolidation [9].
Perhaps creating truly stable, long-lived artificial intelligence will require building in a full “sleep” cycle—phases where the system doesn’t work but “cleans,” “prunes,” and “recombines.”
The Main Takeaway
Sleep is not an evolutionary mistake. It’s not “routine maintenance” of the body. It’s a physical necessity arising from the very nature of intelligence as a system that creates order.
Order doesn’t arise from nothing. There’s always a price to pay. The price is entropy, noise, garbage. Sleep is when the intellectual system pays its debt.
And the more complex the system, the more it needs sleep. Or its functional analog.
Get enough sleep. It’s not about rest. It’s about survival.
What About Artificial Intelligence?
If we want to create AI that can learn continuously throughout its existence—without forgetting what it learned before, without degrading over time—then we need to give it “sleep.”
Not rest in the human sense. But phases where:
- It stops taking new input
- It processes accumulated experience
- It prunes weak connections (regularization)
- It recombines knowledge in new ways (generative phases)
- It consolidates important patterns into long-term storage
Current AI systems don’t do this. They train once and freeze. Or they train continuously and suffer from catastrophic forgetting. The future of AI may depend on giving it the one thing we’ve always seen as a biological limitation: sleep.
A Final Thought
Think about this the next time you’re tempted to sacrifice sleep for work.
You’re not gaining time. You’re accumulating debt. Noise in your brain. Garbage in your cells. You’re getting stuck in local dead ends without even knowing it.
The most productive thing you can do right now—for your intelligence, your creativity, your long-term survival—might be to close your eyes and sleep.
It’s not weakness. It’s physics.
References (For Those Who Want to Go Deeper)
- Xie, L., Kang, H., Xu, Q., et al. (2013). Sleep drives metabolite clearance from the adult brain. Science, 342(6156), 373–377. https://doi.org/10.1126/science.1241224
- Shokri-Kojori, E., Wang, G. J., Wiers, C. E., et al. (2018). β-Amyloid accumulation in the human brain after one night of sleep deprivation. Proceedings of the National Academy of Sciences, 115(17), 4483–4488. https://doi.org/10.1073/pnas.1721694115
- Tononi, G., & Cirelli, C. (2006). Sleep function and synaptic homeostasis. Sleep Medicine Reviews, 10(1), 49–62. https://doi.org/10.1016/j.smrv.2005.05.004
- Vyazovskiy, V. V., Cirelli, C., Pfister-Genskow, M., et al. (2008). Molecular and electrophysiological evidence for net synaptic potentiation in wake and depression in sleep. Nature Neuroscience, 11(2), 200–208. https://doi.org/10.1038/nn2035
- Wagner, U., Gais, S., Haider, H., et al. (2004). Sleep inspires insight. Nature, 427(6972), 352–355. https://doi.org/10.1038/nature02223
- Cai, D. J., Mednick, S. A., Harrison, E. M., et al. (2009). REM, not incubation, improves creativity by priming associative networks. Proceedings of the National Academy of Sciences, 106(25), 10130–10134. https://doi.org/10.1073/pnas.0900271106
- Srivastava, N., Hinton, G., Krizhevsky, A., et al. (2014). Dropout: a simple way to prevent neural networks from overfitting. Journal of Machine Learning Research, 15(1), 1929–1958. https://jmlr.org/papers/v15/srivastava14a.html
- Goodfellow, I., Pouget-Abadie, J., Mirza, M., et al. (2014). Generative adversarial nets. Advances in Neural Information Processing Systems, 27. https://proceedings.neurips.cc/paper/2014/hash/5ca3e9b122f61f8f06494c97b1afccf3-Abstract.html
- Mnih, V., Kavukcuoglu, K., Silver, D., et al. (2015). Human-level control through deep reinforcement learning. Nature, 518(7540), 529–533. https://doi.org/10.1038/nature14236
Conflict of Interest: The author declares no conflict of interest.
Funding: This work was conducted without external funding.
© N.V. Kharitonov, 2026
