What if Sleep is Whole-Body Evolution?

Sleep is no longer widely considered a period of rest, or torpor. Instead, hibernation serves this function. Hibernation, a very different process, actually generates a sleep deficit, which must be periodically paid down in the hibernating animal by “waking up” for sleeping. This demonstrates that sleep has a very different function. But what on earth, then, is sleeping for? I propose that sleep is a repair process that supports the open-ended evolution of living structures.

Sleep simply cannot be for the purpose of energy conservation. First off, it doesn’t conserve very much energy. In humans, sleeping uses 85–90% as much energy as waking life. Second, all known organisms need less sleep when fasted. Think about this. How could a process that is built for conserving energy be turned on less often when all energy is removed? This second point the reader can easily test for herself if she is brave enough to do a water-only fast for several days. During a prolonged fast, humans sleep far less, but we feel rested. This must mean that sleep is accomplishing something very different.

To imagine how this process might work, consider the following: as far as I know, all organs in the body experience atrophy if they are both unused, and alive, for a long period of time. With complete bed rest bones, muscles, even the heart gets gradually weaker, and shrink in size. Logically, this means that either atrophy happens sometimes, or it happens all the time. If it only happens sometimes, then atrophy is a special purpose process that must be turned on when the body senses disuse of a particular structure. If this is the case, my guess about sleep is wrong. But if atrophy is a process that happens all the time, then the reason why organs do not atrophy when used normally is that some other process countervails, and struggles against atrophy. The only way this could happen is that there must be a process (the sleep process) that distinguishes in every organ between the damages that are caused by use and the damages that are caused by atrophy. This process, which must be a bit different in every organ, simply reinforces structures that have been used during active life, and does nothing about the atrophied structures. Atrophy must be, then, a very stochastic form of decay, a random misalignment, that accumulates gradually, but continually, in an unused structure. Any complex molecule will eventually break down if held at a high temperature. This random decay lays the new groundwork for a slightly different set of connections to grow again if the structure begins to be used again later.

This process would tend to have no obvious long-term consequences in most organs, since if they are used, they are always used in much the same way. A muscle recovering from atrophy will work much like a muscle that never atrophied at all. But the brain and the rest of the nervous system is not an organ like this. In the brain, the devil is in the details. Logically, the relative strength of every single connection in the brain should have some slight form of behavioral consequence, so a process that sorts between used and unused connections, strengthening the ones which are used, will have tremendous neural consequences over a long period of time. This would explain results that show that we must sleep on new knowledge up to three nights after learning it before it is really conserved in the memory. This must be the “secret” of intelligence.

Let’s walk through the details here. Let’s say I once knew a person who lived on my street, in the red house on the corner, named Jenny, but years ago she moved away. The pathways between my friend Jenny’s neural correlates (these are the parts of the brain that “light up” when I think of Jenny) and those of the red house have since been weakened by atrophy. In the years since, by thinking about other things, I have associated other neural correlates with that red house location, those of other families, those for sale and rent, those for empty houses, and so forth. Of Jenny I have given little thought, but she is still in there somewhere, for example linked to an expectation for her approximate age. But suddenly one day I pass by this corner and much to my surprise Jenny pops out of the door to the red house. She explains that she is back in town to earn a doctorate and she decided to move into her old place again. Suddenly the old pathway lights up again, causing a thread of trauma in much the same old place.

The brain is full of tenuous, implicit connections. Like a spider mending a gash in her web, it has to “pull” on all the associated threads to strengthen a torn area. That night, the weakened connection from Jenny to the red house is repaired and strengthened, but in a strange way. I dream of Jenny in merchant sailor’s garb. She is leaning over the bow of a tall ship, its mooring cut away, drifting on the high seas. She shouts out to me, in a romantic voice “Four score and seven years ago!” In the morning, I can remember that Jenny is in the red house, but the meaning of this event has changed. Now, since the healed connections to the red house are more various and richer, it feels somehow different to think of Jenny there again. I realize the red house has been “adrift” in the marketplace. I realize Jenny has been in Lincoln, Nebraska, because the phrase she shouted came from the Gettysburg address. I realize that I like Jenny a little more than I knew. This is how implicit knowledge could be knit together by dreaming. Multiply this type of event times more than a billion, and you can begin to imagine how we can construct a world. This fresh interconnectivity is the richness of lived experience.

The nervous system is simply an organ that coordinates the functions of all the other organs, that unites them into one holistic process. If all the organs are following a similar dialectical process, a struggle between atrophy and trauma on the one hand, and sleep on the other, this explains why NREM sleep, which tends to repair the body, occurs first in the cycle, before REM sleep, which tends to repair the mind. Intelligence must be based on the complexity of neural connections and the plasticity of those connections. The ability to learn is simply the ability to change our behavior in the light of new experiences. It makes sense to suppose that behavioral change requires the participation of all the organs and systems of the body, and that they might all profitably be adjusted continually to the rhythms of life to form a coherent new whole each day. This could explain, at long last, why we sleep.

If this process if so basic, how can we have missed it with modern science? What I am describing is a holistic change in an organism, the details of which will not show up in a specific way under a microscope. Without knowing the position and velocity of all the particles in the body and how they work together, we cannot precisely measure either fatigue or atrophy, in its early stages. But we can certainly feel fatigue! We know that it is real. The methods of modern science look first for precise data, and that which does not produce data can often remain mysterious for a surprisingly long time.

…

Neo-Darwinism, the theory of the “selfish gene,” attributes all creativity in organisms to occasional useful mutations in DNA. So if we want to study creativity in a laboratory, we know for sure we can find a creative process by looking at DNA damage. Mutations could arise spontaneously, through thermal breakdown of DNA. But there must also be a place in evolutionary theory for inheritance of traumatic injuries to the somatic germline cell DNA, as well as injuries to the germ cell itself, since the entire germline cell replicates itself through self-templating. Only by including these further types of inheritance mechanism could we apply the theory also to unicellular life, which is the majority of life on the planet. These mechanisms are also required to explain the existence of structures like the mitochondria in eukaryotes, which contain genetic material separate from the nuclear DNA. This means that there are at least two possible sources of creativity in evolving organisms, both genetic mutations and certain very specific sorts of trauma in germline cells.

The mutation rate of DNA varies dramatically across different parts of the genome. This fact suggests that the mutation rate must itself be subject to evolution. In other words, certain locations on the chromosome are more protected from the sources of mutation and trauma than other locations, therefore they change more slowly. Other locations are more exposed to those sources of damage, therefore they change more rapidly. This means in addition to the plasticity of organismic phenotypes being under evolutionary control, the plasticity of different parts of the genome are under evolutionary control as well.

If a higher sensitivity to trauma and higher mutation rate has some adaptive value for parts of the genome, why should it not have some adaptive value for every living structure? Every structure could be impacted by its niche and, in reforming itself according to a repair algorithm, provide a better fit to its conditions. Parts of the structure which survived the rigors of active life could be strengthened, while parts which were broken down might be reformed in a new configuration. This process of fitness is familiar in the musculoskeletal systems. But there is no reason in principle why sleep could not afford an opportunity for a similar open-ended evolution of new network configurations in every system of the body, including the nervous system.

The philosopher of science Karl Popper viewed the building of knowledge as a feedback between conjectures and refutations. Creative conjectures unify a new theory which sets expectations, and refutations erode or fracture it by logical or empirical disproof. This theory maps well onto this epistemology. In my view, any set of biophysical configurations is a conjecture, a guess about the world, which can be refuted (broken) by events and rebuilt into a new conjecture by the process of sleep. The body is a set of physical expectations, and any surprise to those expectations is a form of trauma. Recovering from trauma leaves the organism more knowledgeable. This mechanism could at last answer the question that Popper’s epistemology left open…where do creative conjectures come from?

Intelligence, the behavioral plasticity of whole organisms, appears to have arisen independently in distantly separated lineages in the tree of life. Where it has arisen, it is always accompanied by sleep, and at higher levels by REM sleep or dreaming. Organisms as distant from humans as octopuses, zebrafish, hermit crabs, and even fruit flies show behavioral complexity and patterns of neural activity consistent with some form of sleep, while some especially intelligent members of distant lineages, including mollusks like cuttlefish and octopuses, lizards like the bearded dragon and the tegu, and many families of birds, exhibit both non-REM and REM sleep. Convergent evolution of bi-phasic sleep and intelligence together in these distant lineages suggest that there is something fundamental about bi-phasic sleep that underlies intelligence, and that it is essential.

…

The whole-body evolution mechanism I suggested above suggests that the ultimate determinants of intelligence in species, and in individuals, are not only the underlying physiology, but the overlying characteristics of the niche. Given a process of open-ended evolution in structural networks, organisms would not need to express the same repair algorithms in order to arrive at the same functional characteristics in intelligence, they would only need to be sensitive to traumatic impacts from the same environmental stimuli. The only algorithmic necessity is that the sleep process must distinguish between damage caused by use and damage caused by atrophy. The actual character of the knowledge comes from the environment we are exposed to. Of course the shape and the nature an organism’s body give it access to stimulation from certain stimuli, the stimuli in its niche. A certain type of body can know things that another type of body cannot. This idea was explored by Thomas Nagel in his famous 1974 paper “What Is it Like to be a Bat?”

Consider a lecture hall for a moment. Inside, there are dozens of students, listening to the professor drone away. The words of the lecture are barely audible inside the students minds. One is thinking of his upcoming swim meet, one is thinking about her Instagram profile, another is wishing he knew the girl in the second row better. Each one of them has a set of associations with this particular moment in time that will necessarily be very distinct from the other students. Not only their brains, but their bodies are quite different. One is tall and thin, one is obese but very rugged, and so forth. It simply can’t be true that an identical algorithm, an “intelligence algorithm” is running under the surface. The reason they all learn something from the lecture and can pass a test later is that their overall set of expectations has been overturned by similar knowledge, which because they understand in their body the behavioral and emotional consequences of words, will later be externizable as an active reply to the test questions. That knowledge “lives” in a completely different configuration of atoms in the girl with curly hair than it does in the boy with dreadlocks. What they have in common when they learn in common is holistic behavioral patterns, not the details under their skulls.

If sleep is a repair process for trauma, including neural trauma, then thinking is itself a form of trauma. A thought occurring to us would leave a fingerprint in the neural structure, a thread of damage that would later be strengthened and repaired by REM sleep. That a process like this should sometimes feel like an experience, a dream experience, should be no great surprise. If thinking feels like an experience, and thinking is leaving threads of damage, then the process of tracing those threads while repairing them could be expected to feel like something. The REM phase of sleep, in particular, must support the production of what we experience as “counterfactuals.” Of course there is no real boundary between hallucination and perception, but dreams, like imagination, are recursive self-generated neural experiences. Greater sensitivity and greater plasticity go hand in hand. The recovery might, like the hands of the nurse dressing a wound, be painful.

In fact, this theory reframes pain itself. Pain, like any experience,is just an experience of the intensity of a refuted expectation. This would explain, among other things, the placebo effect. A set of mental and physical expectations is key to any type of cognition at all, and wherever they are overturned, attention will tend to focus. When that expectation (the integrity of the surface of our skin, for example) is particularly deep and fundamental to our homeostasis, we can expect our body to generate an attention-grabbing phenomena like pain.

This theory sees the whole body as necessary to engage in cognition. More specifically, it sees the connections between non-neural networks in the body as the “window onto the world” that the brain reads to generate its understanding. This is why these bodily connections must be prepared ahead of the neural connections by putting NREM sleep before REM sleep. It also sees neural connections as no different in kind from other networks of the body, only in the speed of their mutation rate.

This theory also makes sense of the divided architecture of the nervous systems of intelligent animals. One hemisphere of the brain (in humans, the right) is configured to be sensitive to slight alterations in the expected patterns of a broad array of inputs, while the other (the left) is configured to be sensitive and to focus strongly on the single most noteworthy stimulus in the attention field. This corresponds to the evolutionary need for nearly all organisms to use different attention thresholds for the activities of predation/eating versus social/situational awareness, yet use the same streams of whole-body sensation for both. This bifurcation must be absolutely essential to “get a grip” on the world. It has convergently evolved also in octopuses, which have eight “brains” in their tentacles to attend to details, and one in their main bodies to attend to the whole. This is entirely logical, since attending very vaguely to only to slight surprises in the overall flow of life (the holistic right hemisphere) would lead to obliviousness to salient details, like food or predators, while attending only to the most salient details in order to manipulate them intensively (the reductionistic left hemisphere) would lead to complete inattention to the overall context one is living in. These two forms of sensitivity are incompatible, there is no way for one brain to do both well. This must be why all known highly intelligent nervous systems are divided along this axis.

This view, if correct, opens a new window onto the difficulty of communication, both between people and between species. This theory sees cognition as the result of whole-body changes that result from a repeated physical engagement with the ecological niche. This would explain why it is so difficult to communicate knowledge from one organism to another; so much is “lost in translation.” In order for an organism to communicate with another one, they must share many of their biophysical experiences of the world in common. To get a message across, they must translate it into information or gesture, object, sound, or scent, and the recipient must retranslate it back into a new biophysical configuration. This would explain why the definition of intelligence has remained so elusive, and why the intelligence of other species continues to mystify us. Intelligence is better thought of as a property of the niche outside the organism rather than the algorithms underlying the structural networks of the organism.

…

If approximately correct, this theory could be applied to build truly generally intelligent machines. If intelligence is the result of repeated traumatic engagement with a niche interspersed with repair processes, and the underlying repair algorithm of the network is unimportant, then physical structures could be built with hardware and software that recreated this process of open-ended evolution. The machine would need to be able to engage physically with the world in the way an intelligent organism does, which means that its actions would all be reactions, designed to anticipate incoming data as best it can. All of its actions would be dedicated to reacting to surprises, and it will have motivations.The motivations come from desiring to anticipate the events around it better. These motivations are fundamental to epistemology. There will be no true embodied learning without them. For a thinking being, any unanticipated event it witnesses creates a drive, a drive to discover why.

This could be all be programmed through machine learning, which provides suitable technology to reconfigure the instruction base in the way that stochastic decay and sleep does in a living being. It will just require a program making the distinctions in configuration made above, between atrophy and use in the damage to connections. But this process has been deeply misunderstood by current approaches to artificial intelligence. Cognition is not simply a Bayesian prediction algorithm fed a certain stream of data. Critically, these configurations would have to comprise a set of expectations about the upcoming conditions that could be refuted (traumatically damaged) by unexpected physical events, and by the unexpected consequences of the machine’s own actions. The structure is not encoding statistical likelihoods. It is encoding causal diagrams. Like a child, it would need to experiment.

The robotic body need not be humanoid (it could be shaped like an octopus,) but it would need the equivalent of arms, fingers, eyes, ears, and voice in order to engage with the human ecological niche. It is one thing to have complexity and plasticity in neural connections, but as our disdain for octopuses shows, that is not enough for most people to recognize its cleverness and be amazed. For the general observer, perception of its intelligence would reside in its sensitivity to the aspects of reality which we are likewise sensitive to… in our sensitivity to the ranges of experience available inside our niche. It would also require a divided neural processing apparatus which could attend, one side to the holistic flow of life and the other side to the manipulation of specific details. These two manners of processing the same dataset will both be necessary to “get a grip” on a world in an intelligent way.

This robotic body should not need to undergo atrophy itself. Only the neural connections will require atrophy, in order that they might produce intelligence. In fact, atrophy in the robot would be a nuisance to be avoided. The robot could be made out of interchangeable parts, like all robots. What is significant is that the neural connections follow the procedure of atrophy/trauma and sleep. It could simply use an unchanging robotic body for its “window onto the world” and thus learn more reliably and quickly than a human brain using a malleable human body. What matters, I assume, is the plasticity of the interpretive device, not the plasticity of the medium.

Most surprisingly, this theory anticipates that organisms, all organisms, are already epistemological devices. What differs between them is their surroundings, both now and in the evolutionary past. If this guess is true, then it is fair to expect that all organisms are going about the business of repairing the refutations they encounter to their sets of biophysical expectations about the world inside their niche. They are living curious, sensitive lives, exactly as we humans are, but many of them are locked into niches that are so different from our own that we can barely recognize their sentience, let alone their intelligence. This insight has implications for understanding the nature of morality that I will explore in another paper.

Copyright April 2021 by Charlie Munford