A guess about sleep

I have always found the explanations for sleep and dreaming to be insufficient: the notion that sleep is for saving energy has not stood up to empirical tests…it turns out animals use nearly as much energy while sleeping as while awake, and hibernation, which is an energy-saving process, must be interrupted periodically in order to sleep. Another explanation is that there are “two niches,” one in the day and one at night, and sleeping is an adaptation to avoid the less optimal niche. I find this incoherent…an animal killed during the night is still dead, food not eaten is still missed, and mating opportunities are still lost. Why would mammals like dolphins, who must continually surface to breathe, still sleep with half their brains at a time? And why would animals far from us on the tree of life, intelligent mollusks like octopus and cuttlefish, show convergent evolution on the solution of sleep and REM dreaming? Convergent evolution always indicates that there is something about a niche that is common to another distant niche, like air for wings, water for fins, or light for eyes.

It makes much more sense to me to think of sleep and dreaming as a repair cycle, a comprehensive mode for rebuilding our bodies. But why should it be so essential? Every system of the body fails rapidly without sleep. Why should organisms have evolved to break down so quickly that they need to spend eight hours per day helpless, unconscious and paralyzed in a maintenance mode? I think that the answer is that there is an enormous adaptive value to a certain sort of plasticity, and sleep is the price we pay for that plasticity. What the cephalopod niche has in common with the human niche is that it contains myriad complex causal patterns, and these require the behavioral plasticity we call intelligence.

Neo-Darwinism posits that the bodies of organisms are survival machines for their replicators,--- genes. In this view the organism is caused by its replicators, in order to carry them into the next generation. The sources of change in this view are blind variations in the genome, coupled with selective retention of genes in the population driven by the environment of the niche. There is no algorithm that determines which mutations arise, nor algorithm that determines which gene variants survive and spread. These are the result of open-ended evolution. But there is one odd fact about this: mutation rates vary both across species and across the genome. If the mutation rate is different in different parts of the genome, that suggests that the mutation rate in each part of the genome is ultimately controlled by the niche of the organism during natural selection. In other words, the mutation rate itself is an evolved characteristic, both locally and species-wide. This means that the adaptive value of certain parts of the genome can influence the rate at which that part of the genome changes.

How can the rate of mutation vary? It must be that either some parts of the genome are more shielded from damage than others, or that they are more likely to be repaired than others. In either case, there is a dialectic between damage and repair that results in a certain mutation rate for each region, given its adaptive function. This rate may change over time: it is variable, and driven by open-ended evolution.

Now I have a speculation to make. Suppose that the rest of the body and brain mutate in just the same way as the genome. I call this bodily mutation “blundersplat,” because I think it leads to trial and error. Suppose this dialectic between damage and repair results in a certain mutation rate, a certain plasticity, not just in the genome but in every tissue of the body. This would mean that just like the genome, each tissue has its own vulnerability to trauma of various sorts, and its own repair cycle. How quickly each organic structure changes over time would depend on its adaptive value in the environment of the niche. Some organ systems, like the skeleton, change very little in response to trauma on a daily basis, though they certainly do sustain damage and they do heal from that damage. Other systems, like the nervous system, change rapidly on the order of milliseconds. Blundersplat is what we experience as physical fatigue, and mental forgetfulness.

Each part of the body, then, would be under the same control of the niche environment as the genome. Of course each cell is using the genome to complete its functions, but each tissue has its own plasticity, its own dialectic between damage and repair. The repair algorithm expressed by each tissue may involve the use of genes, but the trauma comes from the niche, and the relative adaptive value of plasticity in each tissue comes from its usefulness in the niche. In this case, ultimately it is not important what the specifics of those repair instructions are, they evolve from a feedback loop with the environment of the niche, just like the genome.

Selective retention of genes in the genome occurs through replication and natural selection. How can selective retention occur in a tissue? Selective retention of mutations in a tissue can occur when a healing algorithm, any healing algorithm, gets applied over and over to a tissue that is subject to regular trauma of certain sorts. If a tissue expresses a healing algorithm, it evolves into a more adaptive structure in response to the cycles of trauma and repair it undergoes. Natural selection still plays a critical role from generation to generation, but the environment is also selecting certain structural components over others. The strongest, most useful parts of a structure are reinforced by a healing algorithm, the weakest parts get destroyed by trauma and reformed in another configuration. The healing algorithm may be mostly the result of genetic natural selection, but the structure that results from the cycles of trauma and repair is unique to each organism, because niches differ slightly from one individual to another.

As the Oxford physiologist Denis Noble points out in his book Dance to The Tune of Life, one can determine a causal relationship between the organism and its niche by asking the question: Does the organism require a niche? Yes. Does the niche require an organism? No. Therefore the niche is the cause of the organism, not vice versa. Likewise, DNA freed from the cell (such as a virus) requires a host, but the host does not require a virus. This shows how, even at the cellular level, organisms cause genes to exist, not vice versa. All this view requires is just that you think of organisms as products of their environment, not the other way around. Noble calls this “biological relativity.”

If this guess is correct, then the question of AGI is really asking: what is the healing algorithm of the human nervous system? But since this healing algorithm itself is under selection based on the persistence of the structure it repairs, it will not be necessary to know all the particulars of this algorithm to build an intelligent creative machine. Instead, it will only be necessary to build a machine that can repair the structure of its nervous system from trauma using any healing algorithm, and be sure that the trauma comes from the same parts of reality that form the human niche, otherwise we will not be able to communicate with it. It would not require a genome, and its body could be made from interchangeable parts of artificial hardware, but its mental system will have to generate structures just as ours does that can be broken by physical experiments of the same sort we conduct with our own bodies. It does not need to be “humanoid,” (it could be shaped like an octopus) but it does need to be able to manipulate objects, move itself, and hear, see, and speak in the ranges of human perception.

Intelligence is different from exchanging information. Information is the way we humans store and communicate some forms of knowledge; most species store and communicate only genetic information. Information is a message “selected from a set of possible messages.” Suppose you encountered a deaf chimpanzee who could not read, yet could play tennis, and adhered to all the rules of the game which it had picked up through trial and error. Would this animal seem intelligent? Surely, yet it could not communicate with you through information. It would seem intelligent in the way an octopus seems intelligent…the solitary octopus, though highly plastic, has not evolved in a niche where external information can be traded between individuals, and does not have the body style that would allow us to train it readily. In order to help build external knowledge with alphabetical and digital information like we do, an AGI will need to share our human niche.

Trauma is the result of the interactions of the organism with its niche. We can think of this as experiments. The data scientist Judea Pearl writes about the “ladder of causation.” An organism can only get so far in explaining the world through observation, because this provides only correlations, not causal models. This is where AI research is today. In order to express a causal model, in order to get an explanation, an organism must conduct experiments with reality inside its niche, and then have the results imprinted on its body somehow. We can think of the nervous system as a structure that expresses an if-then logic to protect the body of an organism from destruction. The configuration of this structure in any moment is a set of imprecise expectations that are the result of all the preceding cycles of trauma and repair…all the background knowledge of the organism. Impacts to this system that differ markedly from expected impacts constitute perception. These are the problems the organism has evolved to solve through its nervous reactions. Problem-solving, and creative thought, is just the evolution of these fast-mutating structures through cycles of damage and repair.

This model of organisms is thoroughly Popperian: the configuration of a living body in any moment comprises a set of conjectures. Perception is our way of being damaged by the parts of reality that refute these conjectures in any given moment. In this view, the only perceptible parts of reality are in fact those that depart from what we have thoroughly expected…this explains why so much of our experience is unnoticed and immediately forgotten, and also why the placebo effect works so reliably. Only the unexpected is noticed and retained for processing.

Dreaming, in this view, is the internal rapid production of counterfactuals…these are the next step on Pearl’s ladder leading to causal knowledge. Our nervous system must uncouple itself from the inputs of conscious perception as well as its locomotor outputs in order to heal its internal structure. The way it heals is by expressing counterfactuals…the internal chains of dominoes expressed in the weights and connections of firing neurons must be tipped independent of inputs and outputs in order to strengthen the connections that have survived the trauma of conscious perception during the daytime. These counterfactuals are the brain’s way of selectively retaining and reinforcing the mutations that best anticipated the refutations encountered during each cycle of waking life. There must be no other way of doing this…that’s why even cephalopods dream in REM sleep.

This niche’s eye view of evolution is a radical revision of Neo-Darwinism…but there are no new superstitions being introduced here, and no theology. This upside-down theory of evolution does not deny any of the facts underlying Neo-Darwinism, only their causal model. It just requires viewing the phenotype, as well as the genotype, as a structure that picks up knowledge through open-ended evolution in a niche. Niches are already irremovable parts of our best scientific explanation of life; this theory just gives them a new and better role to play. If this guess about sleep and dreaming is correct, then it is really our human niche that is creative, not our human brains. This explanation of intelligence is much harder to vary.

Copyright 2020 by Charles Sims Munford