# The Function of Dream Sleep

> An annotated edition of this paper on Fermat's Library, where scientists and readers explain its ideas, equations and history in the margins. Annotated edition: https://fermatslibrary.com/s/the-function-of-dream-sleep

- **Authors:** Francis Crick & Graeme Mitchison
- **Published:** 1983
- **Topic:** Biology
- **Original:** https://s3-us-west-1.amazonaws.com/fermatslibrary/uploads%2F1541871600984-6x1558byixt-657bfbfef6fe92f3555944efb52464ba%2Ffunction_of_dream_sleep.pdf
- **Pages:** 4
- **Annotations:** 12 comments in 12 threads
- **Featured:** Fermat's Library Journal Club
- **Annotated edition:** https://fermatslibrary.com/s/the-function-of-dream-sleep

## Annotations

### Annotation 1 (page 1)

Beside the passage: “The function of dream sleep”

**Annotation** by Micael Oliveira (2019-04-06):

> ### TL;DR
>
> - REM sleep is predominantly found in viviparous mammals and birds
> - REM sleep seems to be associated with homeothermy
> - The authors propose that the role of REM sleep is to suppress "parasitic" cortex modes that are due to brain growth or to experience
> - REM sleep triggers a "reverse learning" mechanism that suppresses incorrect associations formed by neural network learning mechanisms in the cerebral cortex, with the consequence that the last thing that you should do with respect to your dreams is to remember them.
>
>
>
> > ***“Exactly how sleep spindles accomplish this clever memory trick remains unclear. What we have at least discovered is a rather telling pattern of looping activity in the brain that coincides with these speedy sleep spindles. The activity circles between the memory storage site (the hippocampus) and those regions that program the decision of intentionality (in the frontal lobe), such as “This is important” or “This is irrelevant.” The recursive cycle of activity between these two areas (memory and intentionality), which happens ten to fifteen times per second during the spindles, may help explain NREM sleep’s discerning memory influence. ”***  -  Matthew Walker. “Why We Sleep: Unlocking the Power of Sleep and Dreams.” (published in 2017)

### Annotation 2 (page 1)

Beside the passage: “remove certain undesirable modes of interaction in networks of cells in the cerebral cortex. We”

**Annotation** by Micael Oliveira (2019-04-07):

> Additional interesting resources related to sleep cycles and the roles of the different phases of sleep can be found here:
>
> - [Matthew Walker, Ph.D., on sleep – Part I of III: Dangers of poor sleep, Alzheimer’s risk, mental health, memory consolidation, and more](https://peterattiamd.com/matthewwalker1/)
>
> - [Dr. Matthew Walker on Sleep for Enhancing Learning, Creativity, Immunity, and Glymphatic System
> ](https://www.foundmyfitness.com/episodes/matthew-walker)

### Annotation 3 (page 1)

Beside the passage: “Francis Crick* & Graemi Mitchison*”

**Annotation** by Micael Oliveira (2019-04-06):

> Francis Crick is best known for his work with James Watson which led to the identification of the structure of DNA. This discovery earned Crick, Watson and Wilkins the Nobel Prize in Physiology or Medicine in 1962.
>
> Graeme Mitchinson was a mathematician, scientist and pianist with a wide range of interests. He worked with Francis Crick on the meaning of dreams, with Horace Barlow on the neuroscience of vision, with a group at Oxford on quantum computation and on plant development at the Sainsbury Laboratory.
>
> Learn more about the authors here:
> - [Francis Crick](https://www.crick.ac.uk/about-us/our-history/about-dr-francis-crick)
> - [Graeme Mitchinson](https://www2.mrc-lmb.cam.ac.uk/graeme-mitchison-1944-2018-2/)

### Annotation 4 (page 1)

Beside the passage: “We first describe our ideas about the cortex followed by a brief account of neural networks. Next we outline what is known about REM sleep. (For general accounts, see refs 1,2.) We then describe our postulated mechanism and how it might be tested. Finally.we discuss various implications of our ideas. .”

**Annotation** by Micael Oliveira (2019-04-07):

> In "The Interpretation of Dreams" published in 1899 Freud situated the dream  within the brain of an individual. This seems obvious today but at the time it was revolutionary. Freud was right about that but wrong in other aspects. He believed that dreams came from unconscious wishes that had not been fulfilled. One of the major problems with Freud's theory was the lack of any clear predictions. One could not design an experiment that would test it in order to help support or falsify it. Science could never prove him wrong, which is why Freud continues to cast a long shadow on dream research to this day.
>
> More recently in the early 2000s scientists were able to use MRIs to reconstruct three-dimensional visualizations of brain activity during REM sleep. Among other breakthroughs these results undermined the postulates of Sigmund Freud and his nonscientific theory of dreams as wish fulfillment.
>
> ![traumdeutung](https://i.imgur.com/JaVM5LE.jpg)

### Annotation 5 (page 1)

Beside the passage: “The cortex The cortex consists of two separate sheets of neural tissue, one on each side of the head. The neocortex, which has a characteristic layered structure, is found only in mammals (see ref. 3 for recent survey), although a somewhat analogous structure, the wulst, is found in birds. If allowance is made for …”

**Annotation** by Micael Oliveira (2019-04-06):

> ## Sleep Cycle:
>
> During a night of sleep we follow different 90 minute cycles of Non-REM (NREM) and REM sleep. There are 4 stages Non-REM (NREM) sleep stages followed by REM sleep.
>
> **NREM Sleep Stages:**
>
> - **Stage 1:** is the lightest stage of NREM sleep. It is characterized by slow eye movements muscle relaxation and brain wave activity begins to slow from that of awake state.
>
> - **Stage 2:** is the first actual stage of defined NREM sleep. During this stage body temperature begins to decrease and heart rate begins to slow. Awakenings or arousals do not occur as easily as in Stage 1 sleep.
>
> - **Stages 3 and 4:** is known as deep NREM sleep. Stage 3 is most restorative stage of sleep.  Awakenings are rare and it is often difficult to awaken someone during Stage 3 sleep.
>
> **REM Sleep: ** also known as rapid eye movement or dream sleep. During this stage eye movements are rapid, the body does not move and brain waves are more active than in Stages 2  and 3 of sleep. The most vivid dreams happen in this stage.
>
> ![sleep_cycle](https://i.imgur.com/EUE4NJp.png)
> *Figure: vertical axis displays the various sleep stages as they occur over different sleep cycles (horizontal axis) during a night of sleep.*

### Annotation 6 (page 1)

Beside the passage: “Much exploratory theoretical work has been done on such networks of cells (for an introduction see refs 6-8). In these models, information is stored in the strengths of the many synapses and sometimes in the firing thresholds of cells as well. Although the exact behaviour naturally depends on the details of the …”

**Annotation** by Micael Oliveira (2019-04-06):

> The neocortex is part of the brain of mammals that is involved in higher functions such as **sensory perception, generation of motor commands, spatial reasoning, conscious thought, and language.** In humans it accounts for about **76% of the brain's volume.**
>
> ![neocortex](https://i.imgur.com/KiCqoYr.jpg)

### Annotation 7 (page 2)

Beside the passage: “REM periods may of the muscles of the sleeping animal, especially its head and neck muscles, are more relaxed than in non-REM sleep. Its cortex, as judged by the electroencephalogram (EEG) and by the rapid movement of the eyes beneath closed lids, appears to be very active and in a state similar to the waking state. …”

**Annotation** by Micael Oliveira (2019-04-06):

> Thanks to MRIs we are able to highlight the different parts of the brain that are most active during REM sleep:
>
> 1. **the visuospatial regions** which enable complex visual perception.
>
> 2. **the motor cortex** which instigates movement.
>
> 3. **the hippocampus and surrounding regions** which support your autobiographical memory.
>
> 4. **the deep emotional centers of the brain, the amygdala and the cingulate cortex** which help generate and process emotions. These emotional regions of the brain are up to 30 percent more active in REM sleep compared to when we are awake.

### Annotation 8 (page 2)

Beside the passage: “The postulated mechanism We need a mechanism which will tune the cortical system, in the sense of removing parasitic modes which arise after the system has been disturbed either by growth of the brain (when new connections are constantly being made) or by the modifications produced by experience. The mechanism we …”

**Annotation** by Micael Oliveira (2019-04-06):

> ***Newborn babies spend about 16 hours each day sleeping out of which 8 hours are spent in REM sleep.*** Half of their total time asleep is spent in REM sleep and this proportion is even bigger in the third trimester before birth. Interestingly a large amount of REM sleep before and after birth is also found in other mammals.
>
> Learn more here: [Stanford Children's Hospital - Newborn-Sleep Patterns](https://www.stanfordchildrens.org/en/topic/default?id=newborn-sleep-patterns-90-P02632&sa=U&ei=58e3VM6CIIr5yATNnIGwAg&ved=0CG8QFjAT&usg=AFQjCNFvyKhlh5_8yFZvCBirEv-fTY56pQ)

### Annotation 9 (page 2)

Beside the passage: “After this paper had been initially submitted for publication, we learnt from Dr John Hopfield that he and his colleagues had independently arrived at the idea of reverse learning, though not in connection with dreams. In a parallel communication’* they have shown that the behaviout of their very idealized neural net …”

**Annotation** by Micael Oliveira (2019-04-06):

> > ***Put more loosely, we suggest that in REM sleep we unlearn our unconscious dreams. “We dream dream in order to forget.”***

### Annotation 10 (page 3)

Beside the passage: “The effects of REM sleep deprivation are harder to explain. It is well established that REM deprivation often produces a rebound - more REM sleep than usual occurs when the subject is eventually allowed to sleep without interruption. We would have expected that REM deprivation, if severe enough, might cause …”

**Annotation** by Micael Oliveira (2019-04-06):

> According to Crick and Mitchinson's hypothesis the cortex cannot cope with the vast amounts of information received durning the day without developing "parasitic" modes. These "parasitic" modes disrupt the efficient organization of memory. They propose that REM sleep wipes out these unwanted connections in cortical networks.

### Annotation 11 (page 3)

Beside the passage: “Testing the theory As far as we can tell, our theory is broadly compatible with a large amount of experimental data. Starting from a plausible hypothesis about cortical function, it explains in an effortless way both the need for REM sleep in adult life and the large amount of it during the development of the brain. …”

**Annotation** by Micael Oliveira (2019-04-06):

> > ***We normally become conscious of our dreams only if we wake up while dreaming is in progress. If we then pay attention to our dream, some of its content can be maintained in very short-term memory and may eventually be transferred to longer-term memory as the transfer mechanism becomes activated. Otherwise our dream fades. Thus we can speak of forgetting our dreams, meaning that we know that we had a dream, but are somewhat uncertain of its content.***

### Annotation 12 (page 4)

Beside the passage: “from neural modelling. This appr6ach hasits limitations, sin&it is difficult .to produce realistic models and even more difficult to simulate them effectively, especially if the. hypothetical neural nets approach a realistic size, when the computational time becomes prohibitively long. However, such theoretical …”

**Annotation** by Micael Oliveira (2019-04-06):

> Crick and Mitchinson's hypothesis was put to the test in 2009. That study found that contrary to their prediction, it was not REM sleep but rather NREM sleep, and especially the very quickest of the sleep spindles that helped bend apart the curves of remembering and forgetting.
>
> The conclusions of that study:
> > [...] we demonstrate that sleep, relative to time awake, can selectively enhance recall for words previously cued for remembering, without such facilitation of items instructed to be for gotten. Moreover, the efficiency of sleep in differentially responding to cued remembering and forgetting correlated with fast sleep spindles over left parietal cortex. Furthermore, source analysis associated with these spindles identified a network loop of current density between parietal, medial-temporal, and prefrontal memory networks that may support selective offline memory processing. More generally, the concept of sleep differentially modulating memory based on prior cue instructions may have implications clinically, where the capacity for selective retention of certain information, while forgoing maintenance of unwanted experiences, represents a desirable target.
>
> You can find more information here: [The Role of Sleep in Directed Forgetting and Remembering of Human Memories](https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3183424/)

## Full text

Extracted from the PDF. [Annotation N] marks where each annotation above sits.

#### Page 1

p&mJRJz VOL.30 I4 JULY 1983

COMMENTARY

III

The [Annotation 1] function of dream sleep

Francis [Annotation 3] Crick* & Graemi Mitchison*

We propose that the function of dream sleep (more proper& rapid-&ye movement or REM sleep) LT to

remove certain undesirable modes of interaction in networks of cells [Annotation 2] in the cerebral cortex. We

postulate that this is done in REMsleep by a revem learning mechanism (see also p. 158), so that the

trace in the brain of the unconscipus dream is weakened, rather than strengthened, by the dream.

MANKIND has always been fascinated by dreams. As might be expected, there have heen many attempts to assign a purpose Or significance to them. Although we dream for one or two hours every night, we do not remember most of our dreams. Earlier thinkers, such as Freud, did not know this. Modern theories (not reviewed here in detail) have usually proposed that sleep and dreams save energy or have various restorative functions, either to replenish the brain biochemically in some way, or to reclassify or reorder the information stored in it.

Sleep is of several kinds. Dream sleep, or rapid eye movement (REM) sleep, is predominantly found in viviparous mammals and birds. It seems to be associated with homeothermy (a constant internal temperature) and the possession of an appreciable neocortex or its equivalent. It is not unimportant because of the appreciable amount of time we spend in this Peculiar state.

We propose here a new explanation for the function of REM sleep. The basis of our theory is the assumption that in viviparous mammals the cortical system (the cerebral cortex and some of its’ associated subcortical structures) can be regarded as a network of interconnected Cells which can support a great variety of modes of mutual excitation. Such a system is likely to be subject to unwanted or ‘Parasitic’ modes of behaviour, which arise as it is disturbed either by the growth of the brain or by the modifications produced by experience. We propose that such modes are detected and suppressed by a special mechanism which operates during REM sleep and has the character of an active process which is, loosely speaking, the opposite of learning. We call this ‘reverse learning* or ‘unlearning’. This mechanism, which is not the same as normal forgetting,

*TheSalk Institute 10010NorthTorrey PinesRoad, La JOlkX California 92037. USA. Present address (GM.): MRC Laboratory of Molecular Biology and the KmWh Craik Laboratory, Cambridge CBZ 3EG. UK.

is explained in more detail below. Without it we believe that the mammalian cortex could not perform so well.

We first describe our ideas about the cortex followed by a brief account of neural networks. Next we outline what is known about REM sleep. (For general accounts, [Annotation 4] see refs 1,2.) We then describe our postulated mechanism and how it might be tested. Finally.we discuss various implications of our ideas. .

The cortex The cortex consists of two separate sheets of neural tissue, one on each side of the head. The neocortex, which has a [Annotation 5] characteristic layered structure, is found only in mammals (see ref. 3 for recent survey), although a somewhat analogous structure, the wulst, is found in birds. If allowance is made for body weight, it is larger in primates than in most other mammals and larger in man than in other primates. It makes up a substantial fraction of the human brain.

Different areas of the cortex perform different functions, some being mainly associated with vision, touch and so on, while others appear to process more complex information not associated with a single sensory mode. The exact function of the neocortex is unknown but it appears to be closely associated with higher mental activities. It seems likely that it has evolved to perform in a rather special way.

In examining the neuroanatomy of the neocortex one is struck by the very large number of axon collaterals (this is not true, for instance, of the thalamus). In any area of the cortex the great majority of synapses come from axons originating locally and running within it. There is also evidence that the majority of the synapses in the cortex are excitatory in their action. This suggests a capacity for self-excitatory modes of behaviour in the cortex. And indeed, in various conditions, such as epilepsy, migraine and certain kinds of drug-induced hallucination*, parts of the cortex appear

to go into large-amplitude instabilities5 .

Neuronal networks Now, if one asks what functions such richly interconnected assemblies of cells could serve, one attractive possibility is that they could store associations6-R. To see this, suppose an ‘event’ is represented by the activity of a subset of cells in a cell assembly. If all the cells involved in that event form mutual synapses, then when part of that event is encountered again these synapses can cause the regeneration of the activity in the entire subset.

Much exploratory theoretical work has been done on such networks of cells (for an introduction see refs 6-8). In these models, information is stored in the strengths of the many synapses and sometimes in the firing thresholds of cells as well. Although the exact behaviour naturally depends on the details of the particular model, certain general properties can emerge even from relatively simple models. The associations which are stored are not assigned specific locations for each item, as in a digital computer. Instead the information is: (1) Distributed: this implies that a particular piece of information is distributed over very many synapses. (2) Robust: this implies that the information will not be totally [Annotation 6] lost if a few synapses are added or removed. (3) Superimposed: this implies that one synapse is involved in storing several distinct pieces of information.

A properly designed net can be trained (meaning that the strengths of the synapses can be adjusted) so that given an input (a pattern of axonal firings) it can produce the appropriate output (another pattern of axonal firings). It is found that certain general properties will often emerge. (1) Completion: given only part of the input (as a clue) it can produce fairly exactly the whole of the appropriate output (examples aregiven in ref. 7). In computer jargon, the memory is ‘content addressable’. (2) Classification: given an input which is related toseveral of its associations, it may

#### Page 2

112

produce an output which combines many of the common features of its normal outputs.

A major difficulty with all nets of this general type is that they become overloaded if an attempt is made to store simuhaneously too many. different patterns or associations of patterns, or if the stored patterns have too large an overlap. This is because of the superimposed nature of the storage. How the net will behave when overloaded depends on the exact structure of the net, but certain patterns of behaviour are likely to emerge: (1) The net may produce many far-fetched or bizarre associations (‘fantasy’). (2) The net may tend to produce the same state, or one of a small set of states, whatever the input (‘obsession’). (3) Certain kinds of nets, particularly those which feed back on themselves, may respond to inappropriate input signals which would normally evoke no response from the net.(‘hallucination’).

It is against this background of rather tentative and idealized theory that our proposals must be judged.

If the cortex were hard-wired during embryogenesis to an exactly predetermined pattern of synaptic connections, the burden of eliminating parasitic modes in cortical nets would have to be undertaken by the genes alone. Although there is considerable evidence for specificity in the cortical wiring, it is likely that many of the details of the synaptic connections - their exact locations and their strength - are made in a semirandom manner and refined by experience. This is almost a necessity in an organism which is capable of learning very large amounts of novel information. Thus it seems likely that both during cortical growth (when we may say that certain broadly predetermined ‘associations’ are layed down), and also in facing the experiences of adult life, such parasitic modes. will be unavoidably generated.

How would one attempt to eliminate these modes? We suggest the following. The major inputs and outputs of the system should be turned off, so that the system is largely isolated. It should then be given SUCcessive ‘random’ activations, from internal sources, so that any incipient parasitic modes would be excited, especially if the general balance of excitation to inhibition had been temporarily tilted towards excitation. Some mechanism is then needed to make changes so that these potentially parasitic modes are damped down. Such a rough outline description immediately reminds one of REM sleep and the hallucinoid dreams associated with it.

REM sleep It was discovered in the 1950s that in mammals there are two main types of sleep. Periods of REM sleep (also called D sleep or paradoxical sleep) alternate with periods of non-REM sleep (also called S sleep, slow-wave sleep, or orthodox sleep) of which four stages of increasing depth of sleep are usually distinguished. During

COMMENTARY

NATURE VOL304 14 JUI Y I%3

REM periods may of the muscles of the sleeping animal, especially its head and neck muscles, are more relaxed than in [Annotation 7] non-REM sleep. Its cortex, as judged by the electroencephalogram (EEG) and by the rapid movement of the eyes beneath closed lids, appears to be very active and in a state similar to the waking state. On the other hand, the monoamine neurones in the brain stem, especially those in the locus coeruleus, raphe and peribranchial nuclei, reduce their firing rates in REM sleep to only a few per cent of the corresponding rate in the waking state9.

Another major difference between REM and non-REM sleep lies in the dreams associated with them. For most people the few dreams found in non-REM sleep tend to -have a rather thought-like character. During REM sleep, on the other hand, dreams occur more frequently and usually have a perceptual vividness and the illogical episodic character with which we are all familiar. A human adult usually spends a total of 1% to 2 hours each night in REM sleep, spread over several periods. The evidence suggests that most of the dreams during these REM periods do not reach normal consciousness, dreams being remembered only if the sleeper awakes whiledreaming. Even then the memory of a dream is usually very transient, fading quickly if no effort is made to remember it by rehearsing its content.

A most remarkable finding is that newborn humans may have as much as 8 hours of REM sleep per day lo. There is also evidence to suggest that in the womb, especially in the third trimester, REM sleep occurs even more frequently. This large amount of REM sleep before and after birth is also found in other mammals.

AI1 viviparous mammals examined, including primitive marsupials such as the opposum, show periods of REM sleep”*‘*. Even an animal like the mole, which can hardly move its eyes, shows the characteristic EEG of REM sleep. Birds have REM sleep, although often only-a very small amount of it, occupying perhaps 5% of their sleep t3. There are no very convincing reports of REM sleep (as judged by the EEG) in reptiles, amphibia or fish.

If an animal is deprived of REM sleep for one or more nights (but allowed non-REM sleep) then it will usually have more REM sleep in subsequent nights14*t5.

All this evidence suggests that REM sleep has an important function, at least for mammals. Since the majority of dreams are not remembered, that function is more likely to be associated with the unconscious dreaming process - that is, with REM sleep without awakening - rather than with the few dreams which are recalled.

It has been shown that during REM sleep the forebrain is periodically and widely stimulated by the brain stem. This activity in the brain stem can happen even in the absence of the cortex. Hobson and Mc- Carleyr6, following the pioneer work of

Jouvet “, have postulated a ‘dream state generator’ which lies mainly in the pontine reticular formation (the question of which exact cell groups are involved is controver. sial). It produces the so-called PGO waves. They propose that the activity of such cells is the cause of both rapid eye movements and the periodic intrusion of new subject matter into hallucinoid dreams. Our proposals are based on this idea.

In summary, the evidence suggests that in REM sleep the brain is isolated from its normal input and output channels and that it is very active, this activity being promoted by rather nonspecific signals from the brain stem and reflected in the unconscious equivalent of dreaming, which only reaches normal consciousness if the sleeper awakes.

The postulated mechanism We need a mechanism which will tune the cortical system, in the sense of removing parasitic modes which arise after the system has been disturbed either by growth of the brain (when new connections are constantly being made) or by the modifications produced by experience. The mechanism we propose is based on the more or less random stimulation of the forebrain by the brain stem that will tend to excite the inappropriate modes of brain activity referred to earlier, and especially those which are too prone to be set off by random noise rather than by highly structured specific signals. We further postulate a reverse learning mechanism which will [Annotation 8] modify the cortex (for example, by altering the strengths of individual synapses) in such a way that this particular activity is less likely in the future. For example, if a synapse needs to be strengthened in order to remember something, then in reverse learning it would be weakened. Put more loosely, we suggest that in REM sleep we unlearn our unconscious dreams. “We dream in order to forget.”

After this paper had been initially sub- [Annotation 9] mitted for publication, we learnt from Dr John Hopfield that he and his colleagues had independently arrived at the idea of reverse learning, though not in connection with dreams. In a parallel communication’* they have shown that the behaviout of their very idealized neural net is indeed improved by reverse learning. That is, it equalizes the accessibility of stored memories and suppresses most of the spurious ones. We have since repeated their simulations and confirmed their general conclusions. It remains to be seen how well reverse learning acts on other more realistic neural nets. We have revised our paper iI’ the light of their results.

Note that the amount of reverse learninE per step in these simulations was very small (only about 107’0 of the amount needed f@’ complete learning), although several bundred such steps were used. This alerts us t0 the possibility that the changes produced iI1 REM sleep may individually be very small but cumulative over many PGO spikes and

#### Page 3

&uRE VolM 14 JULY Is3

C~MMLNTAR\/

II3 $T gmlly Ilig~ts’+p~ ‘-, 1’ :’ L i :‘.

The ObJection mrght be raised that some .&reriments have shown that REM may appear to help the retention of memory, whereas the process of reverse learning would tend to make the memory fade. The results of Hopfield et aI.‘* show that this need not be the case. After reverse learning the recall of their net was less confused and more uniform. :

If there is indeed a mechanism for reverse learning, many questions arise about its character. Does it act via the same ntechanism as normal learning (whatever that is) or Is a special, quite separate, mechanism involved? Is the mechanism associated with one particular system in the brain stem? Another possibility is that a small amount of reverse learning is always present but is normally overwhelmed by the positive plasticity produced by one or more of the diffuse systems from the brain stem. When their activity is greatly reduced, as it is in REM sleep9, the residual reverse learning can then exert its effect unopposed, at least on recently modified synapses.

learning process.. The latter is a positive mechanism which. does not merely fail to alter synaptic strengths (or other longlasting brain parameters) but changes them so that the dream is not just forgotten but actively ‘unlearned’. The result is that the dream (or some of the elements of it) is less likely to recur in the future.

nature of REM dreams.

The effects of REM sleep deprivation are harder to explain. It is well established that REM deprivation often produces a rebound - more REM sleep than usual occurs when the subject is eventually allowed to sleep without interruption. We would have expected that REM deprivation, if severe enough, might cause hallucinations - that is, structured visual and auditory responses to ‘noise’ - and perhaps delusions and obsessions. There is a little [Annotation 10] evidence for thisz5, but usually the effects are either small or absentz6. This is partly because it is extremely difficult to produce long periods of complete REM deprivation in humans by selective arousal. After a week or two it becomes almost impossible to awaken them promptly at every onset of REM sleep, so that prolonged experiments have not been done. One cannot help but wonder whether similar experiments on food deprivation might lead to the conclusion (if unsupported by other evidence) that food also had no essential function. However, REM deprivation in animals does appear to lower the threshold for cortical instability produced by electroconvulsive shock27-30, which is what we might expect. REM deprivation in humans sometimes produces irritability and an in-, ability to concentrate. One might suggest that these are the effects of the attention mechanism being forced to subdue subthreshold parasitic modes which would otherwise break into consciousness. REM deprivation can also allow feelings and wishes to appear which had previously been kept out of consciousness3r, or, in certain subjects, can show changes towards increased internal fantasy during waking32.

In its simplest form our theory postulates that there is no intelligent supervisor inside the brain which decides in detail which potential neural activities should be left untouched -and which should be damped down. This choice is made solely by the response of the forebrain to the relatively nonspecific signals from the brain stem. In very general terms, the brain stem gives the forebrain a varied pattern of bangs (the PGO waves). Any resulting activity is then modified so that it is less likely to occur in the future.

It would of course be possible to postulate a more complex mechanism. For example, in REM sleep, especially in early development, there could be innate testing programmes, together with a ‘supervisor’ to decide what to store and what to erase, depending on the result of the tests. Various workers have made proposals along these linesrPV2”. As far as we know, nobody has previously suggested that the testing procedure involves the removal of Potentially parasitic modes.

It has been customary to believe that during an unconscious dream the content of the dream is stored in some form of very short-term ‘memory’ but that the mechanism for transferring it into longer term memory is inoperative. We normally become conscious of our dreams only if we wake up while dreaming is in progress. If we then pay attention to our dream, some of its content can be maintained in very short-term memory and may eventually be transferred to longer-term memory as the transfer mechanism becomes activated. Otherwise our dream fades. Thus we can speak of forgetting our dreams, meaning that we know that we had a dream, but are somewhat uncertain of its content.

This forgetting of a dream, which has often been remarked on, does not necessarily involve our postulated reverse

The terms ‘reverse learning’ or ‘unlearning’ are not ideal because they rather imply that one has to learn something first in order to unlearn it. What does a fetus ‘learn’ that has to be unlearned? Our answer is that, during development, the semirandom process of making synaptic cdnnections is likely to produce parasitic modes. it is these which must be ‘unlearned’ in order to obtain a well-behaved system.

We need some explanation for recurrent dreams. We propose the ud hoc hypothesis that a recurrent dream is one which, for one reason or another, tends to wake up the sleeper, perhaps because of the anxiety often associated with them. This will have the effect that the learning process changes sign, passing from reverse learning to positive learning, so that the underlying spurious associations remain, and so a similar dream is likely to occur on some later occasion. This mechanism does postulate a supervisor of a kind but its sole function is to decide whether the sleeper should wake up or not. Thus for a dream to become recurrent it must have two properties. It must be related to a potentially parasitic mode and it must wake up the dreamer in such a way that he remembers it rather vividly.

Our theory, in its present state, says nothing about the function of non-REM sleep. These stages of sleep usually have less of the hallucinoid type of dream which we associate with our reverse learning mechanism. Non-REM sleep is likely to have the restorative function often postulated for it but it may also have some informational function. For example, it might be used for the process of ‘consolidating’ memory in some way. It is worth noting that the first REM period of the night is normally preceded by a substantial period of non-REM sleep.

Testing the theory As far as we can tell, our theory is broadly compatible with a large amount of experimental data. Starting from a plausible hypothesis about cortical function, it ex- [Annotation 11] plains in an effortless way both the need for REM sleep in adult life and the large amount of it during the development of the brain. We believe no previous theory explains this distribution of REM sleep in such a simple manner. Any purely psychological theory (such as Freud’s) is hard-pressed to explain the large amount of REM sleep in the womb, and any purely developmental theory must account for the quite appreciable amount of REM sleep in adult life. Our theory accounts for both. It is also compatible with the hallucinoid

A further difficulty is that some drugs, such as certain monoamine oxidase inhibitors, appear to prevent REM sleep entirely33 without producing very obvious psychological deficits. This is a difficulty for any theory which assumes REM sleep is important and runs in the face of all the other evidence about it. We can only plead that such drugs may have complicated side ‘effects which make the observations misleading.

A direct test of our postulated reverse learning mechanism seems extremely difficult. It would be necessary to show that our unconscious dreams (dreams we do not remember - a new word for this is really needed, we suggest ‘remination’) reduce the probability of such thoughts occurring in the future. This is far beyond the methods we have available today. It would be interesting to know if the threshold for hallucination, induced by drugs or other means, is lowered as a result of REM deprivation. Another approach would be to look for the structural and chemical correlates of the postulated reverse learning mechanism, but exactly how to do this is at the moment unclear. Without further evidence of this kind our theory must be regarded as speculative.

It is clear that useful insights can come

#### Page 4

NATURE VOL.305 14 JULY 19~2

from neural modelling. This appr6ach hasits limitations, sin&it is difficult .to produce realistic models and even more difficult to simulate them effectively, especially if the. hypothetical neural nets approach a realistic size, when the computational time becomes prohibitively long. However, such theoretical studies should at least reveal some of the types of [Annotation 12] networks which would benefit from our proposed mechanism. They might also help to give more life to our otherwise rather vague characterization of the cortical system.

Another approach would be to undertake comparative studies. There is one mammal which, although possessing a well developed neocortex”, appears not to show any signs of REM sleep (at least in young adults), even though it exhibits normal non-REM sleetis. This is the Echidna Tachyglossusaculeatus (the spiny anteater) found in Australia. The Echidnas and the duck-billed platypus are primitive egglaying mammals (monotremes).

Griffiths” has written that “. . . the gyrencephalic cerebrums of the Tachyglossidae have been and are a source of wonder to neurobiologists”. He quotes Elliott SmithM who in ;1902 wrote “The most,obtrusive feature of this brain is the relatively enormous development of the cerebral hemispheres . . . The meaning of this large neopallium is quite incomprehensible . . .“. Griffiths adds: “Determinants of modern neurophysiology also fail to explain how echidnas come by this cortex”. We suggest that Tachyglosncs needs such a large cortex because it cannot tune it up by the process of reverse learning. Experience with idealized neural nets shows that one can usually avoid overloading a net, and thus the confusion such an overloading creates, by making the net bigger.

Tachyglouw can be studied in captivity. It might be rewarding to examine in more detail its behaviour, neurophysiology and neuroanatomy compared to a primitive placental mammal, such as a hedgehog, which does show REM sleep. If REM sleep serves an important function; this should be reflected in some way in its absence in the spiny anteater.

Possible implications If it turns out that our ideas are broadly correct, they could help us to understand the evolution of the neocortex which is so typical of mammals. It seems likely that in order for a highly tuned system to perform efficiently at least two requirements are

necessary: :a fairly constant internal temperature, so that its function is not disturbed by temperature fluctuations, and in addition a cleaning-up mechanism, to remove potentially parasitic modes. In short, without REM dreams evolution could not have produced the highly refined neocortex we have today.

If the reverse learning mechanism we have postulated exists, one might wonder what effects its failure might have. A complete failure might lead to such grave disturbances - a state of almost perpetual obsession or spurious, hallucinatory associations - that it would probably be severely selected against. A partial failure should produce unwanted responses to random noise, perhaps as hallucinations, delusions, and obsessions, and produce a state not unlike some schizophrenias.

‘It has been postulated before that there might be a relation between REM sleep Andy schizophrenia, but studies have shown that there is little oi no connection between the outward signs of REM sleep and schizophrenia3’. However, a partial failure of the reverse learning mechanism would not necessarily alter the amount of REM sleep, since the control mechanisms for the occurrence of REM sleep might be somewhat distinct from the reverse learning process itself. Thus the possibility that some forms of schizophrenia might be

I. Hartmann, E.L. The Functions of S/up (Yale University

Press. London. 1973). 2. Cartmight. R.D. A Primer on sleep and Dreaming

(Addison-Wesley, Reading. Massuchusetts. 1978). 3. Schmitt. F.O.. Worden. F.G.. Addman, G. & Dennis.

S.G. (cds) The Organizalion of the Cerebral Cortex (MIT Press. Cambridge. Massachusetts. 1981). 4. Siegel, R.K. Scienl. Am. 237, 132-141 (1977). 5. Emuntrout. G.B. & Cowan. J.D. Biol. Cybernetics

34.137-Iti(1979). 6. Palm. G. Nrural Assemblies: An Allernalive Approach lo

Artif&llnlclii~en~ (Sarinner. New York. 1982). 7. K&&en. T. A&ciadw Me&y (Springer. New York.

1977). 8. Hinton, G.E. & Anderson, J. feds) Porullel ModeIs ol

Associrrliw Memory (Erlbaum. Hillsdale. New Jersey, 1981). 9. Hobson.J.A., McCarlcy. R.W. & Nelson, J.P. J. Nt-uro-

physiol. (in the press). IO. Roffwarg, H.P., Muzio, J.N. & Demml. W.C. Sciencee’

152.604-619 (1966). II. Allison, T. & Van Twyver, H. Ngr. His/. 79. 56-65

(1970). 12. Allison. T. & Cicchetti, D.V. Science 194. 732-734

(1976). 13. Klein. M., Michel. F. & Jouvet, M. C.r. hrbd.

She. Acad. Sci.. Porti 158.99-103 (1964). 14. Demcnt, W. S-km-e 131,1705-1707 (1960). IS. Kaks. A., Hoedemachcr, F.. Jacobson. A. & Lichrenstein.

E. Nalure 204. 1337-1338 (19641. 16. Hobson. J.A. & McCarley. R.W. Am. J. Psychial.

134.1335-1348(1977). 17. Jouvct. M. Arch% ital. Biol. 100, 125-206 (1962). IS. Hopfield. J.J.. F&stein, D.1. & Palmer. R.C. A’UfUre

304. l58-159(1983). 19. Newman. E.A. &Evans, C.R. Nalure 206,534 (l%S).

caused by a defect in the reverse learning process should not be overlooked.

In this model, attempting to remember one’s dreams should perhaps not be encouraged, because such remembering may help to retain patterns of thought which are better forgotten. These are the very patterns the organism was attempting to damp down.

Finally we should remark that even if it turns out that our ideas are wrong and that nature does not employ the reverse learning mechanism we have postulated, the process may well be useful for artificial intelligence machines of the future, especially those having extensive parallel processing, a learning mechanism and a certain amount of randomness in their construction.

We thank our resident and visiting co]- leagues at The Salk Institute for many useful discussions. We are especially grateful to Drs Allan Hobson and Jim Horne who made detailed comments on our draft manuscript, and to Dr John Hopfield for communicating his work to us before publication and for helpful discussions. This work has been supported by the J. W. Kiechkhefer Foundation, Samuel Roberts Noble Foundation, US Air Force Grant number AFOSR-82-0042 and the System Development Foundation. 0

Received 8 October 1982; accepted I2 May 1983.

20. Gaarder. K: Arch. @n. Psychiut. 14.253-260 11%6). 21. &an. E.M. in Slew and Dreaminn led. Hartmann. E.I

295.jO7 (Little-B&n, Boston, 1970). 22. Greenberg. R. & Pearlman. C. Penp. Rio/. Med. 1.

513-521 (1974). 23. Jouvet. M. Prw. Bruin Rex. 53, 33l-346(1980). 24. Hobson, J.A. in Brain Mechanisms and Percepreol

Awarwwss (eds Pomp&no. 0. & Marsan, C.A.1 379-404 (Raven. New York, 1981). 25. Dement. W. in Academy of Psychoanalysis: Science and

Aychwnolysis Vol. 7 (ed. Masxrman. J.) 129.184 (Grune and Stratton, New York, 1964). 26. Vogel. G.W. Arch. II!. Psychiul. 18.312-329 (1968). 27. Cohen, H.B. and Dement, W.C. Science 150, 1318.1319

(l%5). 28. Hartmann. E.. Marcus. I. & Leinoff. A. Psychono!ll.

Sri. 13. 141.142 (1968). 29. Cohen. H.. Thomas. J. & Dement. W.C. Bruin &‘I.

19,317-321 (1970). 30. Cohen, H.B. & Dement. W.C. Brain Res. 22. 421.4::

119701. 31. G&b&g. R.. Pearlman, C.. Fingar, R.. Kantrowitz. J.

& Kawliche. S. Br. J. med. Psycho/. 43. I-It (1970). 32. Cartwright, R.D. & Ratzel. R.W. Arch. ,een. P$ychioi

27.277-280 (1972). 33. Wyatt, R.J., Kupfer. D.J., Scott, J.. Robinson. D.S. k

Snyder. F. Psychopharmacolopia IS, 236244 (19691. 34. Criffithr. M. ?ht &?lo~y of Monorremes (Academic.

New York. 1978). 35. Allison, T.. Van Twyner. H. & Golf. W.R. Archs. lwl

biol. 110. 145.184 (1972). 36. Smith, G.E. R. Co//. Sarg. Mus. Cur Physiol. Ser. 2.

133.157(19021. 37. Vogel, G.W. & Traub. A.C. Arch. wn. Ps.vrhio’.

18.287-300 (1968).

---

Source: [Fermat's Library](https://fermatslibrary.com/) — https://fermatslibrary.com/s/the-function-of-dream-sleep. More annotated papers: https://fermatslibrary.com/llms.txt
