Below is the unedited draft of:
Revonsuo, Antti (2000), The Reinterpretation of Dreams:  An evolutionary hypothesis of the function of dreaming,  Behavioral and Brain Sciences 23 (6): XXX-XXX.

This is part of a special issue on Sleep and Dreaming containing the following articles:

Hobson, J. Allen, Pace-Schott, E. and Stickgold, R. (2000)
Dreaming and the Brain: Towards a Cognitive Neuroscience of Conscious States [HTML version]
Dreaming and the Brain: Towards a Cognitive Neuroscience of Conscious States [PDF version: BETTER FOR DOWNLOADING]
Behavioral and Brain Sciences 23 (6): XXX-XXX.

Nielsen, Tore A. (2000), Cognition in REM and NREM sleep: A review and possible reconciliation of two models of sleep mentation, Behavioral and Brain Sciences 23 (6): XXX-XXX.

Revonsuo, Antti (2000), The Reinterpretation of Dreams: An evolutionary hypothesis of the function of dreaming, Behavioral and Brain Sciences 23 (6): XXX-XXX.

Solms, Mark (2000), Dreaming and REM sleep are controlled by different brain mechanisms, Behavioral and Brain Sciences 23 (6): XXX-XXX.

Vertes, Robert P. and Eastman, K. E. (2000), The case against memory consolidation in REM sleep, Behavioral and Brain Sciences 23 (6): XXX-XXX.

For information about subscribing or purchasing offprints of the published version, with commentaries and author's response, write to: journals_subscriptions@cup.org (North America) or journals_marketing@cup.cam.ac.uk (All other countries).

 The Reinterpretation of Dreams:
 An evolutionary hypothesis of the function of dreaming
 
 

Antti Revonsuo
Department of Philosophy
Centre for Cognitive Neuroscience
University of Turku
FIN-20014 FINLAND
email: revonsuo@utu.fi

Abstract
 

Several theories claim that dreaming is a random by-product of REM sleep physiology and that it does not serve any natural function. Phenomenal dream content, however, is not as disorganized as such views imply. The form and content of dreams is not random but organized and selective: during dreaming, the brain constructs a complex model of the world in which certain types of elements, when compared to waking life, are underrepresented whereas others are overrepresented. Furthermore, dream content is consistently and powerfully modulated by certain types of waking experiences. On the basis of this evidence, I put forward the hypothesis that the biological function of dreaming is to simulate threatening events, and to rehearse threat perception and threat avoidance. To evaluate this hypothesis, we need to consider the original evolutionary context of dreaming and the possible traces it has left in the dream content of the present human population. In the ancestral environment human life was short and full of threats. Any behavioral advantage in dealing with highly dangerous events would have increased the probability of reproductive success. A dream production mechanism that tends to select threatening waking events and simulate them over and over again in various combinations would have been valuable for the development and maintenance of threat avoidance skills. Empirical evidence from normative dream content, children’s dreams, recurrent dreams, nightmares, post-traumatic dreams and the dreams of hunter-gatherers indicates that our dream production mechanisms are in fact specialized in the simulation of threatening events, and thus provides support to the threat simulation hypothesis of the function of dreaming.
 
 

Keywords: function of dreaming, dream content analysis, nightmares, evolution of consciousness, evolutionary psychology, threat perception.
 
 
 

 INTRODUCTION
 
 

Dreaming is a universal feature of human experience, but there is no convincing explanation as to why we should experience dreams during sleep. Why do we have vivid, intense and eventful experiences while we are completely unaware of the world that physically surrounds us? Couldn’t we just as well pass the night completely nonconscious? The function of dreaming seems to be a persistent mystery, although numerous suggestions have been put forward about the possible functions it might serve. The leading neurocognitive theories, however, seem to have given up the hope of identifying any useful function for dreaming at all. They cannot provide us with an answer to the question "Why do we dream?". Instead, they seem to imply that we dream for no particular reason at all: dreaming is biologically epiphenomenal. Dream consciousness is viewed as some sort of random noise generated by the sleeping brain as it fulfills various neurophysiological functions during REM sleep.

Although the prospects for discovering useful functions for dreaming look rather bleak, the empirical evidence should be reevaluated once more from a truly multidisciplinary point of view, including dream content analysis, the neurophysiology of dream sleep, and evolutionary psychology. The exploration that I undertake in the present paper leads to the slightly surprising conclusion that dreaming does have a well-defined and clearly manifested biological function after all. In the first section, I clarify the nature of the basic question: What exactly is it that we want to understand when we inquire about the function of dreaming? The answer is that we need a clear idea of both what the phenomenon of dreaming is and of the sense in which we are using the word "function". In the second section, we review the currently dominating views on the function of dreaming in the cognitive and neuroscientific literature as well as in the more clinically oriented dream psychology. The most common view in cognitive neuroscience is that dreaming has no function whatsoever. In clinical literature, the function of dreaming has been linked with problem solving and psychological adaptation, but the direct empirical evidence bearing on such functions remains scarce. In section three we point out that none of the previous theories have placed dreaming in the approproate context for evaluating its possible biological functions: the human ancestral environment in which the dreaming brain was evolving for hundreds of thousands of years. If dreaming does have any biologically adaptive functions, they must have been effective in the evolutionary context if anywhere.

In the rest of the paper I argue that switching the context in such a way puts dreaming into an entirely new light, which suggests that the biologically adaptive function of dreaming is to simulate threatening events in order to rehearse threat perception and the appropriate threat avoidance skills and behavioral programs. I emphasize that to claim that threat simulation is the biological function of dreaming is not to claim that every single dream of every single individual should realize this function. It is only to claim that in certain adaptively important situations with certain ecologically valid cues, the system does become fully activated, and this is the principal reason why dreaming was selected for during our evolutionary history.

The threat simulation theory of dreaming is expressed in the form of six propositions, each of which is empirically testable. The propositions can be summarized as follows:
 
 

After presenting the threat simulation theory, other theories that have taken an evolutionary perspective on dreaming are reviewed. While some of them are related to the present view, none of them includes the idea that dreaming is a threat simulation mechanism. In the final section, the theory is compared with neurocognitive theories of dreaming.

Taken together, the Target Paper aims to show that the threat simulation theory of dreaming integrates a considerable body of data from multiple sources in a theoretically meaningful way. The theory treats the conscious phenomenal experience of dreaming as a natural biological phenomenon best understood from the combined viewpoints of psychology, evolutionary biology, and cognitive neuroscience. This multidisciplinary treatment, I hope, manages to clarify the mystery of why we dream.
 
 
 
 
 
 
 
 

1. WHAT IS IT THAT WE WANT TO UNDERSTAND WHEN WE INQUIRE ABOUT THE FUNCTION OF DREAMING?
 
 

We should first make clear what it is that we are asking when we inquire about the function of dreaming. We must explicate what we mean by dreaming and what we mean by function.
 
 

1.1. What is dreaming?
 
 

Dreaming refers to the subjective conscious experiences we have during sleep. We may define a dream as a subjective experience during sleep, consisting of complex and organized images that show temporal progression (Farthing 1992). Questions regarding the function of dreaming must be clearly distinguished from those regarding the function of REM sleep. Dreaming is a subjective conscious experience, while REM sleep is a physiologically defined stage of sleep. Furthermore, as is now clear, REM sleep is neither a necessary nor a sufficient physiological condition for dreaming, although it seems to be the typical and perhaps optimal physiological condition in which fully realized dreams are brought about (Pivik 1991). As Foulkes and Cavallero (1993, p. 9) emphasize, dreaming needs a level of explanation independent of the neurophysiological level at which REM sleep is defined, because "there almost certainly is REM sleep without dreaming and ... there certainly is dreaming without REM sleep. No account of the distinctive physiology of REM sleep could provide either a necessary or a sufficient explanation of dreaming". Thus, the question we will be exploring is: Does it serve any useful function to have, during sleep, the sorts of conscious subjective experiences that dreaming consists of?

In order to make it clear that we distinguish the level of description at which dreaming proper resides from the levels of neurophysiological description, we may say that dreaming is realized at the experiential or phenomenal level of organization in the brain (Revonsuo, 1997). We want to find out whether the realization of this level of organization during sleep serves any natural function. The specification of the functions that lower-level neurophysiological mechanisms serve during REM sleep does not constitute a specification of the functions that the realization of the phenomenal level serves, for the neurophysiological functions can be fully specified without ever mentioning the fact that subjective experience happens to be realized as well^*1*.
 
 

1.2. What is it to be "functional"?
 
 

We must be clear about what we mean by "function" or "functional". The appropriate sense of "function" in this context is that of a biological, adaptive function. According to Tooby and Cosmides (1995) the biological standard is the only standard of functionality that is relevant to analyzing why brain and cognition are organized in one fashion rather than another. A cognitive system is functional in the evolutionary sense if and only if it promotes to the organism’s inclusive fitness. That is, the biologically functional system must solve problems that will increase the probability that the organism possessing the system will produce offspring, or that the organism’s kin will produce offspring. Evolutionary biology gives the concept of "function" a very specific content: the function of a system solely refers to how it systematically caused its own propagation in ancestral environments (Tooby & Cosmides 1995). If dreaming has an adaptive function, then dreaming must solve some adaptive problems whose solution tends to enhance survival and promote reproduction, thus causing the persistence of the brain’s dream-production mechanisms and their spread in the population.

If dreaming does not have any adaptive function of its own, then it is likely to be coupled to properties that do. In that case dreaming is a mere by-product, a non-adaptation that was not selected for (or against) during our evolutionary history, but which was dragged along because the features to which it was coupled were actively selected for. Flanagan (1995) makes an important distinction between "natural" and "invented" functions of dreaming. A similar distinction has been made by other dream theorists between what we do with dreams once we recall them, and what the dream can do itself (Breger 1967, Blagrove 1996). Natural functions are biological, adaptive functions in the sense defined above, whereas invented functions are derivative psychological or cultural functions. We can put our recalled dreams to a variety of personal or cultural uses^*2*, but no matter how enlightening and meaningful such uses may be, they are invented by us, not by natural selection. It is doubtful that any truly natural function of dreaming could be based on the conscious recollection or verbal reporting of dream content, for the natural functions of dreaming, if any, must have been effective in such ancestral conditions and species in which self-reflective dream recollection or reporting were not likely to occur — thus, the natural functions of dreaming cannot have been dependent on them.

Now we are in the position to state our question more specifically. The question we are presently interested in is whether dreaming serves any natural functions: Does the realization of the subjective phenomenal level of organization (the experience of dreaming) solve any adaptive problems? That is, does phenomenal dreaming in any way enhance the prospects of the reproduction of the individual (and/or its close relatives); does dreaming increase the inclusive fitness of the individual?
 
 
 
 

2. CURRENT THEORIES OF DREAM FUNCTION
 
 

2.1. Theories in cognitive neuroscience
 
 

In cognitive neuroscience, recent theories and views on dreaming have led to the conclusion that dreaming as a conscious experience does not serve any useful biological function. Only the neurophysiological events associated with dreaming and REM sleep are assumed to be biologically functional, for they may serve important functions in the development of the brain and in periodically restoring the brain’s neurochemical balance.

The Activation-Synthesis theory (Hobson & McCarley 1977; Hobson 1988) emphasizes the randomness of dream imagery. During REM sleep, PGO waves originate in the pons and activate the forebrain. The forebrain attempts to make sense of this random activation and it synthesizes dream images to fit the patterns of internally generated stimulation. The forebrain selects images that isomorphically correspond to the patterns of eye-movements and motor commands elicited during REM sleep. The images are loaded from memory, in which day residues are particularly salient. The theory delivers no answer to the question why the brain should generate any images at all during REM sleep; it is simply assumed to be an automatic process. The narrative content of dreams remains unexplained as well. More recently, Hobson (1994) has suggested that REM-dreaming might have a function in memory processing, and he specifically regards the rehearsal of motor programs as a possible function of dreaming during REM sleep. In Hobson’s theory dreaming as an experience with vivid phenomenal content is however seen as a kind of random epiphenomenon that merely reflects some totally different events going on at other levels of organization where such events may serve useful neurobiological or mnemonic functions. The Activation-Synthesis theory suggests that the experiential dream imagery itself, the content of consciousness, is functionally as aimless as are the noises emitted by a computer when it processes information. The phenomenal level of organization is not regarded as biologically functional.

The theory presented by Crick and Mitchinson (1983, 1995) is related to Hobson’s views, but contains some original ideas. In this theory, memory in the brain is compared to simple models of associative nets. When such a net gets overloaded, it easily starts to produce outputs that are combinations of actually stored associations. In order to make storage more efficient and avoid overloading, a process of reverse learning can be used. The net is disconnected from its normal inputs and outputs, and random input is given to it. The associations that this random input produces are consequently weakened, and the process is repeated many times with different kinds of random input. According to Crick and Mitchinson (1983, 1995) this is loosely analogous to what happens in the brain during REM sleep: the brain is disconnected from its usual inputs and outputs, and PGO-waves provide it with more or less random input^*3*. The theory explains why REM dreams are full of bizarre intrusions, consisting of mixtures of features previously stored in memory: these are the associations arising in an overloaded network and to be unlearned. The reverse learning theory does not even try to explain the narrative aspect of REM dreams, and it certainly does not assign any independent function to the phenomenology of dreaming; phenomenal dream images merely reflect the functioning of a memory-cleaning process.

David Foulkes (1985) has put forward a cognitive theory of dreaming. He proposes that dreaming originates in diffuse, more or less random activation of semantic and episodic memory during sleep: "Since it seems that the activation of mnemonic elements during dreaming and their selection for dream processing is random and arbitrary, it’s not likely that the particular content of our dreams — in and of themselves — serve any adaptive functions." (Foulkes 1985, p. 200).

Foulkes, however, distinguishes dream content from dreaming as a process. Dreaming, unlike specific dream contents, has very predictable features. It involves an interrelated sequence of events occurring within a "world analog" (or a model of the world) composed of integrated multimodal sensory imagery; the dreamer participates in these events actively and personally; the contents and events depicted in the dream are related to the recent or distant past of the dreamer, not as a simple replay of a past experience but rather as a variation of the past as something that really could have happened to the dreamer. Foulkes suggests that, since the content of dreams seems to be random, what is important about the mnemonic activation is that it is in some way unique, not the precise way in which it is unique. In Foulkes’ theory the phenomenal level of organization is not regarded as functional, apart from the general feature of producing novel and unique mnemonic configurations. Thus, Foulkes’ theory is not essentially different from Hobson’s as to the functionality of phenomenal dream content.

Solms (1997) has recently defended the view originally proposed by Freud: the function of dreaming is to protect sleep. According to Solms, the dream process begins when external or endogenous stimuli activate "the curiosity-interest-expectancy circuits". Inhibitory mechanisms prevent the "appetitive interest", aroused by stimulation, from leading to motor activity; therefore the activity proceeds "regressively" in the direction of hallucinations. In anxiety dreams this mechanism of sleep protection fails. It is clear that this view does not attribute any functions to the specific content of dreams: Solms regards dreams simply as bizarre hallucinations that the weakened frontal reflective systems mistake for real perception.

Owen Flanagan (1995) explicitly denies that dreams as conscious experiences have any biological function. Dream experience or p-dreaming (phenomenal-dreaming) as Flanagan calls it, is "... a likely candidate for being given epiphenomenalist status from an evolutionary point of view. P-dreaming is an interesting side effect of what the brain is doing, the function(s) it is performing during sleep. To put it in slightly different terms: p-dreams, despite being experiences, have no interesting biological function. ... I mean in the first instance that p-dreaming was probably not selected for, that p-dreaming is neither functional nor dysfunctional in and of itself." (Flanagan 1995, 9-11). Flanagan argues that phenomenal experience during dreaming — dream consciousness — has no adaptive significance, because the functions of REM sleep and PGO waves, in early development of the visual system and in the restoration of neurochemicals for the next waking period, do not in any way require mentation of any sort. Furthermore, dream thoughts associated with such biological functions do not seem to be worth remembering. "The visual, auditory, propositional, and sensory-motor mentation that occurs is mostly noise" (p. 24). Antrobus (1993) seems to agree with Flanagan’s analysis. He says that since in REM sleep no sensory information is processed and no association-motor commands are executed, it makes no difference what the association cortex does. Dreaming has no maladaptive consequences, so it has survived.

In conclusion, theorists in cognitive neuroscience tend to regard the phenomenal content of dreaming as a biological epiphenomenon, although at least part of the (nonconscious) cognitive and/or neural activity during REM sleep is regarded as serving useful functions.
 
 

2.2. Theories in dream psychology
 
 

In psychological theories of dream function, the emphasis is on the individual person’s psychological adaptation to his current waking life. The basic assumption behind this approach seems to be that dreaming is functional for the individual in case the dream in some way helps the individual to cope with his current waking concerns, to solve current problems, and to promote psychological well-being. These views can be traced back to Jung (1933) who argued that dreaming helps to maintain the individual’s psychic balance and Adler (1927) who believed that dreaming serves a personal problem-solving function.

These types of theories of the psychological function of dreaming can be divided into two categories. The first holds that dreaming has a problem-solving function in an intellectual or cognitive sense: The function of dreaming is to find solutions to (or to facilitate the solving of) intellectual problems. The second holds that the function of dreaming is related to emotional adjustment, not to intellectual problems. Any real-life event that can be considered an emotional concern for the dreamer can be seen as presenting a problem for psychological adjustment, and dreaming is assumed to contribute to the emotional or behavioral adjustment that is called for in order to solve the emotional problem (e.g. Breger 1967).
 
 

2.2.1. Do dreams solve intellectual problems?

Some studies have directly addressed the question whether we can solve intellectual problems in our dreams or with the help of them. Dement (1972) reports a series of experiments in which 500 undergraduate students were given a copy of a problem, and before going to bed the students were supposed to spend exactly 15 minutes trying to solve the problem. In the morning, they wrote down any dreams they recalled from the previous night and if the problem had not been solved, spent another 15 minutes trying to solve it. In 1148 attempts, the problem was solved in a dream on only seven occasions. This means that less than 1% of the dreams were successful in solving the problem. Montangero (1993) reports a sleep laboratory experiment with six subjects. Four subjects were given a formal problem, while two were trying to solve an intellectual problem relating to their own professional career. Although elements of the problems appeared in the dreams, none of the twenty-nine reported dreams presented the solution to the problem. However, the subjects did find the solutions to the problems with relative ease during the first hour after awakening in the morning. Unfortunately, it remains unclear whether dreaming causally contributed to this problem-solving success at all. Cartwright (1974) compared solutions to problems arrived at either after a period of REM sleep or an equivalent amount of waking. She concluded that "There is no evidence from this study that a period of sleep during which dreaming occurs is regularly followed by a better performance on intellectual tasks" (p. 454). In a study by Barrett (1993) the subjects were allowed to choose the problem that they tried to solve in their dreams. The results showed that problems of a personal nature were much more likely to find a solution through dreaming than problems of an academic or intellectual nature. The personal problems, however, lacked definitive criteria for what should count as a solution, raising the suspicion that at least some of the alleged solutions may have been attributed to the dream during retrospective reflection required during the reporting rather than having been arrived at within the dream itself.

Blagrove (1992) presents a thorough review and critique of the problem-solving paradigm of dream function. The assumption behind this paradigm is that the function of dreaming is to work actively and creatively towards solutions to actual current waking problems, thus going beyond what was known prior to the dream and causally contributing to the solution of a real-life problem. In order to evaluate the evidence for such claims, Blagrove distinguishes three types of problem-solving dreams: (1) Dreams which actually create a new and useful solution to a current problem in waking life; (2) Dreams which contain problem-solving activity that is internal to problems encountered in the dream world, but not relevant to waking problems; (3) Dreams which reflect solutions to waking problems, but there is no evidence that such solutions have not already occurred to the waking mind (i.e. the dream does not contribute to the solution, it merely reflects the solution once it has already been found during waking). Blagrove (1992) argues that there is little evidence for problem-solving dreams of the first type; most of the dreams apparently solving problems either simply reflect solutions already known or solve problems only relevant in the context of the dream. Although a psychological change may be correlated with a dreamed solution to a problem, there is little reason to believe that there is a causal relationship between them. It is most likely that the actual solution first arises during waking, and the consequent dreaming merely reflects the solution, and thus becomes correlated with whatever the beneficial consequences of the solution were. The conclusion from Blagrove’s (1992) review is that whatever the function of dream experience is, it does not appear to be the finding of new and useful solutions to the problems we face in our waking reality.
 
 
 
 

2.2.2. Do dreams solve emotional problems?

Probably the most popular theory of dream function within psychology is the hypothesis that dreaming solves our emotional problems by helping us to adjust psychologically to, and maintain our mental health in, the real-life situations that trouble us emotionally and psychologically. There is an overwhelming amount of evidence showing that dream content indeed reflects the current emotional problems of the dreamer (Kramer 1993; Hartmann 1998). The question is: does dreaming have an effect in reducing the negative affect and other negative psychological consequences induced by our real-life troubles and traumas?

Cartwright (1996) argues that the best way to test this hypothesis empirically is to study subjects who are undergoing a life event that creates genuine affect. She studied subjects undergoing marital separation. Seventy subjects were chosen from a group of 214 potential subjects. Forty of them were depressed as a consequence of the divorce. All subjects slept for three nights in the laboratory, and during the third night, REM dream reports were collected. The depressed subjects’ dreams were emotionally more negative than those of the non-depressed subjects. Furthermore, the depressed subjects were more likely than the non-depressed to incorporate the about-to-be-former spouse as a character in the dreams. In a one-year follow-up, those depressed subjects who had dreamt about their spouse were better adjusted than those who did not. However, it remains unclear how this correlation should be interpreted; on the basis of this study no causal relationship between dream content and adjustment can be established.

Kramer (1991, 1993) argues that during REM sleep there is a surge of emotion, and that the function of dreaming is to contain or to attempt to contain this surge. If the dream is successful in fulfilling this function, it does not enter awareness or memory, but protects sleep. A successful pattern of dreaming first states and then works on and resolves the problem, which leads to a positive affective outcome and no dream recall. Kramer’s (1993) studies show that a successful night’s dreaming is associated with having more characters in the dreams and leads to increased happiness during the next waking period. If the problem is simply restated and not solved, as in repetitive nightmares, then the problem remains unsolved, emotions remain negatively toned, and the dream easily enters awareness. Nightmares and bad dreams are therefore seen as unsuccessful attempts at solving our emotional problems. This theory is called the selective mood regulatory theory of dreaming (Kramer 1993).

Hartmann (1995, 1996, 1998) has recently argued that our dreams deal with our emotions and emotional concerns by making pictorial metaphors of them. Dreaming cross-connects or weaves in new material, which, according to Hartmann (1998), helps us adapt to future trauma, stress and the problems of life. Thus, dreaming and psychotherapy fulfill somewhat similar functions. A stressful real-life experience can be processed in both cases in a similar way: essentially by "making connections in a safe place", i.e. by associating and integrating traumatic experiences with the rest of life in order to facilitate psychological healing. Dreaming "calms" the emotional "storm" going on in our minds. Hartmann calls the class of psychological adaptation views of dreaming consistent with his theory the "Contemporary Theory of the Functions of Dreaming".

Punamäki (1997, 1998) has recently tested the role of dreams and dream recall in protecting psychological well-being in traumatic conditions. She studied the dreams of a group of Palestinian children living in a violent area in Gaza and a control group living in a peaceful area in Galilee. She reports that traumatized children had better dream recall than non-traumatized ones, and the more the children were exposed to trauma, the more negatively emotional and the less bizarre were their dreams. Frequent dream recall was associated with depressive symptoms, whereas infrequent dream recall was associated with somatic and anxiety symptoms. Thus, the pattern of mental health effects associated with dream recall is not straightforward, for both good and bad dream recall were associated with some, although different, psychological symptoms. Furthermore, on the basis of this study it remains unclear whether dream recall was a cause or a consequence of these symptoms, as well as whether frequent or infrequent dream recall in any way serves a positive long-term mental health function in the recovery from trauma.

Thus, the literature on the possible mental health functions of dreaming is inconclusive as to whether dreams truly solve our emotional problems, protect our mental health, or help us to adjust psychologically and to recover from traumatic experiences. The empirical evidence for such psychologically adaptive functions appears to be relatively weak and correlational at best. Furthermore, it is not entirely clear what the predictions of such a theory really are and whether the empirical evidence confirms or disconfirms them. If the idea is that dreaming "protects" our mental health from negative emotional impact by turning the stressful emotional experience into something better and by integrating it with the rest of our lives, it is surprising how often this function deserts us when we need it most. Recurrent dreams occur during times of stress, are accompanied by negative dream content and associated with a deficit in psychological well-being (Zadra et al., 1997-1998). When we live under constant emotional stress or have recently experienced trauma, our dream consciousness typically makes us suffer from intensive nightmares that constantly remind us of the trauma by re-activating powerful negative feelings and other elements from the trauma (see section 3.5. below). If the real function of dreaming is psychological healing, shouldn’t we in fact expect exactly the opposite: pleasant, comforting, manifestly healing dreams; calming, not amplifying the traumatic experience? Intuitively, reliving the emotional shocks over and over again in dreams would not seem to be exactly what traumatized people are psychologically in need of.

The usual explanation for this anomaly is that the assumed dream function has simply "failed"; nightmares are treated as "failures" of dream function (Kramer 1991). But if this is so, then dream function fails a little too regularly, and exactly when it would be needed most.

In opposition to these psychological adjustment theories of dreaming, I shall argue that nightmarish dreams are not ones that failed to perform their function, but, by contrast, prime examples of the kind of dreams that fully realize their biological function. The view that dreams solve our emotional problems and increase our happiness and psychological well-being seems to include the biologically misguided assumption that normal life is free of emotional pain and trauma. Biologically adaptive responses to danger, such as pain and fear, are not there in order to increase our happiness, but to increase our reproductive success. Natural selection cares only about fitness, not our comfort (Nesse and Williams 1997). If dreams are biological adaptations, they may not care about our comfort either.
 
 
 
 

3. THE BIOLOGICAL FUNCTION OF DREAMING
 
 

The discussion above shows that there is no convincing evidence that dreaming would causally contribute to the solving of either intellectual or emotional problems. We must look elsewhere to discover the biological function of dreaming.
 
 
 
 

3.1. Background Assumptions
 
 

The construction of the appropriate context for discovering the biological function of dream consciousness requires clarification of the following two questions:
 
 

(A1) Consciousness can be reconceptualized as the phenomenal level of organization in the brain (Revonsuo, in press). A function attributed to consciousness concerns the causal powers and behavioral effects of events realized at the phenomenal level of organization. The phenomenal level forms the brain’s real-time model of the surrounding world, of the organism’s internal state, and of its external position in the environment. Dreaming as a subjective experience is realized at the phenomenal level.

(A2) The primary evolutionary context for considering the possible adaptive function of dream consciousness is the prehistoric Pleistocene environment in which humans and their ancestors lived as hunter-gatherers for hundreds of thousands of years. If dream consciousness is biologically functional, it should have had adaptive value at least in that original environment, in the conditions in which human ancestral populations lived. Whatever the adaptive role of dream consciousness might have been in that long-gone original context, there is no guarantee that the average dreaming brain today, facing a completely different environment than the one in which it evolved, should fulfill any functions that we recognize as adaptive in the present environment.

I will simply take these answers as background assumptions that are reasonably well established; space does not permit a full defense of these views here (but for more on consciousness see Revonsuo, 1995, 1997, and for an evolutionary perspective in cognitive neuroscience see Tooby & Cosmides 1995, Cosmides & Tooby 1995).

When put into the proper context in this manner, the question "Does dream consciousness have a function" becomes:
 
 

3.2. Dream Consciousness and Threat Simulation
 
 

We are now ready to formulate an evolutionary hypothesis on the function of dreaming. The hypothesis I am putting forward states that dream consciousness is essentially a mechanism for simulating threat perception and rehearsing threat avoidance responses and behaviors.

The threat simulation hypothesis of dreaming is presented below in the form of several independent empirically testable propositions. If each of these propositions is judged as probably true in the light of empirical evidence, then the threat simulation hypothesis will receive considerable empirical support; but if most of them are not supported by empirical evidence, then the hypothesis will be falsified. I try to show that there are good reasons to believe that each of these propositions is actually true.
 
 

3.3. PROPOSITION 1:
 
 

This highly predictable and organized form of dreaming presents a challenge to any view claiming that dream experience is merely an incidental by-product of neurobiological processes operating at a different level of organization. It is extremely implausible that a low-level neurochemical restoration process, for example, should produce as some sort of ‘noise’ a complex and organized model of the world at a higher level of organization (cf. Foulkes 1985). If dreams truly were just noise, they should appear much more noisy and disorganized than they actually are. Random noise in the system is not likely to create organized perceptual wholes, nor is it likely to make a good story, or any story at all^*4*; it would be expected to generate disorganized sensations and isolated percepts. True noise in the brain is produced in connection with an aura of migraine for example. It does not generate an organized perceptual world of objects and events; rather the contrary, it produces for instance white or colorful phosphenes, geometric forms, and scintillating and negative scotomata (Sacks 1992). The visual hallucinations connected with Charles Bonnet syndrome usually consist of static images of people, animals, buildings, and scenery (Schultz & Melzack 1991). Were our dreams closely to resemble these phenomena it would be easy to believe that dreams consist of nothing but random noise reflecting neurobiological processes at other levels of organization in the system.

It could, however, be argued that even random or disorganized processes might activate organized schemas and scripts and thus produce dreamlike phenomenology. For example, in Penfield’s (1975) studies the direct electrical stimulation of temporal cortex produced vivid and realistic perceptual "flashbacks". Still, these experiences were in many ways dissimilar to dreams: they were short (a few seconds) and undramatic excerpts of the patients’ previous experiences, like randomly chosen artificially activated memory traces: "The mechanism is capable of bringing back a strip of past experience in complete detail without any of the fanciful elaborations that occur in a man’s dreaming" (Penfield 1975, p. 34). Thus, the activation of such traces would not produce dreams as we know them. Consequently, there is no evidence that any kind of essentially random activation could produce the phenomenology and narrative structure of fully developed dreams.

Dream phenomenology therefore is likely to be the consequence of an active and organized process rather than a passive byproduct of disorganized activation. This process generates an organized world-model. Foulkes (1985) points out that dreams are coherently organized both momentarily and sequentially. The momentary phenomenal content of dream consciousness is comprehensible and conforms to the kinds of multimodal perceptual experiences that we have during waking perception. These momentary phenomenal contents cohere sequentially so as to constitute narrative stories or temporally extended episodes of experience of the same general form as our waking experience. According to Foulkes, dreams are credible multimodal world analogs that are experienced as life: "The simulation of what life is like is so nearly perfect, the real question may be, why shouldn’t we believe this is real?" (Foulkes 1985, p. 37).

Thus, all of the above shows beyond any reasonable doubt that dreaming is an organized simulation of the perceptual world; a virtual reality (Revonsuo 1995). Even granted this, it could still be the case that the phenomenal content of dreaming is simply a random or indiscriminate sample of the phenomenal content of waking consciousness (or the episodic memories thereof). However, this does not seem to be the case. There are certain experiences that are very frequent contents of consciousness during our waking lives but rarely or never enter our dreams. Hartmann (1998) describes two studies in which it was shown that even subjects who spend several hours daily reading, writing or calculating virtually never dream about these activities. In the first study, two judges examined 129 written dream reports from several studies and found no instances of reading or writing and only one possible instance of calculating. In another study a questionnaire was mailed to 400 subjects who were frequent dreamers and interested in their dreams. They reported spending the average of six hours per day engaged in reading, writing, calculating or typing, but answered that they dreamed "never" or "almost never" about any one of these activities. They furthermore estimated on a seven point scale how frequent different activities are in dreams compared with waking life. Their ratings showed a remarkable dissociation between waking and dreaming life: the average rating was at the "far more prominent in my waking life than my dream life" end of the scale as to the frequency of writing, reading and typing.

This shows that dreaming is not only an organized but also a selective simulation of the world. Not every type of event or activity is simulated by the dream production mechanisms, no matter how prominent they may be in our waking lives. Given that reading, writing, typing and calculating are excluded from, or at least grossly underrepresented in, dream experience, what kind of phenomenal content is overrepresented in it? Which events is dream experience really specialized in simulating? This question leads us to Proposition 2.
 
 
 
 

3.4. PROPOSITION 2:
 
 

If dreams are specialized in simulating threatening events, then we ought to find that dream content is biased towards including various negative elements (reflecting threats) rather than positive elements. Several prominent features of dream content suggest that this bias indeed exists.
 
 

3.4.1.1. Emotions in Dreams. In the normative study by Hall & Van de Castle (1966) 500 home dream reports from female and 500 from male college students, aged 18-25, were content analysed. About 80% of the over 700 emotions expressed in the dream reports were negative, and only 20% positive. The figures remain similar if the dreamer’s own emotions only are considered. About half of the negative emotions experienced by the dreamer were classified as "Apprehension", the other half consisted of "Sad", "Anger" and "Confusion".

In the first normative laboratory study, Snyder (1970) collected 635 REM dream reports from students and found that over two thirds of the emotions mentioned in the reports were negative, fear being the most common and anger the next most common. Strauch and Meier (1996) report a sleep laboratory study in which they not only collected REM dream reports from 44 subjects but also asked them how they had felt during the dream. The emotions described in response to this question were analysed. Specific emotions were mentioned in connection of every other dream. Negative emotions appeared twice as often as positive ones, with anger, fear and stress being the most frequent types of negative emotions. In contrast to specific emotions, general mood states were found to be more often positively than negatively toned.

Foulkes et al. (1988) and Revonsuo & Salmivalli (1995) have shown that emotions in dreams are in most cases appropriate to the dreamed situations in which they are experienced; therefore, the high proportion of negative emotions is a sign of frequent unpleasant dream events that should be expected to produce negative emotions if they were real. Emotions are evolved adaptations that increase the ability to respond appropriately in adaptively important situations. Negative emotions such as anxiety, fear and panic, can be seen as adaptive responses that increase fitness in dangerous situations threatening a loss of reproductive resources (Marks & Nesse 1994). When emotions are experienced or expressed in dreams, they are much more likely to be negative than positive ones, and very likely to be appropriate to the dreamed situation. These findings are well consistent with the hypothesis that dream content is biased towards simulating threatening events.
 
 

3.4.1.2. Misfortunes in Dreams. "Misfortune" names a class of dream event in which a bad outcome happens to a character independent of anything the character has done (Hall & Van de Castle 1966). Misfortunes include e.g. mishaps, dangers and threats. The opposite is called "Good Fortune". In the Hall & Van de Castle (1966) normative study, there were altogether 411 cases of Misfortune in 1000 dream reports, and only 58 cases of Good Fortune. Thus, Misfortunes in dreams are seven times more frequent than Good Fortunes. Furthermore, about 70% of the misfortunes happen to the dream self, and it is accidents, losses of possession, injuries or illnesses, obstacles, and threats from environment that comprise almost 90% of these misfortunes, whereas death and falling are rare types of misfortune (Hall & Van de Castle 1966; Domhoff 1996). Misfortunes therefore typically reflect situations in which the physical well-being or the resources and goals of the dream self are threatened.
 
 

3.4.1.3. Aggression in Dreams. Aggression is the most frequent type of social interaction found in dreams, the other classes in the Hall & Van de Castle (1966) scale being Friendliness and Sexual Interactions. About 45% of the dreams in the normative sample included at least one aggressive interaction. Dreamers are involved in about 80% of the aggressions in their dreams, and when they are involved, they are more often the victim than the aggressor (Hall & Van de Castle 1966; Domhoff 1996).
 
 

3.4.1.5. Summary. Negative emotions, misfortunes and aggression are prominent in dreams. These findings indicate that normative dream content frequently contains various unpleasant and threatening elements, which supports the view that dreams are specialized in simulating threatening events.
 
 
 
 

3.4.2 Dream content is consistent with the original evolutionary environment of the human species rather than the present one
 
 

3.4.2.1. "Enemies" in our dreams. Domhoff (1996) defines "Enemies" as those dream characters with which the proportion of aggressive encounters of all aggressive + friendly encounters is over 60%. This calculation on the Hall & Van de Castle (1966) normative sample reveals that animals and male strangers are the enemies in men’s as well as women’s dreams. (Men vs. Animals 82%; Women vs. Animals 77%; Men vs. Male Strangers 72%; Women vs. Male Strangers 63%). Encounters with Female Strangers are not at all so aggressive, but predominantly friendly (Men vs. Female Strangers 40%; Women vs. Female Strangers 43%) (Domhoff 1996). According to Hall & Domhoff (1963), unknown males are responsible for the high proportions of victimization and physical aggressions with male characters.

Hall (1955) content analysed 106 dreams of being attacked and found that the attacks predominantly represented situations in which the dreamer’s life or physical well-being was at stake. The attacker was usually human or a group of humans (70%) but not infrequently an animal (21%). When the sex of the human attacker was identified it was virtually always male. The dreamer usually reacted to the attack by running, escaping, or hiding (unless she woke up). Hall and Domhoff (1963, 1964) analysed aggressive and friendly interactions in over 3000 dream reports. They found that interaction was aggressive with 48% of the animal characters in men’s dreams and with 29% of the animals in women’s dreams.

Van de Castle (1983) compared college students’ dreams (over 1000 dream reports altogether) in which humans are the dominating dream characters with those ones in which animals predominate. He found that dreams with animal figures typically take place in an outdoor setting, have a great deal of activity that is often of a violent nature, and that the dreamer typically experiences fear. If an animal figure initiates an interaction with the dream self, the nature of the interaction is aggression in 96% of the time and friendliness only in 4%. Van de Castle writes that "almost without exception, if the animal figure initiates any response to the dreamer, it is some form of threat or hostility" (p. 170).

Why are animals and male strangers our enemies in dreams? Ancestral humans lived in environments in which many animals (e.g. large carnivores, poisonous animals, parasite-carrying animals) presented an ever-present mortal threat for humans. Therefore, behavioral strategies to avoid contact with such animals and to escape or hide if attacked by them obviously were of high survival value. Some deep-rooted human fears and phobias of snakes, spiders, rats, and open spaces are indications that ancient threat avoidance programs still remain with us (Marks & Nesse 1994). Dreaming simulates and rehearses these ancestral threat-avoidance programs in order to maintain their efficiency, because the costs of a single failure to respond appropriately when the danger is real may be fatally high, while the costs of repeated threat simulations during sleep are remarkably low.

Our present-day encounters with unfamiliar males in the waking life are not predominantly aggressive. In the ancestral human environment, however, intergroup aggression and the violent competition over access to valuable resources and territories is likely to have been a common occurrence. Since intergroup warfare and violence was and still is almost exclusively practised by males (Wrangham & Peterson 1996; see also Campbell 1999), encountering male strangers is likely to have been a potentially threatening situation in the ancestral environment, comparable to the threats presented by dangerous animals. Indications that unfamiliar males often present a mortal threat to offspring comes from other primates where infanticide by genetically unrelated males is common (Hrdy 1977). Furthermore, human infants universally develop stranger fear at about six months of age, and even in the modern world are much more likely to be killed or abused by genetically unrelated adults than by close kin (Daly and Wilson 1988). Thus, although an overwhelming majority of our current waking-life encounters with animals and male strangers are not particularly aggressive or threatening, dream content still reflects the ancestral conditions in which such encounters were potentially life-threatening. Dreams are biased towards simulating threats that were common in our ancestral environment.
 
 

3.4.2.2. Children’s dreams. If dreams are naturally biased towards simulating ancestral threats, then we should expect that the traces of these biases are strongest early in life, when the brain has not yet had the chance to adjust the biases in order to better fit the actual environment. This seems to be the case when it comes to the appearance of animals and aggressions in children’s dreams. One of the most prominent differences between child and adult dreams is the much larger number of animal characters in children’s dreams. Hall and Domhoff (1963, 1964) analysed about 500 dream reports from children aged 2-12 years; Hall later increased the sample to 600 dreams and Domhoff (1996) reports the results from this larger sample. Animal characters make up about 25-30% of all characters in the dreams of children 2-6 years of age, and about 15% in 7-12 years of age, whereas the normative finding for adult dreams is about 5% (Domhoff, 1996).

Also Van de Castle (1983) reports studies on children’s dreams. The 741 dream reports (one from each child) were written down by schoolteachers or directly reported by the pupils themselves. The general trend toward a decrease in the frequency of animal dreams as a function of age is clearly manifested. Two-year averages in the percentage of animal dreams for children 4-16 years old were 39.4% for 4-6 years olds, and 35.5%, 33.6%, 29.8%, 21.9% and 13.7% for the next consequtive two-year age groups. In an earlier study on a smaller sample of dreams, Van de Castle (1970) reported closely similar figures (Fig. 1).

In their dreams children surprisingly often encountered such animals that they never or only relatively seldom could encounter in the waking world. Wild or frightening animals (e.g. snakes, bears, monsters, lions, spiders, gorillas, tigers, wolves and insects) comprised nearly 40% of all animal characters in children’s dreams in this study, but under 20% in college students’ dreams. Dogs, horses and cats accounted for 28% of animals in children’s dreams but for 38% in college students’ dreams (Van de Castle 1983). Thus, the proportion of domestic animals increases and that of wild animals decreases with age.

Due to the methods of collecting the dream reports, the studies mentioned above may have included a somewhat biased sample of dreams^*5*. However, also in the laboratory study of Foulkes (1982) animals were the major characters in the dream reports of children ages 3 to 5 and 5 to 7, appearing in 30-45% of the reports. Also the decrease in the number of animal characters with increasing age was confirmed. Strauch (1996) reports results from both home and laboratory REM dreams in Swiss children aged 9-11. Both types of dreams involved more animals than young adults’ dreams, again confirming the decrease of dream animals with increasing age. Home dreams contained animals about twice as often as laboratory dreams, which was explained by dream report length: home dreams were longer and included more characters. Girls’ dreams contained more animals than boys’ dreams. In the REM dreams, 102 animals were found. In girls’ dreams, tame animals and pets prevailed (63%) over wild native or exotic animals (37%), whereas in boys’ dreams, wild animals were much more common than tame ones (61% vs. 39%). Taken together, on the average one out of two animals encountered in the children’s dream world is an untamed wild animal. For boys around 10 years of age this is the most common type of dream animal.

Hall & Domhoff (1963) showed that children also have a higher rate of aggression in their dreams than adults. The greatest amount of aggression occurs in the dreams of children aged 2 to 12 years. According to Domhoff (1996), much of this larger amount of aggression is with animals and the child is usually the victim of an attack by the animal. In Strauch’s data (1996) of combined REM and home dreams, about 30% of all the animals appearing in 10-year-old children’s dreams were in the role of aggressors, compared to 10% in adults’ dreams.

Levine (1991) studied the representation of conflicts in the dreams of 77 children who were about 10 years of age and came from three different cultures: Bedouin, Israeli, and Irish. Conflictual dreams accounted for about two thirds of the reports and were reported about twice as often as non-conflictual dreams in all three cultures. The Bedouin children, who were living in a traditional seminomadic tribe, had dreams that were realistic and concerned with threats to physical survival, usually from the natural world.

Children’s dreams thus show strong biases towards simulating a world that contains animals (especially wild animals), aggression, conflicts, animal aggressors, and victimization to a greater degree than their own waking world or the dream world of adults does. These biases decrease with age if the child’s real environment is largely devoid of them. It seems unlikely that young children should have had more frequent real waking experiences of such things than teenagers or adults have; therefore, it is difficult to explain these biases by referring to the waking lives of the children^*6*. These biases seem to be another sign of the fact that the dream production system is prepared to simulate threatening events consistent with and prevailing in the human ancestral environment. The biases decrease with age, as the perceptual world proves to be quite different than what was anticipated by the dream production mechanisms.
 
 

3.4.2.3. Recurrent dreams and nightmares. Robbins and Houshi (1983) asked 123 university students whether they had ever had recurrent dreams, and if yes, they were asked to describe them. Sixty percent reported that they had had recurrent dreams, many beginning in childhood. A content analysis revealed that only one type of recurrent dream occurred with any frequency, an anxiety dream in which the dreamer was being threatened or pursued. The threatening agents were wild animals, monsters, burglars, or nature forces such as storms, fires, or floods. The dreamer was watching, hiding, or running away. The authors regarded these descriptions as reasonably close to nightmares (Robbins and Houshi 1983, p. 263). Recurrent dreamers reported more problems in their lives and more physical symptoms than those who did not report recurrent dreams, indicating that recurrent dreams may be related to increased levels of stress. Feldman and Hersen (1967) found that frequent recurring nightmares in adults were related to conscious waking concerns about death and to having experienced the death of a close relative or friend before the age of ten. Zadra et al. (1997-1998) reported that in both late teenagers and older adults recurrent dreams with negative content occur during times of stress.

Nightmares, or long, frightening dreams that wake the dreamer, are the paradigm cases of highly unpleasant dreams. It is estimated that almost everyone has had a nightmare, that children, especially during the ages 3 to 6, very frequently experience some, and that adults quite commonly have occasional nightmares. In a study of 1317 subjects, 5% reported having nightmares once per week and an additional 24% once per month (Feldman & Hersen 1967). The themes in the dreams of lifelong nightmare sufferers are remarkably similar to the themes of recurrent dreams, and the most frequent theme is, again, that of being chased or attacked (Domhoff, 1996; Hartmann 1984). Such dreams usually begin in childhood and involve being chased by a monster or a wild animal. In adulthood, the chaser was more likely to be a large unfamiliar man, a group of frightening people, or a gang. These dreams can be frequent, seem very vivid and real, but still do not usually reflect any actual events that ever happened to the dreamer (Hartmann 1998).

Recurrent dreams and lifelong nightmares not directly connected with any real-life traumas appear to be very powerful simulations of rather primitive threats. Again, we should note that the origin of these simulations apparently is not in the real life of the dreamer. Where do these recurrent themes come from? In the light of the human ancestral environment, it makes great sense to simulate violent encounters with animals, strangers and natural forces, and how to escape from such situations^*7*. Therefore, these simulations are incorporated as default values in the threat simulation system, and they can be activated in almost anybody, at least occasionally. In lifelong nightmare sufferers the trigger seems to be the fact that, because of their highly sensitive personality ("thin boundaries"; Hartmann 1984), for them even everyday experiences may be highly stressful or traumatic (Hartmann 1998; Domhoff 1996) and, as we will see in later sections, such emotional triggers can have profound effects on subsequent dream content.
 
 

3.4.2.4. Absence of reading, writing, typing, and calculating. One explanation for the fact that we do not dream about reading and writing is that they include little if any emotional charge for us. However, Hartmann (1998) found that walking, talking to friends, and sexual activity are represented in dreams about as often as in real life, although these activities differ considerably as to their emotionality. Therefore, the principal reason why we do not dream about writing, reading and doing arthmetics probably is that all these activities are cultural latecomers that have to be effortfully hammered into our evolved cognitive architecture. They were not present in ancestral environments nor are they neurally hardwired in the human brain in the way that other complex cognitive functions, frequently present in dreams, are (e.g. speech comprehension and production). Furthermore, they are highly dependent on abstract symbol systems rather than on the recognition or manipulation of concrete objects. Thus, they are in many ways activities fundamentally different from the ones that the human brain was selected for in its original environment.
 
 

3.4.2.5. Brain activation during REM sleep reflects the neural correlates of threat simulation. If the essence of dreaming is threat simulation, then we should find that the brain areas active during REM sleep are ones involved in generating emotional and perceptual experience.

According to Hobson (1999) PGO waves are believed to be the neural generators of the internal stimulation that results in dream phenomenology. They occur as bursts of waves during REM sleep, activating, in particular, the thalamocortical circuits involved in vision, but also radiating to the limbic lobe and amygdala. In the waking state PGO waves are triggered by strong, novel stimuli and are associated with surprise and fear: "PGO waves prepare us for fight or flight should these prove necessary. The startle reactions provoked in us by real or imaginary intruders are mediated by PGO-like signals" (Hobson 1999, p. 169). Thus, the function of PGO-waves during waking is clearly consistent with internal threat simulation during dreaming.

Research on emotionally charged memories and memory under stress has recently come up with the idea that there is a separable "hot" amygdala-centered emotional system distinct from the "cool" hippocampally centered episodic memory system (for a review, see Metcalfe & Jacobs 1998). The two systems work in cooperation, the "hot" system highlighting those species-specific or learned elements of memory traces that are highly emotional by nature. The "hot" system is believed to have a role in releasing species-specific behaviors such as fear or defensive responses to emotionally charged stimuli. As the stress levels of the organism increase, the "hot" memory system becomes increasingly activated. When a person is in a stressful and dangerous situation, the hippocampal "cool" system may not be optimal for responding to threat. Instead, the "hot" system may very efficiently process the threatening cues and immediately activate threat avoidance mechanisms. The "hot" system is considered to be more automatic and primitive than the "cool" system, thus allowing the organism to realize rapid protective responses. In accordance with this view, a recent PET study suggests that the human amygdala modulates the strength of conscious episodic memories according to their emotional importance (Hamann et al. 1999).

Recent functional brain imaging studies of sleep show that brain areas involved in the processing of emotionally charged memories are strongly activated during REM sleep and dreaming. The dream production mechanisms thus seem to be in close interaction with the primitive "hot" memory system, preferably selecting memory traces with high emotional charge. A study of regional cerebral blood flow distribution showed that during human REM sleep, activation and functional interaction occurs between the amygdaloid complexes and various cortical areas but the prefrontal cortices are deactivated (Maquet et al. 1996). The authors concluded that these interactions might lead to the reactivation of affective components of memories. A similar pattern was found in another study, concluding that pathways which transfer information between visual cortices and the limbic system are active during REM sleep (Braun et al. 1998).

In sum, neurophysiological studies and functional brain imaging reveal the dream production mechanisms at work during REM sleep, searching for and processing emotionally charged memory traces in the evolutionarily ancient, "hot" memory system. The dream production mechanisms, guided by the dominant emotional concerns of the dreamer, create the content of dreams in interaction with other long-term memory systems (Cavallero & Cicogna 1993) and perceptual cortical areas.
 
 

3.4.2.6. Summary. Many elements abundant in contemporary life (e.g. reading, writing) are absent from dreaming, whereas many such elements that are not common in waking life, but well consistent with simulating primitive threats (e.g. aggressive interactions with animals and male strangers), are universally present in adults’ dreams, children’s dreams, recurrent dreams and nightmares. Furthermore, brain activation during REM sleep is consistent with the activation of brain areas required to simulate emotionally charged threatening events.
 
 
 
 

3.5. PROPOSITION 3:
 
 

Real experiences of actual dangers or life-threatening events are very likely to be incorporated into dreams (Hartmann 1984, 1996; Barrett 1996). This is most clearly manifested in cases of post-traumatic nightmares. These are reported by people who have undergone, e.g. wartime battles, natural catastrophies, terrible accidents, or assault, rape or torture. The frequency of post-traumatic nightmares depends, among other things, on the degree of threat perceived to be targeted at self and significant others. It appears that the greater the sense of threat created by the experience, the more likely it is that nightmares will follow (Nader 1996). For example, 100% of 23 children who were kidnapped and buried in a truck trailer; 83% of six children that underwent a life-threatening medical operation; 80% of ten children witnessing their mothers being raped; 63% of children exposed to sniper fire on their schoolyard, and 40% of children whose suburb was exposed to radioactivity after a major nuclear power plant accident reported nightmares related to the respective incidents (for a review, see Nader 1996). Ninety-six percent of 316 Vietnam combat veterans described a combat nightmare in an interview (Wilmer, 1996).

These are very impressive figures, especially in view of the fact that laboratory research has failed to find any strong determinants of dream content. Presleep stimuli, such as films depicting violence, are only marginally if at all incorporated into dreams. The conclusions of Vogel (1993) in a review of stimulus incorporation in dreams, are revealing:
 
 

Once the dream production system encounters the memory of an event combined with a deep sense of threat, how does it handle that? There seems to be a more or less universal pattern involved in the ways in which post-traumatic dreams are constructed (Hartmann 1984, 1998). In the first stage, immediately after the traumatic event, the frequency of trauma-related dreams and nightmares increases and the event is being replicated a few times in the dream world, in a form often closely similar but not exactly identical (Brenneis 1994) to the original experience. The first stage normally lasts a few days or weeks, but in severe post-traumatic disorder it may persist for years^*8*. In a longitudinal study on children who were exposed to a sniper attack on the school playground, 42% continued to have bad dreams 14 months after the incident (Nader et al. 1990). In another study on a bushfire disaster, 18% of the children continued to have posttraumatic nightmares when studied 26 months after of the actual event (McFarlane 1987).

Gradually, the nightmares change into increasingly modified versions of the event. At this stage, the original experience is associated with and connected to other similar contents in memory. The resulting dreams may be small variations of the original threat, the original threat mixed with previously experienced ones, and with classical nightmare themes such as being chased or escaping powerful natural forces. Eventually, perhaps after a few weeks or months, the content of dreams returns to approximately normal. Even long after the original trauma, events that remind of it or also induce a deep sense of threat may trigger the recurrence of the trauma-related nightmares. In post-traumatic stress disorder, this normal development of the dream sequence does not occur, but instead, replications and different variations of the original trauma recur over and over again, even for years (Hartmann 1984, 1996; Stoddard et al. 1996).

Ordinary as well as post-traumatic nightmares are especially frequent in children (Hartmann 1984). Nader (1996, p. 16-17) mentions the following types of trauma-induced threat simulations in children’s dreams. Kuwaiti and Croatian children exposed to war dreamed of being personally endangered by someone trying to kill them with a knife, a gun, or bare hands, and of being captured or tortured. Children from L.A. who had witnessed their mothers being raped dreamed of the rapist returning, of being threatened, of being severely physically harmed, of directly confronting the assailant, or of taking revenge. A girl who was chased and groped by an unfamiliar man had recurrent dreams in which people or animals chased her. Children who were in a cafeteria when a tornado knocked the wall down with serious consequences dreamed of the wall falling again, of houses being destroyed by a tornado, of branches falling, of being hit by glass, and of trying to find bandages for dead people. After a hurricane, both parents and children dreamed of being threatened by winds or tornadoes coming directly at them.

In a study of 11 to 13-year-old Arab and Jew children’s dreams (Bilu 1989), all the dreams representing the "other side" were extracted and analysed. In these 212 dreams, aggression appeared in about 90% of the interactions, while friendliness was virtually nonexistent (4%). Jewish children dreamt about Arab terrorist attacks and camouflaged detonating explosives in public places. In these dreams, the dreamer was usually the recipient of an unprovoked assault initiated by an adult adversary who was typically defeated in the end. Arab children living in a refugee camp dreamt about brutal physical aggression which resulted in death on either side in 25% of the dreams. The dreamers were typically harrassed, expelled, arrested, beaten, injured or killed. Bilu (1989, p. 385-6) comments that the conflict between Arabs and Jews seems to have an even stronger presence in these children’s dreams than it does in reality (i.e. it is overrepresented in dreams): "...the intensity and pervasiveness of the conflict as reflected in the dreams cannot be taken for granted even by those well-acquainted with the situation."

Dreams after trauma reflect the dream production system working at full capacity, producing a regular pattern that proceeds from near-identical replications to increasingly modified variations to gradual fading and possible recurrence. Hartmann (1996, 1998) suggests that dreams after trauma should be seen as the paradigm case of dream formation. He makes the important observation that:
 
 

3.5.2. Real threats as cues that activate the threat simulation system

The view that emerges can be summarized as follows: Experiences of real threats are the only ecologically valid cues for the threat simulation system. Encountering real threats powerfully activates the threat simulation system: first, they may intensify the neurophysiological events underlying threat simulations, second, they tend to render the threat simulations more realistic, and third, they may even influence the development of the dream production system.

There is some evidence indicating that real threats may intensify REM sleep. In normal subjects the presence of stressful life events is associated with increased intensity of REM sleep (Williamson et al. 1995). One study (Ross et al., 1994) found that patients with posttraumatic stress disorder and frequent anxiety dreams showed elevated tonic REM sleep measures: they spent a higher percentage of total sleep time in REM sleep and their REM sleep periods were longer than those of control subjects. Furthermore, they had heightened phasic event generation in REM sleep: they manifested increased rapid eye movement activity. However, another study (Hurwitz et al. 1998) did not find any differences in polysomnographic sleep between Vietnam combat veterans and normal controls. Thus, more studies are needed to establish the relationship between stressful or threatening life-events and the intensity of REM sleep.

In a study on Palestinian children living in traumatic conditions it was found that the more the children were exposed to trauma, the more negatively emotional and the less bizarre their dreams were. The children exposed to trauma also had better dream recall than other children (Punamäki 1997). These findings indicate that the dream production system creates especially vivid and realistic simulations of threatening events encountered in the real world.

Real threats might even trigger the ontogenetic development of dreaming. Foulkes (1999) argues that adult-like "true" dreaming appears relatively late in childhood, between 7-9 years of age. By contrast, clinical case reports suggest that the earliest nightmares may be experienced as early as during the second year of life (Hartmann 1998). Some traumatized preschool children report fully developed nightmares, unlike those typical of the age (Nader 1996), suggesting that traumatic experiences may actually stimulate the development of the dream-production system, or, conversely, that a lack of real life-threatening events might hold it back, or at least preserve the dream production system in a resting state although it would already be capable of generating threat simulations if only exposed to the ecologically valid cues. (See also Note 5 on children’s dreams).

It could be argued against Proposition 3 that both positive and negative real emotions are equally strong in activating the dream production system. This alternative hypothesis is not supported by evidence. Hartmann (1998, p. 73) observes that "... even when people experience a happy event, they are more likely to dream about problems associated with it than the pure happiness of the event itself." Thus, dreams tend to represent even happiness in the light of the possible threats that might endanger it.

According to the present hypothesis, the brain’s dream production system selects traumatic contents not because they represent unsolved emotional problems, but primarily because such experiences mark situations critical for physical survival and reproductive success. What from a psychological point of view is a "traumatic experience" is, from a biological point of view, an instance of threat perception and threat avoidance behavior. Negative emotions, such as fear and terror, accompanying the perception of serious real-life threats, serve to label such events as critical to one’s own survival and future reproductive success. The contents of the threat simulations are selected by the dream production system from long-term memory, where recent memory traces associated with threatening emotional impact are the most salient ones to enter the dream production mechanisms. The stronger the negative emotional charge, the more threatening the situation is likely to have been, and the more likely it is that it will be selected by the dream production system as a recurrent theme for threat simulation. The dream production system is highly sensitive to situations critical for the physical survival and future success of the individual: violent attacks, being chased by strangers or animals, finding intruders in one’s private territory, losing valuable material resources, being socially rejected, encountering untamed natural forces or dangerous animals, being involved in accidents and misfortunes. Such dream contents involve, from a biological point of view, threat perception, threat avoidance, antipredatory behavior and coping strategies against threats.
 
 
 
 

3.6. PROPOSITION 4:
 
 

So far we have shown that dreaming specializes in the repetitious simulation of threatening events. Next, we need to show that these events constitute realistic rehearsals of threat perception and avoidance, for otherwise they would not be useful simulations. First of all, dreams and especially nightmares consist of vivid images that seem perfectly real. Second, during dreaming we are in an uncritical, delusional, isolated state of mind which very efficiently prevents us from realizing that it is all just a hallucinatory simulation (Rechtschaffen 1978). The relatively rare exception of lucid dreaming (Gackenbach & LaBerge, 1988) notwithstanding, we take the dream world for real while it lasts, totally lacking insight into our true condition. Thus, these two factors, perceptual realism and delusional lack of insight, guarantee that the simulation is taken most seriously. If that were not the case, we might instantly recognize the dream world for what it is and not be motivated to defend ourselves against the simulated threats. Lucidity has in fact been recommended as a possible cure for recurring nightmares (e.g. LaBerge 1985; Zadra & Pihl 1997).
 
 

3.6.2. Motor Realism.

What about the dreamed action: what is its relationship to real motor behavior? It should be neurally realized in the same way as real actions are, otherwise it could not be regarded as an efficient rehearsal of what to do in a comparable real situation. Mental imagery of motor actions uses the same motor representations and central neural mechanisms that are used to generate actual actions (Decety 1996; Jeannerod 1995); moreover, dreamed action is experientially far more realistic than mere imagined action. Therefore we have good reasons to believe that dreamed action is equivalent to real action as far as the underlying brain mechanisms are concerned.

Classical neurophysiological studies in the 1960's (reviewed by Hobson 1988) showed that the pyramidal-tract cells of the motor cortex increased their firing during REM sleep (compared to NREM), having firing rates as high as those during waking with movement^*9*. Thus, motor commands are generated during REM sleep at the cortical level but they are not realized in the periphery because of the operation of an inhibitory system that blocks the activity of motor neurons in the spinal cord, resulting in muscular atonia. According to Hobson (1999), the experience of movement in dreams is created with the help of the efferent copying mechanism, which sends copies of all cortical motor commands to the sensory system. The brain thus receives internally generated information about issued motor commands and computes the expected consequences of those commands. The sensory system is not informed that these commands were not in actual fact carried out by the muscles, and therefore the illusion of movement comes about.

If the inhibitory mechanisms that produce atonia during REM sleep are malfunctioning, the result is a recently described sleep disorder called REM Sleep Behavior Disorder (RBD) (Schenck et al. 1986). These patients manifest violent behaviors during REM sleep, which are the acting out of the motor imagery they are dreaming about. Thus, dreamed action corresponds to real action as far as the forebrain is concerned. The difference between dreamed and real motor action depends only on the inhibitory cell groups in the pons. Thus, within the forebrain, dreamed action has the same neural realization and the same causal powers as real action does. Dreamed action is experientially and neurophysiologically real. (For a similar view on all motor imagery, see Jeannerod 1994).

Some illustrative cases of RBD have been reported in the medical literature. Dyken et al. (1995) describe the case of a 73-year old man. During an episode of RBD, the patient leaped from his bed, fell, and struck the right side of his face on a corner of a chest, awakening him immediately. This resulted in subdural hemorrhage. He had dreamed of working on a loading dock and saw a man running. Someone yelled "Stop him!", and the patient had tried to do just that when he jumped out of his bed with the unfortunate consequences. In the sleep laboratory during REM sleep, the patient suddenly exhibited explosive running movements, followed by an arousal. The patient’s actions again clearly corresponded to what he was dreaming about at the time of the observed behaviours. Boeve et al. (1998, p. 363-370) describe a patient who, on one occasion, "held his wife’s head in a headlock and, while moving his legs as if running, shouted: ‘I’m gonna make that touchdown!’ He then attempted to throw her head down toward the foot of the bed. When awakened, he recalled a dream in which he was running for a touchdown, and he spiked the football in the end zone". Comella et al. (1998) describe a group of patients with RBD. If these patients were awakened during an episode of abnormal sleep behavior, none of them realized that they had executed violent movements, although all recalled violent dreams at the time of awakening: being pursued by an enemy; trying to protect family members from unknown intruders; or fighting off unidentified assaillants. Schenck (1993) describes a patient whose EEG, EMG and EKG were polysomnographically recorded during an attack of violent behavior. The muscle tone was increased and the arms and legs showed bursts of intense twitching, accompanied by observable behavior. After a spontaneous awakening, the man reported a dream in which he was running and trying to escape skeletons that were awaiting him.

It is noteworthy that most cases of RBD involve intensive threat simulation dreams, and the behaviors manifested are (mostly adequate) responses to these threats. It may be that threat simulations are associated with increased cortical activation, leading to intensive motor imagery that breaks through the malfunctioning inhibitory mechanisms.

There are other sleep disorders that can be interpreted as an inappropriate activation of the threat simulation system, leading to sleep-related behaviours. Night terrors^*10*, sleepwalking (somnambulism) and nocturnal wandering appear to be, at least in some cases, threat simulations that take place during NREM sleep and lead to an altered state of consciousness, a mixture of wakefulness and NREM sleep (Mahowald & Schenck 1992, Mahowald et al. 1998). In this state, one’s subjective consciousness is focused on one internally generated, usually terrifying image or belief. Appropriate threat avoidance behavior is often realized automatically, violently and efficiently, in the absence of reflective thought, without an awareness of one’s altered state, one’s actions, or their actual consequences^*11*. One patient described by Schenck & Mahowald (1995) once left the house in pajamas by running through a screen door, entered his automobile and drove eight kilometers to his parents’ home where he awakened them by pounding on their door. This episode of somnambulistic automobile driving was initiated by the subject’s belief that someone was in the house and about to attack him. Another subject with nocturnal wandering once threw his wife on the floor, ran to his two children, took them into his arms and ran outside. He afterwards said he had believed that the house was on fire (Guilleminault 1995). When aged ten, one patient had risen from sleep, rushed into the sitting room where his parents were still sitting, and thrown the butter dish out of the window, believing it to be a bomb (Oswald & Evans 1985).

Thus, both RBD and NREM-related sleep disorders show that threat simulation during sleep includes realistic and adequate motor activation in the brain in response to the perceived threats.
 
 

3.6.4. Summary.

The evidence reviewed above shows that dreaming constitutes a realistic simulation that we tend to believe without questioning and that dreaming about an action is an identical process for cortical motor areas as actually carrying out the same action. In some pathological cases, the actions generated in the dream world are inadvertently performed in the real world. Thus, to dream about threat perception and threat avoidance behaviours is to realistically rehearse these functions in a safe environment.
 
 
 
 

3.7. PROPOSITION 5:
 
 

It is a commonplace that training and repetition lead to enhanced performance. However, can actions only performed at the phenomenal level and not overtly executed lead to any kind of learning? Research on the effects of motor imagery and mental training to motor performance show that repeated motor imagery can lead to increased muscular strength (Yue & Cole 1992), improvement in the learning of new motor skills (Yaguez et al. 1998; Hall et al. 1992), and improved performance in sports (e.g. Lejune et al. 1994). These learning effects are thought to arise at the cortical programming levels of the motor system (e.g. through activation of Brodmann area 6 where the premotor and the supplementary motor areas reside), not from neural changes at the execution level (Yue & Cole 1992; Jeannerod 1994, 1995). Since even motor imagery and mental training can have these effects, there is every reason to believe that the intensive and thoroughly realistic motor imagery in our dreams can also lead to similar effects. Thus, repeated simulation of threat avoidance behaviors should lead to enhanced threat avoidance skills by increasing the efficiency of the programming and execution of motor activity required in the responses to perceived threats.
 
 

3.7.2. Implicit Learning and Implicit Memory

There is one difference, however, between "mental training" and dreaming: we do not explicitly remember the learning and training episodes, nor do we have any idea of what the skills we are training in our dreams really are. Thus, doubts may be cast on whether it is possible to learn something in the absence of an intention to learn and memory of what one has learnt. Extensive literature on implicit learning, however, confirms that many skills important for human performance are in fact learned without any conscious access to their nature (for reviews, see Berry 1994; Lewicki et al. 1996; Cleeremans et al. 1998). A person may have no idea that she uses certain types of acquired knowledge when performing a certain task. Even amnesic patients can learn motor skills despite their inability to remember having ever done the task before: their performance becomes faster and more accurate, showing implicit skill learning in the absence of any conscious memory of the learning episode. Furthermore, amnesic patients can have implicit memory also for emotional experiences that they cannot remember explicitly (Schacter 1996; Glisky & Schacter 1989, 1990). Therefore, like any other skills, also threat avoidance skills may be learned and rehearsed without explicit access to what has been learned.

Implicit learning is very sensitive to correlations and covariations between different features of perceived objects. If two features are associated in our experience a few times, an initial coding rule can be acquired which biases perception to detect both features whenever only one of them is directly perceived (Lewicki et al. 1996). Dream experience might bias waking perception in the direction where certain perceived features are automatically associated with certain other ones in order to be prepared for possible threats. Furthermore, we are predisposed to learn certain reactions to certain stimuli. Especially stimuli that reflect ancient threats easily come to be feared (Marks and Nesse 1994).

If the function of dreaming is realized through implicit learning and memory, then we should predict that REM sleep deprivation has a detrimental effect on tasks requiring implicit but not explicit memory. This is what in fact has been found: Smith (1995) reports that memory for explicit tasks is not affected by REM sleep loss, but memory for procedural or implicit tasks is impaired by REM deprivation.
 
 

3.7.3. Summary.

I conclude that rehearsing threat avoidance skills in the simulated environment of dreams is likely to lead to improved performance in real threat avoidance situations in exactly the same way as mental training and implicit learning have been shown to lead to improved performance in a wide variety of tasks. It is not necessary to remember the simulated threats explicitly, for the purpose of the simulations is to rehearse skills, and such rehearsal results in faster and improved skills rather than a set of explicitly accessible memories. Furthermore, REM sleep physiology appears to selectively support implicit, procedural learning.
 
 
 
 

3.8. PROPOSITION 6:
 
 

So far we have shown that dreams are specialized in threat simulations, effectively triggered by real-life threats and engaging the appropriate cognitive and neural mechanisms in ways which have been shown to lead to improved performance in other learning contexts. In order to complete the argument, we now need to show that the human ancestral environment was the kind of place that contained the relevant ecologically valid cues for constantly activating the threat simulation system, and that there was likely to be a selectional advantage from improved threat avoidance skills so that repeated threat simulations were likely to lead to increased reproductive success.

We need to show, firstly, that there was a high selectional pressure in the ancestral environment. How long did people live in those conditions? Which proportion of the population survived to reach the reproductive age? As far as we know, mean life expectancy was remarkably low compared with that of modern times, only 20-25 years. According to one estimation (Meindl 1992), of those who reached five years of age in ancestral hunter-gatherer populations, about 25 % died before entering the reproductive period and about 70% died before completing it. Thus, mortality rates were high, and only the selected few ever got the chance to reproduce successfully.

Secondly, we need to show that the real threats in the ancestral environment were the kind of events the threat simulation system is good at simulating. What were the most likely threats to survival in the ancestral environment? How severe were they? Among the primary causes of death in hunter-gatherer populations have probably been exposure to predation by large carnivores, exposure to the elements, infectious disease, poor conditions and risky activities during hunting and gathering, and aggression or violent encounters especially in defence of personal resources or group territories (Landers, 1992; Meindl 1992).

These estimates render it quite obvious that the life of an average ancestral human was constantly at stake in the original environment^*12*. The death or serious injury of close relatives and local group members must have been a not uncommon event. Confrontation with extremely dangerous or even life-threatening situations is likely to have been part of everyday life rather than a rare exception. In order to reproduce successfully in those conditions, an individual must have been quite skilled at perceiving and recognizing various threatening situations (e.g. predators, aggressive strangers, poisonous animals, natural forces, social rejection by own group members), at avoiding unnecessary dangers, and when a threatening situation could not be avoided, must have been able to cope with it by using efficient cognitive and behavioral strategies that promote survival.
 
 

3.8.2. Activation of the threat simulation system in ancestral conditions.

The key question is: What was the dreaming brain dreaming about in those circumstances? In view of the extremely harsh conditions in which our ancestors lived, it is likely that every individual was continuously rather severely ‘traumatized’, at least by modern standards. Therefore, their threat simulation systems must have been repeatedly activated by the ecologically valid cues from threatening real-life situations, resulting in a continuous flow of threat simulation dreams. In effect, the dream production system must have been in a more or less constant post-traumatic state. In fact, that probably was the normal state of the system then, although we who mostly live free of immediate threats to physical survival have come to regard it as a peculiar pathological state.

As Tooby and Cosmides note (1995, p. 1190), natural selection retained neural structures on the basis of their ability to create adaptively organized relationships between information and behavior, e.g. the sight of a predator activates inference procedures that cause the organism to hide or flee. Threat simulation rehearses and improves performance in processing exactly such organized relationships, specifically between information interpretable as a threat to survival and efficient cognitive-behavioral procedures which need to be activated in response to such information. In the light of our present knowledge, it seems very likely that the dream production system had more than enough threatening experiences to work with in the human ancestral environment. Therefore it was likely to simulate realistic threats thousands of times during an individual’s lifetime, which was bound to result in improved threat avoidance skills. Individuals with improved threat avoidance skills were more likely to leave offspring. Since the neural basis of the dream production mechanisms is innate, dreaming came to be selected for during our evolutionary history. Individuals without the threat simulation system would have been in a disadvantageous position, and would have been selected against in the ancestral environment. Now that most humans live in environments far removed from the ancestral ones, and face threats completely unlike the ancestral ones, it may be that the threat simulation system is not properly activated or not able to construct useful simulations of most of the present-day threats. But dreaming still is an important part of universal human experience, and its persistence and universality can now be explained by referring to the advantages in threat avoidance it provided our ancestors with.
 
 
 
 

4. THE DREAMS OF HUNTER-GATHERERS AND ANIMALS
 
 

4.1. Threat simulation in the dreams of contemporary hunter-gatherers
 
 

If, as we have argued, the dreaming brain is a phylogenetically ancient threat simulation system with default values reflecting ancestral rather than modern conditions, then we should expect to see this mechanism naturally activated in individuals who live in conditions closely resembling the ancestral ones. We should predict high levels of survival themes, threat simulation and animal characters in the dreams of such individuals. Fortunately there are some studies of dream content in hunter-gatherer populations. Dreams from the Yir Yoront, an aboriginal society in Australia, were collected in the 1930's and later analysed by Calvin Hall. Some of the results have now been published in Domhoff (1996). Compared to American males, the Yir Yoront males dream significantly more about animals, have a higher proportion of aggression with animals, and a very high percentage of physical aggression. They also often dream about sharing meat from the animals they have killed.

Gregor (1981) reported a content analysis of 385 dreams collected among the Mehinaku Indians in Central Brazil whose life had remained essentially traditional at the time of the study. The Mehinaku are an exceptionally fruitful group of informants for a study on dreams, for they have the habit of carefully recalling and often recounting their dreams in the morning. Gregor found that the dreams of the Mehinaku contain significantly more physical aggression, especially with animals, than dreams from the American normative sample. However, gender differences are similar in Mehinaku and American dreams: there is more aggression in men’s than in women’s dreams, and women are more likely than men to be the victims of aggression. The most frequent attackers are men and animals. There are many themes in the Mehinaku dreams that could be interpreted in the framework of evolutionary psychology: For example, women often dream about being the targets of sexually violent men, and men dream about having sex with women other than their own spouse, and consequently being attacked by jealous male rivals or angry female lovers.

In his paper, Gregor (1981) provides short summaries of the Mehinaku dream reports. Here are some examples of typical threats in the Mehinaku men’s dreams:
 
 

A woman attempted to have sex with him, the jealous husband assaulted him;

Lost his belt and could not find it;

Desired and approached girl, struck by his jealous wife;

Attacked by a jaguar;

Stung by wasps while in woods;

Stung by ant;

Chased by snake, he turns and kills it;

Daughter almost drowned, rescued her;

Stung by bees;

Had sexual relations with girl friend, wife saw them and became angry;

Rescued drowning brother;

Attacked by a herd of wild pigs;

Shot at threatening jaguar but missed;

Killed a threatening snake.
 
 

The prediction from the threat-simulation hypothesis is that threatening events are overrepresented among dream events, and that non-threatening, peaceful activities are underrepresented. In accordance with this prediction, peaceful and realistic non-threatening and non-aggressive activities [e.g. "Went to river and saw birds"; "Worked in the forest"; Watched as the sun rose"; "Went to the garden"; "Went to the field to get corn"] make up only about 20% of the 276 dreams reported by the Mehinaku men. In contrast, about 60% of the dreams have a threatening situation as a theme^*13*. Even if their waking lives contained more threats than ours, it is unlikely that sixty percent of their waking time would consist of overtly threatening episodes; for that they would have to spend no less than almost 10 hours per day in situations involving threats (i.e. 60% of the total estimated waking time of 16 hours). Therefore, the prediction holds in the Mehinaku dreams: dream production mechanisms selectively overrepresent threatening events and underrepresent peaceful activities.

Dream samples from contemporary hunter-gatherer groups are probably as close to ancestral dreams as it is possible for us to get, which is not to say that they should be identical. In any case these studies show that threat simulation is very frequent in the dreams of such individuals, and that the dream production system tends to generate fairly realistic threat simulations when the world it simulates is not very dissimilar from the ancestral human environment.
 
 

4.2. Dreaming among other mammals: evidence for the rehearsal of survival skills.
 
 

This reinterpretation of the function of dreaming is consistent with the inferences we can make concerning possible dream contents and the function of dreaming in other mammals. Although we cannot know with absolute certainty that other mammals have subjective experiences during sleep, we do know that they can manifest remarkably complex behaviors during REM sleep. In humans the comparable condition is the acting out of dream experience (REM sleep behaviour disorder, see section 3.6.2. above). Therefore we may assume that to the extent these animals have conscious phenomenal experience when awake, they are likely to have similar experiences, i.e. dreams, when in REM sleep^*14*.

REM sleep without atonia induces complex species-specific behaviors in the cat, for example motions typical of orienting toward prey, searching for prey and attacking (Morrison 1983). In several species of mammals the hippocampal theta rhythm is associated with behaviors requiring responses to changing environmental information most crucial to survival: e.g. predatory behavior in the cat and pray behavior in the rabbit (Winson 1990). The theta rhythym disappears in slow-wave sleep but reappears in REM sleep. Winson (1990, 1993) suggests that information important for survival is accessed during REM sleep and integrated with past experience to provide a strategy for future behavior. Thus, there is empirical evidence that also in other mammals the dreaming brain rehearses species-specific survival skills, consistent with the present hypothesis that the human dream-production system is primarily a threat simulation system^*15*.
 
 
 
 

5. TESTABILITY AND PREDICTIONS
 
 

We can now summarize the central claims of the threat simulation theory of dreaming, all of which are supported by the available evidence, and present some testable empirical predictions:
 
 

1. Dream consciousness is an organized and selective simulation of the perceptual world.

In order to disconfirm the threat simulation theory (or some part of it), it must be shown empirically that the above predictions are false. If it can be shown e.g. that dream generation is a truly random physiological process (as stated by several theories), or that even experiences completely different from threat-related ones regularly lead to intensive, recurrent simulations, or that there are cultures in which threatening experiences do not lead to threat simulations and nightmares, then the threat simulation theory is in serious difficulty^*16*.

As an evolutionary hypothesis, the threat simulation theory of dreaming concerns historical events, and the historical events themselves of course cannot be subjected to observation or experimental manipulation. But it would not be correct to say that e.g. theories on what caused the mass extinction of the dinosaurs 65 million years ago, or on why Asia has got the Himalayas, are not empirically testable just because the original events cannot be observed or experimented on. Therefore, the threat simulation hypothesis is open to empirical testing, confirmation, and disconfirmation to the same extent as any other hypotheses regarding the causal mechanisms at work in the past, leading the natural world to be as it is in the present^*17*.
 
 
 
 

6. THREAT SIMULATION AS A BIOLOGICAL DEFENSE MECHANISM
 
 

This section summarizes and clarifies how the threat simulation mechanism is hypothesized to operate in dream production, and in what sense this operation can be regarded as biologically functional.
 
 

6.1. Dream production and threat simulation
 
 

Dream production is an automatic, hard-wired, regularly activated feature of human brain function. The sources of dreams are selected from long-term memory by reactivating and recombining memory traces that are the most salient for the dream production system. Saliency is a function of at least two factors: the degree of threat-related or negative emotional charge and the recency of the encoding or reactivation (or other priming) of the memory traces. Therefore, the most salient memory traces for dream production consist of the ones encoding the most threatening events most recently encountered (or whose memory traces have been otherwise most recently reactivated). The saliency of a set of memory traces gradually declines by time or may be overcome by that of other traces: memory traces compete for access to dream production through their saliency.

Simulations including elements of the selected threatening memories are then reconstructed by the dream production mechanisms. In this process the dream production mechanism tends to use dream settings and stereotyped scripts that are compatible with threats similar to ancestral ones (composing events that involve e.g. attacks, fights, pursuits, escapes, intrusions, losses of valuable resources, and events during which the dream self or close kin are endangered). Typical threat simulation dreams, such as nightmares and recurrent dreams, are thus composed of a variable mixture of salient episodic memory traces and suitable threat simulation scripts. This mechanism is biologically functional (i.e. it solved adaptive problems for our ancestors) because in the original environment the dream production system regularly generated simulations of such real events that directly or indirectly threatened the reproductive success of ancestral humans.

Simulating these events rehearses performance at two stages: Threat recognition and threat avoidance. The simulation of threat recognition is supposed to proceed in the following way. Salient emotionally charged mem