The Evolution of the Self

This paper is about the biological and cognitive origins of the self. The target papers in the JCS special issue ‘A new approach to studies of the self’ give insights into the experience of the self and release from self. But some scientific questions remained unanswered, notably: What is the self? Why is it so destructive in its impact on our daily lives? If release from the ego-bound self is so beneficial, why have we not evolved to be naturally free from self? Why do mindfulness and spirituality need to be cultivated – why are they not just a normal part of the human psyche?

These questions can be posed in more traditional forms: What is the cause of our fall from grace? or: What is the origin of evil in mankind? Religions have struggled to answer these questions, but have not done so. Finding a convincing answer could, to say the least, be important for mankind.

An answer can now be found, from evolutionary biology.

In brief: the human self is something not found in any other animal brain (with a very few possible exceptions). It is a symbol used in a set of simple, automatic (and often destructive) computations in the brain; a set of knee-jerk reactions to our social situation, a little clockwork computing engine which never stops. This engine has evolved in mankind as a deliberate handicap – as an ‘honest signal’ of fitness, through forces of sexual selection. Because these computations are unconscious, we do not observe them directly. But we can observe their effects at all times; and in doing so, we can become more free from them.

In more detail: Animals differ from plants by having muscles and the ability to move. Their brains exist primarily to control their movements – to ensure their survival by feeding, finding shelter, and reproducing. To do this, brains compute. In nearly all respects, to ensure its owner’s survival, the brain’s computations must be truth-seeking; they must represent the real world as accurately as possible. Brains need to be truth-seekers, because that is the best way to survive. For instance, as you look around, your conscious awareness is a model of the space around you – and it is a remarkably faithful model of that space. You need that faithful model to plan and carry out your every movement. If it was a distorted model, you could not stand up, swing on branches, pick fruit, or do any of the things a primate needs to do; or rather, you could not do them so well.

The truth-seeking propensity of brains can now be understood mathematically, through the theories of Bayesian inference and evolutionary optimisation. Brains seek out the most truthful account of the world from their noisy, incomplete sense data, by using Bayesian inference – because, as can be proved, Bayesian inference does the job best. Anything else would give a less accurate internal model of the world, and give lower fitness – and so would have been displaced by natural selection. Bayesian inference is now the mainstream approach to studying computation in the brain [e.g. Rao et al, 2002].

However, our brains are not truth-seekers in all respects. Why do they sometimes depart from the optimal-fitness, truth-seeking behaviour? To understand why, we need to look at other animal traits which have evolved in spite of an obvious cost in fitness to their owners. The paradigm case of this is the peacock’s tail.

Male peacocks have evolved to have large, ludicrous tails; and female peacocks have evolved to find them attractive. Both sexes are stuck with this, in spite of its obvious costs in fitness to the peacock species. A male with a short tail would be unattractive to females; and a female who fancied males with short tails would have unattractive children. How this happens is now understood, using the mathematics of evolutionary game theory [Maynard Smith, 1982]. The result is that sexual selection – a conspiracy of evolution between the sexes – is not a process of convergent improvement leading to greater fitness. It is a runaway, positive feedback process of evolution of both sexes, whose results are (a) locked in to specific species; (b) partly arbitrary and unpredictable; and (c) a real handicap to individuals, because they must be an ‘honest signal’ of superior fitness, indicating a good prospect for a mate.

The results of sexual selection can be seen in almost any species – in riotous and ostentatious plumage in birds, in male mating displays, in choices of breeding sites, in bizarre courtship rituals. These all illustrate the arbitrary, species-specific, and fitness-reducing results of sexual selection. The peacock’s tail is just an extreme case. Without sexual selection, biology would be a much duller subject.

How has sexual selection played out in the human species? Where peacocks have ludicrous large tails, we have ludicrous large brains. The most prominent overt ‘display’ for our large brains is language; no other animal can express unlimited numbers of sensible or nonsensical propositions. Language is a result of sexual selection, which started from primate social intelligence. It is a key part of the over-extended social intelligence, which is the peacock’s tail in our brains.

In order to succeed in sex, any primate needs to attain high status in their group. This is not just a matter of brute strength, because alliances with one’s peers can help to overcome a stronger rival; so a powerful social intelligence is needed to understand a shifting social scene of rank, kinship, alliances, social signals, and tokens of friendship [Worden 1996]. However, most primate social intelligence, while complex, is still truth-seeking – not constructing any more, in its model of the social world, than there is evidence to believe. This was the starting point for the runaway growth of human social intelligence, through sexual selection.

Because the course of sexual selection is somewhat arbitrary – arising as it does from a runaway positive feedback of evolution between the sexes – there is no single true ‘Just-So-Story’ for the evolution of sexually selected traits. Their evolutionary history, like the resulting traits, is somewhat arbitrary – in which small random changes are reinforced by positive feedback (like a pencil balanced on its point – you cannot say which way it will fall, because any small movement produces a positive feedback). All we can say is that human social intelligence has now gone well beyond the social intelligence of any other primate, in a cluster of related respects:

  • Complex internal ‘script’ structures, describing the development of social situations over time
  • Discrete symbolic tokens for the elements of those scripts
  • A special token representing ‘self’ in scripts (most primates do not need this token)
  • Fast computations to combine scripts to work out the consequences of new information
  • A Theory of Mind, to represent (and infer or guess) what others think – especially what they may think about oneself
  • A system of ‘self-esteem’, to infer one’s own likely future status from what others might think about the ‘self’ token
  • Complex learned emotions , driven by expected script outcomes and the urge to maximise self-esteem
  • A constant concern for one’s own emerging life story and status of the ‘self’ token; no outcome ever being good enough to reassure us

This cluster works as a little automatic computing engine (I picture it as a clockwork motor), which goes on in our heads night and day, which makes us play the human social game, and which drives our emotions. The social game, being the result of sexual selection, is somewhat arbitrary in its rules. We keep score of the game with ‘counters’ of self-esteem, measured by what we think other people will think of our lives, through some strange and arbitrary criteria of status. The computational engine which plays this game is not a truth-seeker; for instance, it makes up propositions about ‘what others think’ which have no basis in fact or evidence. Its purpose is to play an arbitrary game, not to seek the truth about the world.

We could easily build a simple computer program to model our social computing engine, which would do a good job of mimicking its knee-jerk reactions to everyday events, its type-casting of others, its spurious life histories, and its negative comparisons of the self with others – all the core features of our habitual dualistic thought, and of the social game which determines our status and reproductive success. The ‘self’ is a central symbol in this cluster of automatic computations, and it is little more than that. The symbols for other people are caricatures,  like wooden chess pieces in our game.

This cluster of computations – whose results we can observe in ourselves every day – already goes well beyond the social intelligence of most primates. This cluster may already have been a result of sexual selection, as our ancestors chose mates based on assessment of their social game-playing ability (and damn the impact on the childrens’ fitness!). These computations already constitute the ‘real handicap’ required for sexual selection, in the personal unhappiness and alienation caused by a constant gnawing concern for the ‘self’ token and its status, as imagined in the minds of others; the costs of competing rather than cooperating in the group; and the metabolic costs of an over-sized brain.

However, we have not yet mentioned the pinnacle of human social intelligence, which is our complex language. All the evidence points to language being a part of our social intelligence. As soon as any language was possible – even in sentences of a few words, which we can now train chimps to use – then superior ability in language could immediately become most obvious signal of greater social intelligence. The stage was set for a runaway evolution between the sexes, of the kind: “Use elaborate language; find elaborate language attractive in a mate”. Our language capacity today is so prodigious, and has evolved so fast, that it must have been the result of some runaway, positive-feedback process such as sexual selection [Worden 1995]. Better group hunting, or other practical advantages in the wild,  are not big enough benefits to account for complex language. The big advantage lies in getting a mate.

As language evolved to have greater power, it would have dragged other parts of social intelligence along behind it. For instance, to use language effectively, you need to make good guesses about what others know and do not know – so you know what is worth saying to them, and what can be left out. This drives the evolution of our Theory of Mind – to the point where our mental models of ‘what others think’ may go far beyond the evidence. This in turn reinforces the tendency to dwell on ‘what others think of me’ in the extended life story and the struggle for self-esteem, with all the negative emotions that they entail. Also, more fluent language requires faster unconscious pattern-matching of scripts – including the ‘self esteem’ scripts which drive our emotions. Language has supercharged our social computing engine.

These evolutionary changes need not be in any positive direction which is good for the individual, or which increases individual fitness. On the contrary – they are expected to take some arbitrary negative direction which is bad for the individual, and to increase the individual handicaps which are the ‘honest signal’ of general fitness, required for sexual selection. The social game we play can be arbitrary, as long as both sides have evolved to agree the same rules, and it has real costs to play it.  The message to potential partners is: “I have this ludicrous handicap which you find attractive; but still I am fit enough to win you and mate with you.” That is how sexual selection works.

To describe some of the ways in which the social game, and the computer in our heads which plays it, are arbitrary and not truth-seeking:

  • The scoring for the game, which the computer seeks to maximise, consists not of getting enough to satisfy one’s own needs, but of getting the maximum amount. The highest scores, consisting of high self-esteem, are not delivered by “I have enough of this now”, but by “I want more, more, more”.
  • Self-esteem points are determined not by comparing oneself and one’s actions against any objective external criteria, but by estimating what one thinks other people think about oneself, using the Theory of Mind facility.
  • The Theory of Mind, whereby we compute the mental states of others, goes far beyond any evidence available to it to determine their mental states, and so is highly unreliable – particularly in respect of what we think other people think about us
  • The things which count for scores in the game are only loosely related to our actual needs. Usually we find things valuable not because we really need them, but because we think other people find them valuable (e.g. we value large collections of pieces of paper with identical intricate designs on them).
  • The computations have a large asymmetry between the symbol for ‘self’ and the symbols for others – in spite of the fact that the self and the others are all individuals of the same species, whose similarities outweigh their differences. The same behaviour can be found acceptable in the self and reprehensible in others.
  • Other people are generally represented by simple tokens with a few key attributes, clustered around the self-esteem scoring system, and taking little account of their actual diversity.
  • Language is a highly productive tool for framing and remembering complex propositions about the world and the people in it. But it is in no way matched by our capability to check the truth of those propositions. So our system of language-based thought, like our Theory of Mind, is highly unreliable. Often we believe propositions for no other reason than that we think other people believe them. These two unreliable facilities, language and the Theory of Mind, cling together and lead us far from truth.
  • Behaviour patterns which are clearly seen to be reprehensible in other people (such as greed, envy, and other sins) are often the first recommendations of the social computer for our own actions
  • In assessing our level of success in the self-esteem stakes, the social computer is programmed and biased always to make us compare ourselves unfavourably with other people, real or imagined – so as to keep us competing for higher status.
  • One of the scoring tokens in the social game is long-term security; our craving for more of it leads to permanent anxiety about imagined outcomes far in the future which are, in reality, impossible to predict.

These  errors and distortions have evolved in our brains by sexual selection. We all need to play the same social game as other people, in order to reproduce. This is an evolutionary positive feedback, an unstable process which produces arbitrary, quasi-random results. The game does not have to be beneficial to us as individuals. Quite the opposite; it has to be a handicap, to serve as an ‘honest signal’ to potential partners of our general fitness.

This, then, is the evolutionary origin of the self, and of its harmful effects on our emotional lives. The harmful emotional effects of the self are not a mystery; they are the inevitable effect of sexual selection, which has to impose real handicaps on individuals.

To say some more about how our social computing engine works, and how we become aware of its workings: it works by rapid unconscious pattern matching of symbolic life-story ‘scripts’ against our current social situation, using a large number of conditioned emotional scripts, which we have learned since childhood. The ‘self’ symbol figures in nearly all these scripts. Pattern-matching of scripts happens  very fast (as it does also in language understanding, which needs to be very fast) so it usually ‘kicks in’ faster than conscious thought. The script patterns are crude symbolic patterns, with little specific detail. They often lead to ‘Theory of Mind’ results, implying that others will think more or less of us, based on how we guess they will see our actions and judge our lives. Because the engine has evolved to keep us forever competing for more status, the resulting emotions are more negative than positive.

All these are unconscious computations, so we are not directly aware of them. We become aware of them though emotions – bodily feelings and imagined postures and movements, and by other conscious mental images. Emotions may be sudden waves of feeling, or background moods. Then, as the social situations trigger slower, language-based conscious thought, those thoughts are influenced the unconscious scripts, and they may provide conscious justification for what we have emotionally decided to do. Often they involve thoughts about oneself, particularly about how others may regard oneself, along the lines of: “How do I look now?”. Any conscious rejection of the results  of scripts does little to stop them  matching again, with the same emotional impact; we cannot ‘unlearn’ our emotion scripts just by trying to.

Emotion scripts require little detail from current reality, but have developed to cascade automatically, triggering further scripts to take us along habitual emotional railway lines. For this reason, they need to divert our minds away from close attention to the present moment. Noticing new things in the moment might take you off your script sequence; living fully in the moment is not needed in order to play the social game, and could divert us from it.

That is the picture of our ordinary un-mindful minds – which occupy most of us for most of our days. It is why, to quote Cynthia Bourgeault, ‘millions on this planet unsuspectingly sleep their lives away’, allowing their lives to be directed by a little unconscious computer in their heads. This is the evolutionary origin of the ‘egoic operating system’ she describes, and of the self which all spiritual traditions have described, and propose ways to escape from it.

If the self results from an unconscious automatic computation in the brain, which carries on whether we want it to or not, what is release from self, and how can it come about? Is release from the self and its negative emotions even possible?

As the target papers of this issue attest, release from the ego-bound self can happen. There is a set of mindfulness techniques, starting with meditation, which work – although it is  hard to say in words just what ‘work’ means. Some schools of mindfulness, such as  Zen, assert that a permanent, complete release from the self is possible. All paths to mindfulness must address its inbuilt paradox – that any deliberate attempt to find release from the self can turn into just another goal feeding the self, which will then control us in more subtle ways. This may be the root of John Welwood’s ‘spiritual bypassing’.

I do not know where these ideas will lead. But we live in a wonderful time, when scientific understanding of the mind and its evolution is starting to converge with ancient wisdom about the mind. If this paper’s account of the self is near the mark, we now have the elements of a scientific understanding of the origin and workings of the self. It will help us understand why we are not naturally mindful, and why our emotions can be unnecessarily painful, persistent and destructive. That understanding could help to make mindfulness less puzzling and more accessible, in ways we cannot yet foresee. It might show how language, being linked to the evolutionary origins of the self, confounds any resolution of the self in the paradoxes and distortions of words.  Other vehicles for thinking – scientific, mathematical, pictorial, or computational – might resolve these paradoxes.  This could help show a way towards greater mindfulness for many people, beyond the dedicated few.

These ideas are described at greater depth in papers in preparation. I would greatly appreciate feedback on the ideas and the papers. If you are interested, please contact


Maynard Smith, J. ‘Evolution and the Theory of Games’, Cambridge University Press, 1982

Rao, R.P.N, Olshausen, B.A, and Lewicki, M.S. (Eds) ‘Probabilistic models of the brain’, MIT press, 2002

Worden, R.P. ‘A speed limit for evolution’, Journal of Theoretical Biology, 176, 137 – 152, 1995

Worden, R.P. ‘Primate Social Intelligence’ , Cognitive Science, 20(4): 579-616, 1996

Stay Informed, Stay Inspired

Discover the frontier of academic insights with our newsletter. Imprint Academic brings a world of scholarly discourse right to your inbox. Stay updated on the latest publications, special offers, and the vibrant conversations shaping today’s academic landscape. Your quest for knowledge deserves a companion as dedicated as our newsletter. Sign up now and become a part of a community driven by curiosity and intellectual exploration.
Something went wrong. Please check your entries and try again.