This is the third and final instalment in my series of articles about the brain and rationality. The first article looked at the troubles encountered by artificial intelligence researchers. The second looked at the complexity of the environment and the complexity of the brain and concluded that any attempt to rationally calculate decisions was entirely impossible. This article picks up the thread and answers the question of how the brain actually functions.
The brain is a generalisation machine. It functions by coming up with generalisations of the patterns of inputs that it receives and mapping them to generalised behavioural patterns. It tests these generalisations against reality – the ones that are better than randomness are retained, through natural selection which operates on multiple levels, and the ones that are worse die out. It accumulates generalisations on top of generalisations, constantly working to reduce the complexity of the processing it has to deal with by summarising, ignoring the particular, ruthlessly throwing away detail and substituting general patterns. It is like an overworked executive screaming at his subordinates: “shut up with that detail, summarise!” Human brains are the accumulation of 600 million years or so of generalisations that were simply more successful than randomness. The closest analogy is science – the brain’s accumulated and inherited generalisations are equivalent to the existing body of scientific theory, the brain’s generalisation process is the scientific method in action.
If we go back 600 million years to the emergence of the first animals, we can see how the process works. Animals are, more than anything else, defined by the presence of neurons. The electrical communication supported by neurons has two great advantages when compared to chemical communication. It is fast and it can be precisely directed. When plants experience some trauma to some part of themselves, this is signalled to the rest of the organism by a flood of chemicals produced by the damaged cells. Similarly, when some part of a plant encounters a rich source of nutrients in its environment, this is communicated to the rest of the plant by the relevant cells producing chemical signals to the rest of the organism. This process is relatively slow and diffuse as chemicals spread out like a cloud through the organism.
Neurons change everything – they allow electrical signals to be conveyed between very precise parts of the organism almost instantly – small charged ions rather than complex hormonal molecules are the signal carriers and the signals travel almost as quickly as the speed of propagation of electro-magnetic waves in a vacuum – many orders of magnitude quicker than chemical diffusion. With the evolution of neurons it became possible for organisms to react to very immediate changes in their environment; thus a whole new evolutionary arms race began – as animals became capable of detecting and reacting to one another’s behaviour.
So, imagine we have a simple animal with a single neuron whose input comes from a light-sensitive cell and whose output is connected to a muscle. The neuron translates inputs from the light sensitive cell into contractions of the muscle – before long one of these animals will be constituted so that this connection serves to make the animal move towards light. “Move towards the light” is a very simple generalisation. It is very far from a fail-safe rule of life: it will cause these animals to hop merrily into fires, bounce nonchalantly into deserts, move away from food, and many other death-producing behaviours. But all it has to be is better than any other equally simple generalisation (e.g. move away from light, move orthogonally with respect to light, if you detect light, don’t move, move randomly, etc) because very simple generalisations are all that is possible with such a simple creature. Survival is the measure for the quality of generalisations. Those animals whose wiring expresses the best generalisation will survive in greater numbers than the others and the pattern will be passed on to their offspring and the generalisation will eventually come to be encoded in the genes of the species.
This process of generalisation operates on all levels. If we consider the neuronal pathways that are responsible for visual processing: photons strike the retina, causing neurons to fire. Each firing neuron represents a dot of light. As the electrical signals pass upwards through the brain, dots are generalised into lines, lines into shapes, shapes into objects, objects into scenes, scenes into situations. At each stage of processing, the details are thrown away and only the general, higher-level pattern is passed upwards. Details can be brought into focus either through attention or salience (i.e. generalisations that depend on patterns in the lower level inputs). Otherwise the details are just thrown away and the generalised pattern is all that is passed onwards and upwards for further processing. By the time the signals reach the level of consciousness, all that is delivered is the ‘gist’ of the visual scene – an extremely simplified generalisation with all the details rendered fuzzy. The process of progressively generalising visual scenes is built into the neuronal architecture – it is how the brain deals with the overwhelming complexity of visual inputs and turns it into simplified patterns that can be processed at a much higher level of abstraction. Without this process of simplification and generalisation, the brain would remain eternally overloaded with the constantly changing waves of photons hitting the retina.
A good example of the pattern matching and generalisation processes that occur on the visual pathway relates to the identification of dangerous elements in the environment: specifically snakes and spiders. The patterns of motion that are characteristic of these animals is encoded deep in our brains and produces a general ‘recoil’ behavioural impulse. This occurs before the visual signals have processed the stimulus and generalised it into an object – a ‘slithery’ or ‘spindly’ pattern of motion or general shape, is sufficient to produce the recoil. The reaction is based on the low level visual pattern, which is only later processed into an object. In some cases, we will realise that it was not, in fact, a spider or a snake that we have recoiled from. The generalisation that is encoded into our brains is “as soon as you see anything slithery or spindly – get away – you can figure out what it is later.” The cost of waiting the few milliseconds that it takes for the brain to turn the motion pattern into an object is greater than the cost of recoiling falsely.
Generalisations are equally important on the human social level as they are in neuronal processing. For example, one of the most basic generalisations that humans use in order to navigate the vast complexity of the social world is imitation. Children imitate their parents, their older siblings and their peers. In unfamiliar social situations, imitating the behaviour of those around you is a great general rule that is instinctively adopted by most people, to a certain extent at least. Imitation works as a successful generalisation because it increases the overall complexity that individuals can handle – it is rarely possible, never mind tractable, to work out social rules for an unfamiliar world from first principles. Imitation allows such situations to incorporate unfamiliar people without everybody having to renegotiate the basic norms for interpersonal interactions and group dynamics. Without broad use of the imitation pattern, all large social events would have to devote significant parts of their time and energy to figuring out ways in which to organise the basic foundations on which higher level communication depends. Imitation is also the basic cognitive building block on which inter-generational cultural transmission depends.
successful generalisations build upon older, successful generalisations
Successful generalisations build upon older, successful generalisations. Each new generalisation frees up the brain for dealing with higher level, more complex and longer-term problems. There is a basic transmission chain which goes from the individual decision to the individual personality to the broad culture to the genetic. When individuals come up with a general strategy that works successfully for them in a broad class of situations, they will tend to apply it again and again. Repeated generalisations that prove successful become encoded into their neuronal architecture and become part of their basic behavioural programme – because that’s how brains work. For example, the mechanics of riding a bicycle becomes effortless once one has mastered it – they are encoded as a generic sub-routine in the subconscious parts of the brain, freeing up the higher levels of the brain to deal with other problems. Generalisations that prove broadly successful across large numbers of people will become incorporated into culture and the most broadly successful of them will diffuse from culture to culture. Broad generalisations that prove successful over many generations will become encoded into the genetic code that dictates the pattern that brains form and will, in effect, become part of human nature (or at least part of the nature of a subset of humans).
So, how does the brain actually do this? If it is, as I claim, a machine for storing and generating generalisations, there must be some physical mechanism by which neurons achieve this outcome.
At the very lowest neuronal level, there are two basic mechanisms which effectively lead to a tendency towards generalisation. The first one is activation. Activation strengthens neural pathways. Every time that an electric signal passes through a particular neural network in the brain, the pathway becomes minutely stronger – in that the connections between neurons that the signal passes through become more firmly established. So, if a stimulus (for example a particular smell) causes an electrical signal to pass through a particular pathway, when the user encounters a slightly different stimulus, it is more likely that the same pathway will be activated – a very simple automatic implementation of generalisation.
There is another, very low-level principle which underlies the brain’s tendency to form generalisations. It is on a slightly less solid scientific basis as the mechanism remains slightly mysterious. But it remains one of the most basic principles of neuro-science: that neurons that fire together wire together. That is to say, if two neurons tend to fire at the same time, the brain will create a connection between them. So, for example, if we take two types of visual object (for example a can of coca cola and a healthy, slim and attractive young person looking extremely contented), each of them will cause a different neuron to fire in the brain representing that class of object. If the brain repeatedly comes across the two objects in the same visual scene, the two neurons will tend to fire together and, eventually the brain will wire them together. This means that, if one of them is encountered on its own, without the other, the other one may fire (due to the principle of spreading activation) as if the other one was present. The stimuli are associated at the neuronal level. So, when we see a can of coca cola, our brains may conjure up the mental image of a young healthy, attractive person, despite the fact that any rational analysis of the two objects would suggest that their association does not run very deep (in fact, one might speculate that only people who never drink coca-cola appear in coca-cola advertisements).
Another basic mechanism relates to energy. The brain represents concepts by generating ‘action potentials’ in neurons (which can be expressed informally as neurons ‘firing’). There is an energy cost of neurons firing. The fewer neurons that have to fire to represent a given concept or set of concepts, the lower the overall energy cost. The more general the representation, the fewer neurons that must fire to support it. This produces a very broad and general evolutionary advantage to representational generalisation – energy efficiency.
It would, however, be a mistake to think that the brain performs generalisation purely through such low-level, undifferentiated mechanisms. Neurons are not a great big undifferentiated mass that wire together in response to stimuli. They are arranged in particular structures which cooperate to produce higher level functions. These might be relatively simple arrangements with little functional differentiation: for example, deep belief networks are artificial constructs based upon neural patterns which are used in computer science for pattern matching and are extremely good at aggregating general patterns out of fuzzy data. It is plausible that similar arrangements of neurons in the brain are responsible for producing higher-level generalisations.
Higher-level modular functionality
However, there are other more complex and differentiated structures in the brain. From the top down, the brain can be divided up into numerous functional units, each with a different job in the overall information processing . These functional units map to regions of the brain – generally the deeper down in the brain, the closer to the brain stem, the more basic the function and the earlier it evolved. But this mapping between function and region only happens in a fuzzy, inexact way, given the presence of multiple connections between neurons in different regions which mean, unlike computer circuits, the components can never be considered to be entirely independent of one another. Nevertheless, as a first approximation, the functional units can be considered to be discrete and to operate largely independently of one another, regardless of their physical layout.
There are individual sub-units that have specific responsibility for speech and language; for processing the different types of sensory inputs: hearing, vision, touch, smell, taste; for managing motor functions; for processing social interactions and so on. These units depend on neuronal arrangements that are specialised for particular functions. At first glance, this contradicts the model of the brain as a generalisation machine, but the first glance gives a false impression. When a generalisation becomes incorporated into the genes, what it means is that the neuronal pattern that encodes it is formed more or less automatically as part of the organ’s development and does not have to be formed afresh through experience.
From time to time, a random genetic mutation occurs which causes the individual’s brain’s wiring to form in a pattern that meant it takes less work to learn the most successful, important and persistent generalisations. The individual with this mutation has a competitive advantage, as their brain is freed up to deal with other problems while their peers’ brains are busy moving the generalisation from culture or experience into their neuronal architecture. Hence, such mutations proliferate and eventually become incorporated as a normal part of the species’ genome and the neuronal patterns that they encode become part of the species’ basic wiring.
The specialised functional sub-units of the brain are mostly examples of generalisations that worked their way into genes a long time ago. The ability to turn a torrent of photons hitting the retina into a coherent and simple scene is an information processing pattern that is useful to virtually all animals with some form of vision, so it is likely that our cognitive architecture for processing such signals is very similar to that of other animals and evolved long before humans or primates did. The ability to interpret social signals, read emotions and navigate complex social hierarchies and networks is less broadly useful to animals and thus the patterns that encode such generalisations world are less deep down in our brains and likely to be less hard-wired. Language and speech evolved even more recently and are still less instinctive and automatic.
These specialised functional units of the brain encode general purpose functions that have to be calibrated through experience to deal with the specific environment that individuals find themselves in. For example, the brain’s language centres encode an abstract and fuzzy generalisation of language patterns. Linguists have been trying, for more than half a century to identify a universal grammar – with limited success. It seems likely that the brain’s wiring, rather than having any such universal rules hard-wired into it, is biased towards certain patterns of grammar. The general patterns of abstract language processing that are encoded in the genes solidify into a more specific pattern as the individual learns a real language in their early years.
The generation and accumulation of generalisations by the brain creates its own problems – very quickly situations arise where generalisations overlap and clash. For example, a table with both a delicious cake and a snake on it might invoke one response pattern of “eat” and another response pattern of “run away” – both behavioural generalisations cannot be fulfilled at once. This is a very simple and contrived example, but the complexity of accumulated generalisations is such that similar situations are, in practice, the norm. There are almost always multiple competing impulses acting on the individual in any given situation. So, how does the brain deal with this competition? “humans do not direct their behaviour towards rationally calculated evolutionary advantages. They seek to attain more desirable emotional states and avoid less desirable ones”
The answer is emotion. Emotions are a fuzzy way of measuring multiple simultaneous generalisations and producing an aggregate effect. Here I am using the broadest possible definition of emotion: any sort of feeling or sensation that has some teleological or purposeful element. From this point of view the experience of pain, pleasure, hunger and thirst are all emotions (or at least they are sensations with emotional components). What makes emotions so important in cognition is that they provide direction. Humans do not direct their behaviour towards rationally calculated evolutionary advantages. They seek to attain more desirable emotional states and avoid less desirable ones. This is what provides direction to human behaviour. Evolution, calibrated by experience, is what makes certain emotional states desirable and others undesirable. Thus, the way that the brain experiences acute pain is almost universally considered to be undesirable, while the way that it experiences sexual intimacy with an attractive partner is almost universally considered to be desirable.
What makes emotional states such a good solution to the problem of competing and simultaneous generalisations is their multi-faceted nature. People frequently experience situations where strong impulses are competing with one another and emotions can encapsulate the ambiguity of such situations. For example, a potential sexual liaison may provoke strong feelings of desire, mingled with strong feelings of fear – fear for the potential loss of an existing relationship and aversion to future feelings of loneliness and abandonment if it is discovered. Simple rules are insufficient for dealing with this ambiguity. For example, there are some situations where, on balance, the urge to avoid extreme pain may be over-ruled by some other imperative. A rule that states ‘avoid pain’ is insufficient.
This account of behaviour driven by generalisations and emotion may seem deficient as it is manifestly the case that people sometimes suppress immediate emotional impulses in favour of longer term, rational goals. However, this contrast is deceptive. What makes the long term goal more desirable in such cases is the imagined future emotional state that it will lead to – without an emotional state to provide direction, there is no way of comparing outcomes and no purpose to behaviour (this is why, incidentally, the most profoundly depressed people are almost incapable of movement). In some cases the emotional satisfaction of being self-disciplined enough to avoid immediate emotional temptations may itself be sufficient to drive behaviour – but it is still emotionally driven.
Even when considering ultra-logical endeavours such as mathematics or computer programming, generalisations and emotions play a critical role – almost nobody works out 2+2 = 4 from first principles. People apply general arithmetic patterns which they have learned and arrive at answers that ‘feel right’. In both programming and mathematics, the search for elegance is one of the primary driving factors and determinants of what seems correct. Elegance is not a concept that makes any sense from a purely rational point of view. The most elegant solution flows most directly from the deepest generalisation. The act of coming up with an elegant solution to a logical problem can inspire feelings of intense pleasure in a mathematician or programmer – and that is what provides direction to the task.
Emotion is just as important in social life. Cooperation over wide groups – ultra-sociality – is one of the most characteristic and unusual features of human behaviour which has allowed our species to attain unparalleled mastery over our environment. People can recognise intellectually that cooperation is good in the abstract, but that is not what motivates them to cooperate. Humans cooperate because cooperation feels good. Cooperation feels good because groups with ultra-social traits out-compete groups without them. Evolution has built that pattern into our cognitive architecture – and that’s why we cooperate (although, of course, this is only true in general – both game theory and observation demonstrate that anti-social humans can persist alongside the overwhelming pro-social majority).
Modern research in robotics is increasingly turned towards the use of emotional based control systems, known as affective or emotional-behavioural systems in the computer science jargon. Initially they were focused on allowing robots to interact more easily with humans by displaying emotions to allow the humans to better understand their internal states. However, as robots become more complex, with multiple different behavioural modules, they face the same problem as humans do – which modules should they invoke in any given situation, when there are multiple competing stimuli that could be addressed through incompatible responses? As soon as you get multiple competing goals, and non-linear interactions between the solutions, it quickly becomes intractable to describe behaviour as rules, so you need something flexible and fuzzy like emotions. Such affective controllers are, as it turns out, good solutions for any domain in which the complexity is such that stimulus-response rules are insufficient. For example, a few years ago I published a research paper which described a multi-level affective control system for managing adaptive networks.
The way in which evolution uses emotions to drive behaviour in order to promote gene-propagation is an endlessly fascinating topic, to me at least, and I could go into it at great length, but this article is long enough already, so I’ll leave that for another day. To finish, I’ll just mention one particularly interesting general feature of emotions. Satisfaction has a built-in decay function – emotional animals will never arrive in a state of permanent satisfaction. As soon as a desired state is attained, the satisfaction of reaching this state will start to fade and fresh desires will emerge to drive behaviour towards further gene-proliferating behaviours. This is an inescapable conclusion both of evolutionary logic (because a brain that has this characteristic will always out propagate one that does not have it) and of simple observation of behaviour – millionaires soon come to envy deca-millionaires who come to envy billionaires….
A very important aspect of the brain’s functional sub-divisions and emotional direction is that the very great majority of this functioning is sub-conscious and, mostly beyond the awareness of the consciousness. The various sub-units process information, generate emotional states and in some cases generate physical responses before the consciousness even becomes aware of the stimulus. This has been well-established by modern cognitive science – the consciousness is slow acting and it takes a significant amount of time for information to reach it from the senses. It can be demonstrated, for example, that in tennis, the receiver of a serve will have started their return stroke before the consciousness could possibly have become aware of the ball’s trajectory – a sub-conscious functional unit has been trained to produce these responses in an autonomic fashion.
The consciousness receives a high-level summary of the situation, complete with emotional weights calculated by these sub-conscious functional units. So, if all of this behaviour is based on generalisations, emotion and sub-conscious processing, what does the consciousness actually do and is there any space for rational calculation? The basic generalisation process of the brain is more or less an automatic feature of the brain’s neural wiring. The functional sub-units of the brain are organised in such a way that this process is tuned towards specific types of stimulus – the language centre generalises over language inputs, the visual centre over visual inputs and so on.
Conscious, deliberate reasoning, by contrast, uses the same basic process of generalisation but it is a general purpose mechanism that can be applied to any type of stimulus, on any level of abstraction, including the highest level, most abstract and longest term generalisations. This makes it slow and resource intensive – it needs to marshal inputs from across the brain’s sub-units and deliberately reason over them in order to weigh up their emotional outcomes. This does resemble the traditional model of the rational actor but the rational aspect of it is relatively superficial – the reasoning process can be best understood as orchestrating the sub-consciousness to process imagined future states.
It is reasonably clear that the consciousness is required in situations where problems must be solved by following rules or algorithms that have been supplied to the brain from without (for example, when learning a new technique that is, in some respects, counter-intuitive). However, the interesting thing about consciousness is that the more one looks into it, the less involvement in controlling behaviour it has. Sophisticated experiments to measure cognition show that the brain typically initiates responses before the conscious mind has arrived at the decision to do so. The most plausible roles for consciousness are: firstly, coming up with a retrospective narrative to explain the individual’s behaviours to create the illusion of a unified identity under the control of consciousness, acting according to a rational plan; secondly, orchestrating sub-conscious processes together in long term planning. However, despite the application of some of humanity’s greatest minds to the problem, it remains an elusive entity.
The modular model of the brain illuminates many of the apparently paradoxical observations about human cognition and behaviour. The functional sub-units can be considered more or less independently of one another. So, for example, an individual may have extraordinary language skills, or outstanding social skills, or outstanding motor skills, while being almost totally incapable of sophisticated high-level reasoning – their genome may contain sequences which encode exceptionally well developed wiring plans for functional sub-units but relatively poor functioning of the high-level general purpose reasoning functions of the consciousness – in everyday language, there are many types of intelligence and they can vary independently of one another.
Brain injuries and brain-developmental disorders provide voluminous evidence which prove the modular nature of cognition. There have been many cases where individuals have received traumatic head injuries which have destroyed a very localised region of the brain and they have lost very specific functions as a result, without greatly affecting any of their other capabilities. Williams syndrome and autism are good example of gene-based developmental disorders which have effects that are opposite in many respects. Individuals with Williams syndrome have typically low IQs and poor reasoning abilities, but excellent social and verbal skills, while individuals with autism often have the opposite characteristics.
This observation answers the puzzle that I mentioned earlier, which I encountered while working in Artificial Intelligence research. How is it that teams of the world’s smartest people are incapable of writing programmes to emulate natural language when some of the world’s apparently thickest people can master it effortlessly? At the time I thought it was proof of the innate smartness of ordinary people and the arrogance of experts. There is something to this, but only to a very limited extent. Abstract reasoning and problem solving capabilities are fairly rare – very few people possess the cognitive architecture to carry them out to a very high level and, even then, they usually require significant amounts of training. They are the capabilities that we normally associate with intelligence. They are rare for a few reasons: firstly because they are the highest level of information processing and secondly because they don’t give people that much of an evolutionary advantage. An individual with exceptional modules in their brain for mastering language or motor control, or social skills, or music is highly likely to out-compete somebody who has excellent abstract reasoning capabilities. To put it crudely, Ronaldo has more opportunities to propagate his genes than any physicist ever will.
It is increasingly widely accepted, at least among cognitive scientists, that humans are not rational calculation machines. However, it remains little understood that the brain is essentially a generalisation machine driven by emotion, with rational calculation playing a very minor role. It is a model that is, naturally, something of a simplification, but it is a model that is a much more accurate representation of reality than the rational actor.
Modern psychology and behavioural sciences (such as behavioural economics) have experimentally explored many of the ways in which human behaviour departs from rationality. This has given rise to categorisations of “cognitive biases”. However, these biases are often considered to be undesirable deviations from a rationalistic ideal; biases that we should strive to suppress in order to better navigate the world in a rational manner. Such interpretations are quite wrong and the remedies which they propose would be disastrous. These biases are the behavioural manifestations of some of the generalisations that are buried deep within our cognitive wiring. These generalisations are widely shared by the population because they are broadly useful in reducing the complexity of the information processing task faced by our brains. Without them we would be quite incapable of navigating the overwhelmingly intractable complexity of the world. Furthermore, many of these generalisation are embedded in sub-conscious processing modules which are difficult or impossible to consciously control.
One consequence of the brain as generalisation machine is that people tend to over-generalise from the limited number of examples that they come into contact with. There are two good evolutionary reasons for this. Firstly, large-scale social groups are a relatively new phenomenon in human history – no older than 12,000 years or so. The great majority of human history was spent in relatively small groups, with relatively limited mobility. Thus, people came into contact with a much greater proportion of the people, places and things that were relevant to their survival than they do in modern, mass societies. Thus, generalisations from experience were much more likely to hold when scaled up. Secondly, in a world of endless complexity, over-generalisation has an inherent advantage over under-generalisation. Over generalisation creates false positives (situations covered by the heuristic which should not be), under-generalisation creates false negatives (situations not covered which should be). If the error-rate is the same in both directions, we get the same number of mistakes in each case, but with under-generalisation, we have more cases left over that require independent cognitive investment, leaving us with less cognitive resources to apply to other problems. And there will always be other problems demanding cognition.
The advent of massive-scale societies in modern times has changed the cognitive environment significantly. Models of group dynamics that are generalised from personal experience do not apply very well to organisations and societies that number in the millions and billions. Understanding large scale social and institutional dynamics requires both significant amounts of specific knowledge about their functioning and deliberate, algorithmic reasoning – it cannot flow from intuitive generalisations. Public commentary about large institutions or broad social dynamics, whether by professional media commentators or uninvolved members of the public is almost invariably based on misapplied generalisation, as can be observed through casual observation of virtually any media output – perceived systemic problems are attributed to aspects of human nature: laziness, greed, arrogance, pride and so on. Under the guise of democratic debate, such generalisations have a corrosive effect on both social cohesion and democracy, because they are entirely unrelated to the actual problems of system and institutional design: of course people are lazy, greedy, arrogant, proud and so on – but they are also energetic, selfless and humble. Which aspects of human nature are expressed depends on the institutional environment.
Another major problem that our generalisation based cognition faces in the modern world relates to science. Nowadays we have the tools to allow us to systematically deconstruct human psychology to identify broadly shared generalisations – biases – and to exploit the fact that such generalisations are only generally true. This is the domain of advertising, marketing, public relations and much of practical psychology. It is relatively easy to exploit cognitive false-positives in order to implant certain ideas in the brain that are not true. The example above of coca-cola advertising creating associations between fit, healthy and happy people and their product is a particularly simple and obvious example, but similar approaches are pervasive. For example, a common generalisation that people make is that organic produce is produced by small-scale, local, traditional farmers and not by large scale, industrial capitalist agri-businesses. Once this generalisation is understood, it is a trivial matter for agri-businesses to attach the word ‘organic’ to their produce and to exploit the brain’s propensity to over-generalise.
Such ‘gaming’ of the brain’s generalisations is hardly new. Every generalisation creates a niche that can be exploited and it is almost a rule of evolution that available niches will be filled. However, small-scale societies contained within them the ability to defend themselves against such gaming – simply by virtue of the fact that surveillance and enforcement of collective ethics is easy when everybody knows everybody else. Furthermore, such societies typically had strong collective identities which tolerated suppression of behaviours and individuals considered to be anti-social. In our modern, large-scale, market-based, individualist societies, such mechanisms are vastly weaker.
This discussion has only briefly touched upon some of the implications of an understanding of the brain as a generalisation accumulation and generation machine and the problems that it faces in modern, large-scale societies, without offering anything in the way of solutions. In future articles, I will return to these problems in greater detail and look at ways in which they can be addressed, if not solved. The most important point to grasp is that rationalism is not a viable solution.