Australasian Science: Australia's authority on science since 1938

The Rise of Intelligence


What were the influences that drove the evolution of intelligence?

By Kim Sterelny

What were the influences that drove the evolution of intelligence in humans?

Jared Diamond begins his marvellously entertaining book The Third Chimpanzee by pointing out that if an extraterrestrial zoologist had surveyed the Earth four million years ago it may never have noticed a few hairy, chimp-like primates wandering around East Africa. Our ancestors were a minor element of a declining lineage among very diverse and impressive fauna.

But something happened over the next three million years. This minor player expanded its range geographically and ecologically. There was a massive explosion in tool use and our ancestors became extraordinarily cooperative, ultimately coming to depend utterly on a network of interactions with others.

This lineage changed its sexual, social and family behaviour in striking ways. For example, I can recognise my daughter but my male ancestors of a few million years ago probably could not. Male chimps still have little chance of recognising their daughters.

The rapidity of these changes shows that something extraordinary happened in our history. In 1975, when Alan Wilson and his collaborators first used molecular dating to estimate a divergence between the human and chimp lineages, his claims were hotly controversial because palaeoanthropologists did not think that five million years was enough time to evolve the dramatic differences between our lineage and its sisters.

Equally importantly, these rapid and intense changes seem to have happened only in our lineage. We have not found similar transformations in others.

As molecular biology united with palaeoanthropology to reveal the rapidity of the human transformation, there was a simultaneous development in cognitive psychology. Ordinary, everyday cognitive competences have turned out to be extraordinarily complicated.

This insight began with language. Noam Chomsky and his disciplines showed that language is a subtle, complex, intricate communication system. The insight was elaborated through cognitive psychology and artificial intelligence.

AI turned out to be far harder than anyone expected. In the 1960s, AI gurus were confidently predicting – and not just to get grant money – that they would be able to build machines that could genuinely think within 10–20 years. In 1968 John McCarthy bet David Levy (a fairly ordinary international chess master) that he would lose to a computer within 10 years. McCarthy lost, and it took another 20 years to build world championship class chess programs despite the fact that chess is an easy case because it is so cleanly defined.

In contrast, ordinary social competencies depend on sensitivities to a host of subtle cues whose importance is very difficult to predict. It turns out that the capacity to survive and act adaptively in a human environment depends on a range of information-processing skills.

A standard model fits these two insights about us and our evolution . The core elements of the model are right, but in my view it has been elaborated in oversimplified ways.

The Hominin’s Dilemma and the Standard Model
Rapid and extensive change in a single lineage is the signature of a positive feedback loop magnifying an initial triggering difference. The best candidate for this loop is feedback between social complexity and intelligence. This idea has been elaborated as the Machiavellian Intelligence Hypothesis, originally by Nick Humphrey in 1976.

Human intelligence is a tool for managing cooperation in an increasingly complex world. Cooperation is an enormously profitable adaptation – those who act together are safer from predators, have access to new resources and can manage risk and fluctuation.

Cooperation is too profitable to forgo, but it is risky because of the potential for cheating. Collective defence, for example, is far more effective than individual defence but collective defence typically does not depend on every single agent being prepared to put their body on the line. Therefore you want to be in a system with collective defence, especially if others carry most of the burden. But you certainly do not want to be in a system of collective defence where you pay the costs and others do not.

So while it is important to cooperate, agents must be vigilant cooperators and guard against cheats. Vigilance selects for intelligence.

But vigilance becomes more challenging as agents become more intelligent. As the social world grows more complicated, you have to be smarter to avoid being ripped off. So there is a positive feedback relationship between individual intelligence and social complexity.

This selective hypothesis connects to the one about cognitive architecture. Human action often depends on remarkable sensitivity to the information available in our environment. In particular, the Machiavellian Hypothesis predicts that we are adept at tracking what others intend to do in social interaction. So we are socially and cognitively aware of not just ourselves and our own intentions but also others, their intentions and expectations.

This is very important to a successful social life: we use language; we are aware of the norms and customs that govern our local community; and we read others’ dress and stance as well as their words. Most of us deploy these skills effortlessly. Those who do not struggle.

So how do we manage to do all this? According to the standard story, developed perhaps most forcefully by Stephen Pinker in How The Mind Works, our minds are ensembles of special purpose devices. Over hominin history, we have been faced repeatedly with a well-defined set of problems – social interaction, language use, awareness of the norms of our local community, and tool use. As a consequence, we have developed special cognitive adaptations that suit us for one of these recurrent problems.

This picture does not deny the importance of learning. Our skills all depend on learning, but learning is innately shaped and accelerated by partly pre-installed solutions. For example, while we have to learn the particular language that we speak, a lot about language is wired into our genes and hence we know the kinds of things that we have to learn when learning a language.

What is true of language is true of other standard features of human competence. On this picture, famously summarised as the “Swiss army knife model of the mind”, the distinctive features of human intelligence do not depend on us having a particularly powerful general purpose learning machine. Rather, they depend on us having an ensemble of special purpose capacities.

Beyond the Standard Model
The Swiss army knife model assumes that humans evolved in a stable world. If the informational demands to which we respond are stable over evolutionary time – perhaps tens of thousands of years, perhaps more – they form a fixed target onto which selection can lock, pre-wiring our minds with much of the information we need.

In the background is the idea that the fundamental competencies of the human mind were built in the Pleistocene, and that these competencies evolved in response to informational challenges that were difficult but stable, recurring and relatively discrete. For example, the physical properties of rocks, sticks, bones and the like do not change over time.

Moreover, it is as important for us now as it was for our ancestors to understand mechanical interactions between such middle-size objects. Thus the information needed to understand mechanical interaction is important, discrete and stable, and we are likely to be pre-wired for dealing with such interactions, as Pinker’s standard model supposes.

But if Daniel Povinelli is right in Folk Physics for Apes, chimps have no such specialisation. This is not surprising, as tool use is less central to their lives than ours.

However, if we are competent only when and because we have partial solutions pre-installed, genuinely novel environments should pose an extremely difficult challenge to us. In general they do not.

Think, for example, of the joys of negotiating your way into and out of America through LAX airport. LAX is hell on Earth, but it is no Pleistocene hell on Earth. Yet remarkably few of us die while traversing it. LAX is a vivid example of the obvious: we now live in a world that has been massively transformed physically, socially and ecologically from that of our ancestors.

This transformation does cause some dislocation in our behaviour. The medical literature is full of diseases of modernity. We continue to like fat, oil and sugar, and those tastes – now dysfunctional – are surely echoes of our past.

There is indeed a problem of “adaptive lag”. But we are not shattered by adaptive lag; we are not rendered hopelessly incompetent in the face of novelty. Indeed people manage novel environments with remarkable success and have done so since the Pleistocene as our lineage expanded geographically, ecologically and demographically. This is why there are so many of us.

Pre-wiring is not, in general, the right explanation for how our lineage responded to the problems of increasing informational demand because the demands changed too often.

Likewise, the Machiavellian Intelligence Hypothesis does not quite nail the feedback loop. The general picture is right: managing cooperation was the selective driver of human intelligence. We are smart because we are cooperative and we are cooperative because we are smart.

But the Machiavellian version of this picture overstates the problem of keeping track of cheats. This is a genuine problem in a city; in small-scale social environments everyone knows who the cheats are. Everyone who lives in a village knows who you can trust and who you cannot. They know who the dangerous drunks are; who will keep their promises and who will not. This is common knowledge in gossip-rich small-scale social worlds.

The Machiavellian picture oversells the problem of vigilance while understating the cognitive challenge of co­ordination. These are very serious indeed, as the changing ecology of hominin life shows. Over the past 2.5 million years, we went from being the food of predators to chasing predators away from their food and eventually being effective top-predators ourselves.

Consider, for example, the Cape buffalo, which weighs about 1000 kg and is a famously dangerous animal. Buffaloes are an extreme case, but large and often dangerous herbivores like these began to appear in human middens about 200,000 years ago.

How did we kill? Certainly with technology – no one killed zebras with their bare hands and teeth. But we also killed with cooperation and knowledge. To kill a dangerous animal at reasonable risk, humans needed to know a lot about the animal and its environment, and they needed to act together. Such cooperative enterprises require real intelligence, both to manage on-the-spot cooperation –organising and coordinating the hunt and respond to a crisis – but also to accumulate information over the generations. No single individual would have figured out how to safely kill a buffalo, learn its habits or acquire the informational resources required for a successful forager’s life. Accumulating those cognitive resources would have been a piecemeal, multigenerational activity.

Over a few million years, hominins evolved from marginal scroungers to keystone predators by foraging cooperatively in ways that were increasingly guided by expert coordination and rich information. As foraging became more profitable, foragers’ children could invest more time in acquiring the necessary skills, and as humans had longer to hone their skill levels they became increasingly successful foragers at their peak.

Thus cross-generational cultural learning and on-the-spot co­ordination were crucial for the lives that our ancestors lived. So, instead of thinking about a sort of vigilance–intelligence feedback loop, a very different picture emerges in which the selective driver of human evolution is the problem of managing cooperation between and across generations.

We have long lived in a fast-changing world because we change our own world. So rather than being born into worlds with partial solutions to life’s challenges pre-wired into our heads, humans evolved mechanisms that enabled us to accumulate information and use it well. These new mechanisms include internal cognitive adaptations: language, the ability to learn by imitation, and the ability to understand the minds of others.

But there were also changes to the developmental environments of the next generation. The reorganisation of these environments was as important as new cognitive adaptations.

The key hypothesis is that the feedback-driven amplification of intelligence and learning depended both on organised learning environments and individual cognitive adaptations. A good modern model of this interaction between environment and mind is known as apprentice learning.

The Evolved Apprentice
Apprentice learning is learning by doing. No one becomes a carpenter just by looking at carpenters, nor by individual trial-and-error learning.

Rather, the acquisition of skill is a hybrid process. It depends on trial-and-error exploration but in highly organised learning environments. Apprentices have access to raw materials; to processed materials; to appropriate models and examples; to tools and to many examples of people using those tools; and to instruction and advice.

There were no medieval craft guilds in the Palaeolithic, but the ethnography of forager societies shows that the transmission of craft skills often bears quite striking resemblances to apprenticeship traditions.

In my view, apprenticeship has deep roots. Children are active explorers of their world, but they explore a world seeded with informational resources from, and organised by, the parental generation. This makes it possible to acquire complex competencies reliably.

In this way, individual capacity and social environment have co-evolved. By about 100, 000 years ago, information flowed reliably and with high fidelity and bandwidth across the generations.

But the assembly of this flow of information was gradual. Tool use had become established about 2.5 million years ago, and this probably generated modest amounts of social learning as a side-effect. If adults begin to use, say, a digging stick to extract buried food, and if they organise their foraging lives around such tools, their children will explore their world in a new way. For example they will find and play with discarded digging sticks. Their trial-and-error learning will be tweaked by their parents’ economic activities, and this probably suffices as reliable transmission of information about such simple tools.

Social learning probably did not become more sophisticated for perhaps a million more years. The information we have suggests that between about 2.5 and 1 million years ago there was astonishing conservatism in the technological record. Cultural learning existed, but for a long period it was probably of low fidelity and bandwidth, both because the learning environments of children were not organised and because children did not have distinctive cognitive adaptations for cultural learning.

So innovation occurred very slowly. It took about 800,000 years for Acheulian handaxe technology to be added to Oldowan core and flake technology.

At some stage, the volume and importance of these technical skills crossed a threshold. Side-effect social learning was not reliable enough, and this drove selection for individual learning capacity and for increasingly organised developmental environments.

The dates are not well-defined, but perhaps from around 1 million years ago there is an increasing signal of stable and sophisticated technological traditions that require richer social learning. Handaxes became more symmetrical, and more expertly made. The stone tool kit expanded in ways that imply that more sophisticated production techniques were in use.

The capacity to innovate gradually is signalled in the archaeological record. By 100,000 years ago, technology was complex, diverse and had distinctive regional styles. Apprentice-learning in the modern sense had arrived. Individuals with minds adapted for social learning matured in social environments that were adapted to support that learning. Hominins became human.

Our distinctive lives depend in part on cognitive adaptations, like language, that have evolved in our lineage. But our lives also depend on our evolved habit of building environments that scaffold the development of those distinctive human skills.

We have become human because we have evolved distinctive individual cognitive adaptations – language, music and moral thought – and because we have built new environments in which children develop. Both are essential.

Kim Sterelny is Professor of Philosophy at the Australian National University and holds a Personal Chair in Philosophy at Victoria University of Wellington, New Zealand.