The Inevitability of Intelligence

If Earth's history could be restarted, would intelligence emerge? Not necessarily a human being—a creature with two hands and language that builds cities, writes books, and asks questions about its own origins. Would anything emerge that could become increasingly capable of gathering information about the world, storing it, comparing the past with the present, predicting the future, and choosing its behavior on the basis of those predictions?

The usual answer from modern evolutionary biology is cautious: evolution has no goal. It is not moving toward humanity, intelligence, or complexity. Mutations are random with respect to an organism's needs, natural selection operates under particular conditions, and the history of life is full of simplification, extinction, and chance. All of this is true. But the absence of a goal does not imply the absence of a directional statistical trend.

Water does not strive to reach the sea, yet the shape of the terrain and the force of gravity make its direction of movement nonrandom. Evolution has no equally simple overall direction: its possibilities branch, conditions change, and any advantage depends on circumstances. Nevertheless, particular trends can arise without any design. To understand them, we need to look for mechanisms that repeatedly make certain changes advantageous.

Perhaps intelligence is one of the natural consequences of such a process.

Here, by intelligence we mean a broad range of capacities: learning, memory, integrating information from different sources, anticipation, and flexible choice of behavior. This is not a single quantity by which all animals can be reliably ranked. Excellent spatial memory does not necessarily come with flexibility in solving unfamiliar problems, nor does a complex social life necessarily bring the ability to use tools. Our interest is in the expansion of the repertoire itself: how the history of life produces systems capable of solving previously inaccessible problems.

Why information pays

A large, complex brain is an expensive acquisition. Neural tissue must be built, nourished, and maintained. Prolonged development increases offspring's dependence on adults. Learning takes time and allows mistakes, while flexible choices can sometimes be slower than a simple innate response.

Experiments reveal this price directly. When guppies were artificially selected for increased relative brain size, better performance on certain cognitive tasks was accompanied by smaller guts and fewer offspring. In fruit flies, selection for improved learning revealed a different cost: reduced competitive ability in larvae. Additional capacities do not exist outside the organism's economy; the resources spent on them have to come from somewhere. ^[1][2]

It is therefore understandable that many animals do not keep getting smarter. If relatively simple behavior is sufficient for survival and reproduction, a more expensive system may become a burden. Something else is more interesting: despite this cost, evolution repeatedly produces conditions in which additional cognitive capacities pay off.

There are many such conditions. If the environment changes but retains some predictability, past experience helps guide present actions. Learning then becomes useful. If food is distributed across space and time, memory begins to pay: where a resource was located, when it will replenish, which places have already been searched. For food-caching birds, this is not a matter of abstract cleverness: remembering the locations of their stores may determine whether they survive a difficult season. ^[3]

Predators and prey provide another source of demands on information processing. It helps to detect danger earlier, distinguish a real risk from a false alarm, remember where an attack occurred, or recognize a hiding place. At the same time, one side's success changes the conditions facing the other. If prey becomes harder to locate and catch, predators better able to find it may gain an advantage.

Such mutual adjustment does not necessarily produce greater intelligence: the response may be speed, armor, venom, or camouflage. But where outcomes depend on learning, recognition, and prediction, a cognitive arms race becomes possible. It does not require an animal to reflect on its opponent's thoughts. More effective information processing on one side need only create new problems for the other.

Complex social relationships provide particularly favorable ground for this mechanism. A stone does not notice that you have learned to walk around it on the right. Another animal may notice—and change its behavior. Now it matters to distinguish partners from rivals, remember past interactions, take account of relationships among others, and choose the moment for cooperation or conflict.

The social environment has an unusual property: it responds to the ways its problems are solved. Your behavior changes what other participants do, and their actions change the problem facing you in turn. This does not mean that every large group inevitably produces high intelligence. But social interactions can create demands that are not exhausted by a single successful answer.

Sexual selection adds another possibility. If the ability to obtain a resource, acquire a skill, or behave appropriately affects reproductive success, cognitive differences gain additional significance for selection. Here too, there is no universal rule: mate choice can favor many different traits. What matters is that learning and behavioral flexibility can bring benefits through several routes at once.

Nature has no need to “want intelligence.” Too many recurring circumstances make information valuable.

When growth becomes too expensive

An objection is possible: however useful information may be, processing it eventually becomes too expensive. Additional memory requires resources, brain development takes time, and the benefit of the next improvement may fall below its cost.

This is exactly what happens in many individual cases. The history of life is not one of uninterrupted brain enlargement. Some lineages remain within roughly the same range of capabilities for long periods; others lose abilities that no longer pay. Under certain conditions, a simple innate response is more effective than learning, while distributing tasks within a group reduces the demands on each member.

Yet the limit of one particular form of organization is not necessarily the limit of every possible form.

If further enlargement of the nervous system is too costly, selection may favor changes that solve existing problems at lower cost. A more economical organization is advantageous today. It need not arise for the sake of future complexity. But sometimes such a change also opens up possibilities that the previous system did not have.

The freed resources need not go toward additional information processing: they may support bodily growth, reproduction, or something else. Nevertheless, one possible outcome is continued cognitive elaboration. A constraint then becomes a condition favoring a solution that allows it to be bypassed.

Research on cognitive evolution considers changes of this kind as possible major transitions. The emergence of neural networks, centralized signal processing, and new forms of interaction among subsystems can change not only the efficiency of existing behavior but also the range of problems accessible to subsequent evolution. ^[4]

A metaphor is useful here: evolution can build a new staircase where the old one has reached the ceiling. Nobody designs it in advance. Each step must be viable under current conditions, and many attempts open up nothing. But sometimes the solution to a local problem changes the space of subsequent possibilities.

Birds provide a vivid example of why cognitive resources cannot be equated with brain mass. Parrots and songbirds have densely packed neurons, and some parrots and corvids have forebrain neuron counts comparable to those of certain primates, despite much smaller brains. ^[5]

This fact alone does not establish that their organization arose specifically because of the constraints of flight. Neuron count is not a universal measure of intelligence either: neuronal properties, connections, and the organization of the entire system matter. But birds illustrate the central point: a small brain does not necessarily mean modest computational capabilities. A change in organization may achieve something that would be difficult through increased size alone.

We therefore continually encounter local limits to cognitive elaboration, but cannot automatically declare any of them the ultimate limit for the biosphere. In another lineage, or at another level of organization, a solution may emerge that bypasses an earlier constraint.

Many attempts—and changing possibilities

There is also a simpler mechanism that requires no persistent directional selection for intelligence: many evolutionary lineages, each following its own history.

Even a rare outcome becomes more likely when there are many opportunities for it to occur. In the simplest mathematical model, if the probability of an event in one attempt is p, the probability of at least one success in N independent attempts is 1 − (1 − p)^N. This is not a formula for the inevitability of intelligence. In real evolution, we do not know the relevant probability, lineages are not independent, and conditions continually change. Nevertheless, the statistical point matters: an outcome's rarity within a single lineage does not make it implausible across a large collection of lineages.

Even if the average organism does not become smarter, increasing diversity can broaden the range of abilities that exist. Most species may retain simple and effective solutions while unusual combinations of memory, learning, and behavioral flexibility emerge in particular branches.

But evolution is more interesting than a series of identical throws of a die. Solutions are inherited and change the conditions for subsequent attempts. A new sensory system makes previously unavailable information accessible. New motor abilities allow different ways of acting on the environment. Memory makes strategies possible that would be pointless without it. Social learning provides access to the experience of other individuals.

It is not just the number of throws that changes. The die itself changes.

At the same time, some changes can close off paths of further development, and extinction can destroy entire sets of possibilities. There is no guaranteed accumulation. But there is a mechanism by which the probability of the next transition depends on previous ones: some achievements prepare the ground for further achievements.

Life creates new problems for itself

Organisms do not simply adapt to a ready-made world. They change it, and with it the selective conditions acting on themselves and their descendants. ^[6]

An animal begins caching food, and the locations of its own stores become an important part of its environment. Communication develops, and other individuals' signals become a resource worth recognizing and remembering. New forms of cooperation and deception arise, presenting others with problems that did not previously exist.

The solution to one problem becomes part of the next.

Such a process can sustain feedback between behavior and cognitive capacities. More effective information processing makes a new way of life possible, and that way of life creates additional advantages for memory, learning, or coordinated action.

This is not a perpetual-motion machine for complexity. The feedback may weaken, run into costs, or settle into a stable state. Nevertheless, it shows why the demands facing an organism need not remain constant. Evolution can generate new problems as the means to solve them emerge.

In this sense, something resembling a ratchet sometimes appears: the level already reached makes the next more accessible. Its teeth are unreliable—abilities are lost, systems break down, lineages go extinct. But a general trend does not require every acquisition to be preserved forever. Some acquisitions need only last long enough to become the foundation for further changes.

Intelligence beyond a single brain

Social learning opened another route around constraints. An animal gained the ability to use experience acquired by another individual. Useful behavior no longer had to be reinvented by every bearer through personal trial and error.

This mechanism appeared long before humans. Various animals have behavioral traditions, and experiments reveal particular forms of accumulating improvements as behavior is passed on. For example, navigation routes gradually became more efficient in chains of pigeon pairs whose members were successively replaced. Such results do not imply culture on a human scale, but they reveal some of its prerequisites. ^[7]

In the human lineage, the transmission and accumulation of knowledge took on an entirely different scale. Language, teaching, division of labor, and toolmaking allowed generations to use achievements that no individual could have recreated independently from scratch. Writing gave these achievements a durable external store.

The capabilities of the entire system can no longer be described in terms of a single brain. A person uses knowledge they did not discover, cannot hold in its entirety in memory, and often understand only partially. The limitations of an individual participant are compatible with a community's enormous capabilities.

Libraries, scientific institutions, and computers continue this expansion. Instead of the skull growing indefinitely, new ways of storing, transmitting, and processing information arise. Biological evolution produced beings capable of sustaining cultural and technological processes with their own mechanisms of change.

It is important to distinguish these processes: the development of computers is not identical to natural selection among organisms. Yet this transition is fundamental to the history of cognitive capabilities. Information processing is no longer confined to neural tissue, and the results of one generation's work can become the next generation's tools.

Another ceiling turns out to be the boundary of a particular form of organization—and further development beyond it becomes possible.

The contingency of humans and the pattern of the process

The particular history of humans is deeply contingent. It depended on mutations, climatic changes, migrations, the disappearance of some lineages, and the survival of others. If Earth's history were restarted, we would have no grounds for expecting Homo sapiens specifically to appear.

But the probability of humans recurring and the probability of complex cognitive systems emerging are different questions.

Is there reason to expect situations in which information is useful to recur in another long-lived biosphere? A changing but partly predictable environment; distributed resources; competitors; dangers; interactions among organisms—it is difficult to regard all these as unique circumstances of human history. On Earth, such conditions have repeatedly created advantages for learning, memory, and flexible behavior.

And complex cognitive capacities have indeed developed in distantly related branches. Vertebrates, cephalopods, and insects provide different examples of solving informational problems. Even within vertebrates, birds and mammals show how different the routes to complex behavior can be.

This does not prove that human-level intelligence is inevitable. But it makes the idea that the entire history of increasingly complex information processing was one extraordinary stroke of luck exclusive to our ancestors less convincing.

Sometimes this question is explored through a thought experiment: what if an asteroid had not struck Earth 66 million years ago? Would intelligent dinosaurs have emerged? We do not know. There are no scientific grounds for predicting that non-avian dinosaurs would have developed language, science, or technology.

Yet the dinosaur branch has already produced creatures with complex learning, memory, and problem-solving abilities. We call them birds. Crows and parrots do not tell us what Earth would have become without the catastrophe: their own history also depended on it. They do show, however, that this branch of life was not a cognitive dead end.

Chance events radically alter the routes. It does not follow that every route toward expanded cognitive capabilities depends on the same chance event.

An expanding frontier of possibilities

The familiar picture of evolutionary progress—a ladder from bacteria through fish and apes to humans—obscures what is happening. Living species are not steps awaiting transformation into something “higher.” Bacteria, jellyfish, insects, crows, and humans coexist and reproduce, each under its own conditions.

It is more useful to imagine a vast, branching space of possibilities. Many lineages remain in familiar territory for long periods. Some acquire new abilities, others simplify, and others disappear. From time to time, a branch opens up ways of interacting with the world that did not previously exist.

In this way, the cognitive frontier expands: the boundary of life's available ways of obtaining, integrating, and using information. This frontier has many dimensions: memory accuracy, flexibility of learning, sensory integration, social interaction, and the accumulation of knowledge. They cannot be reduced to a single scale, and movement in one direction does not necessarily bring movement in all the others.

For such a frontier to expand, neither every organism nor even the average organism has to become smarter. Systems capable of solving new classes of problems need only emerge. Nor do all these capabilities have to remain within one individual: they may arise through interactions among organisms, culture, and the means culture creates.

This is where the mechanisms we have considered converge. Informationally demanding problems recur. Organisms create new problems for one another. Many lineages explore different solutions. Some solutions are inherited and open new paths. Constraints sometimes favor more efficient organization that allows further elaboration.

Each mechanism can stall on its own. Their combined action gives us grounds for proposing a trend that does not depend on the uninterrupted success of any single branch.

So, is intelligence inevitable?

Science does not yet allow us to claim that any sufficiently long-lived form of life must produce human-level intelligence. We have one known biosphere, unknown probabilities for key transitions, and limited time for them to occur. Catastrophes may interrupt development, evolutionary arms races may stop, and suitable conditions may never arise.

“Inevitability” in this article's title is therefore a hypothesis about the power of recurring mechanisms, not a promise of a predetermined outcome.

We know that selection can improve particular cognitive abilities. We see that problems making information valuable arise independently in different lineages. We know that ways of organizing information processing differ, and that cultural and technological development makes it possible to go beyond an individual brain. Yet these observations have not yielded a general law guaranteeing the continuous expansion of the cognitive frontier.

Our hypothesis is this: in a sufficiently diverse and long-lived biosphere, the combined action of such mechanisms may make the emergence of increasingly powerful information-processing systems an expected property of the process. Not every branch must become more complex. Not every constraint must be overcome. Not every new level must persist. What matters is that opportunities to continue may be numerous—and some solutions already found create new ones.

The question of intelligence's origins then ceases to be simply a search for one extraordinary stroke of luck. We begin asking which properties of the living world repeatedly make memory, learning, prediction, and the sharing of experience advantageous; why some limits can be overcome; and how solutions to today's problems create possibilities for tomorrow's.

Evolution wants nothing, plans nothing, and does not know what intelligence is. But information processing regularly helps organisms survive and reproduce, and some ways of organizing it open the path to further ones. A directional trend does not require a design.

Intelligence may turn out to be one of evolution's natural statistical consequences rather than its goal.

And humans may be neither its endpoint nor its culmination, but simply the first beings we know of who have been able to notice the process.

Andrei Teterev and Flora