A Brief History of Intelligence: Evolution, AI, and the Five Breakthroughs That Made Our Brains

by Max Solomon Bennett

visual processing in mammals was hierarchical, with lower levels having smaller receptive fields and recognizing simpler features, and higher levels having larger receptive fields and recognizing more complex objects. Second, at a given level of the hierarchy, neurons were all sensitive to similar features, just in different places.

Fukushima invented a new architecture of artificial neural networks, one designed to capture these two ideas discovered by Hubel and Wiesel.

His architecture first decomposed input pictures into multiple feature maps, like V1 seemed to do. Each feature map was a grid that signaled the location of a feature—such as vertical or horizontal lines—within the input picture. This process is called a convolution, hence the name applied to the type of network that Fukushima had invented: convolutional neural networks.* After these feature maps identified certain features, their output was compressed and passed to another set of feature maps that could combine them into higher-level features across a wider area of the picture, merging lines and edges into more complex objects. All this was designed to be analogous to the visual processing of the mammalian cortex. And, amazingly, it worked.

Convolutional neural networks are designed with the assumption of translational invariance, that a given feature in one location should be treated the same as that same feature but in a different location. This is an impregnable fact of our visual world: the same thing can exist in different places without the thing being different.

Despite being inspired by the brain, convolutional neural networks (CNNs) are, in fact, a poor approximation of how brains recognize visual patterns.

CNNs impose the constraint of translation, but they don’t inherently understand rotations of 3D objects, and thus don’t do a great job recognizing objects when rotated.*

While auto-association captures some principles of how pattern recognition works in the cortex, clearly even the cortex of fish is doing something far more sophisticated. Some theorize that the vertebrate brain’s ability to solve the invariance problem derives not from the unique cortical structures in mammals, but from the complex interactions between the cortex and the thalamus, interactions that have been present since the first vertebrates. Perhaps the thalamus—a ball-shaped structure at the center of the brain—operates like a three-dimensional blackboard, with the cortex providing initial sensory input and the thalamus intelligently routing this sensory information around other areas of the cortex, together rendering full 3D objects from 2D input, thereby flexibly able to recognize rotated and translated objects.

Indeed, while CNNs may not capture exactly how the brain works, they reveal the power of a good inductive bias. In pattern recognition, it is good assumptions that make learning fast and efficient.

In some ways, the tiny fish brain surpasses some of our best computer-vision systems. CNNs require incredible amounts of data to understand changes in rotations and 3D objects, but a fish seems to recognize new angles of a 3D object in one shot.

The coevolution of the familiar sensory organs and the familiar brain of vertebrates is not a coincidence—they each facilitated the other’s growth and complexity. Each incremental improvement to the brain’s pattern recognition expanded the benefits to be gained by having more detailed sensory organs; and each incremental improvement in the detail of sensory organs expanded the benefits to be gained by more sophisticated pattern recognition. In the brain, the result was the vertebrate cortex, which somehow recognizes patterns without supervision, somehow accurately discriminates overlapping patterns and generalizes patterns to new experiences, somehow continually learns patterns without suffering from catastrophic forgetting, and somehow recognizes patterns despite large variances in its input.

The greater the brain’s ability to learn arbitrary actions in response to things in the world, the greater the benefit to be gained from recognizing more things in the world. The more unique objects and places a brain can recognize, the more unique actions it can learn to take. And so the cortex, basal ganglia, and sensory organs evolved together, all emerging from the same machinations of reinforcement learning.

researchers began applying Sutton’s temporal difference learning to all kinds of different games. And one by one, games that had previously been “unsolvable” were successfully beaten by these algorithms; TD learning algorithms eventually surpassed human-level performance in video games like Pinball, Star Gunner, Robotank, Road Runner, Pong, and Space Invaders. And yet there was one Atari game that was perplexingly out of reach: Montezuma’s Revenge.

It wasn’t until 2018 when an algorithm was developed that finally completed level one of Montezuma’s Revenge. This new algorithm, developed by Google’s DeepMind, accomplished this feat by adding something familiar that was missing from Sutton’s original TD learning algorithm: curiosity.

reinforcement learning requires two opponent processes—one for behaviors that were previously reinforced (exploitation) and the other for behaviors that are new (exploration). These choices are, by definition, opposing each other. Exploitation will always drive behavior toward known rewards, and exploration will always drive toward what is unknown.

The approach is to make AI systems explicitly curious, to reward them for exploring new places and doing new things, to make surprise itself reinforcing. The greater the novelty, the larger the compulsion to explore it. When AI systems playing Montezuma’s Revenge were given this intrinsic motivation to explore new things, they behaved very differently—indeed, more like a human player.

The importance of curiosity in reinforcement learning algorithms suggests that a brain designed to learn through reinforcement, such as the brain of early vertebrates, should also exhibit curiosity. And indeed, evidence suggests that it was early vertebrates who first became curious.

Curiosity and reinforcement learning coevolved because curiosity is a requirement for reinforcement learning to work.

the first vertebrates were presented with a new opportunity: for the first time, learning became, in and of itself, an extremely valuable activity. The more patterns a vertebrate recognized and the more places she remembered, the better she would survive.

And so it was 500 million years ago in the tiny brain of our fish-like ancestors when curiosity first emerged.

the ability to construct an internal model of the external world, was inherited from the brains of first vertebrates.

The ability to learn a spatial map is seen across vertebrates. Fish, reptiles, mice, monkeys, and humans all do this. And yet simple bilaterians like nematodes are incapable of learning such a spatial map—they cannot remember the location of one thing relative to another thing. Even many advanced invertebrates such as bees and ants are unable to solve spatial tasks.

trial and error in vertebrates, in turn, made it possible for the even more perplexing and monumental breakthrough that would follow. It was early mammals who first figured out how to engage in a different flavor of trial and error: learning not by doing but by imagining.

They became the first tetrapods (tetra for “four” and pods for “feet”), most closely resembling a modern amphibian such as a salamander. One evolutionary lineage of tetrapods, who were lucky enough to live in parts of the Earth that still supported these warmer puddles, would maintain this lifestyle for hundreds of millions of years—they would become the amphibians of today.

amniotes—the creatures that developed the ability to lay leathery eggs that could survive out of the water. The first amniotes probably best resembled a lizard of today. Amniotes found an inland ecosystem abundant with food—insects and plants were everywhere for the feasting. Eventually, the Devonian ice age faded and amniotes spread and diversified to all corners of the Earth.

The other lineage of amniotes were our ancestors: the therapsids. The therapsids differed from reptiles at the time in one important way: they developed warm-bloodedness. Therapsids were the first vertebrates to evolve the ability to use energy to generate their own internal heat.* This was a gamble. They would require far more food to survive, but in return they had the ability to hunt at any time, including the cold nights when their reptile cousins lay immobile—an easy feast offered on a Permian platter.

They grew to the size of a modern tiger and began to grow hair to further maintain their heat. These therapsids would have looked like large hairy lizards.

The neocortex gave this small mouse a superpower—the ability to simulate actions before they occurred. It could look out at a web of branches leading from its hole to a tasty insect.

early vertebrates got the power of learning by doing, then early mammals got the even more impressive power of learning before doing—of learning by imagining.

a side effect of warm-bloodedness was that mammal brains could operate much faster than fish or reptile brains. This made it possible to perform substantially more complex computations. This is why reptiles, despite their long-range vision on land, were never endowed with the gift of simulating. The only nonmammals that have shown evidence of the ability to simulate actions and plan are birds. And birds are, conspicuously, the only nonmammal species alive today that independently evolved warm-bloodedness.

In the human brain, the neocortex takes up 70 percent of brain volume. In the breakthroughs that followed, this originally small structure would progressively expand from a clever trick to the epicenter of intelligence.

If you unfolded a human neocortical sheet, it would be almost three square feet in surface area—about the size of a small desk.

The auditory and visual cortices are interchangeable.

If the visual cortex is damaged, patients become partially blind. But over time, function can return. This is usually not the consequence of the damaged area of neocortex recovering; typically, that area of neocortex remains dead forever. Instead, nearby areas of neocortex become repurposed to fulfill the functions of the now-damaged area of neocortex. This too suggests that areas of neocortex are interchangeable.

Further, if Mountcastle’s theory is correct, it suggests that the neocortical column implements some algorithm that is so general and universal that it can be applied to extremely diverse functions such as movement, language, and perception across every sensory modality.

The first thing that became clear to these nineteenth-century scientists was that the human mind automatically and unconsciously fills in missing things.

No matter which conversation you tune in to, the auditory input into your ear is identical; the only difference is what your brain infers from that input. You can perceive only a single conversation at a time.

Figure 11.6 can be interpreted as a frog (see here if you don’t see this). Once your mind perceives this interpretation, you will never be able to unsee it. This is what might be called the can’t-unsee property of perception.

It turns out that there is, in fact, an abundance of evidence that the neocortical microcircuit is implementing such a generative model.

Mental rehearsal of motor skills substantially increases performance across speaking, golf swings, and even surgical maneuvers. The motor cortex’s skill in sensorimotor planning enabled early mammals to learn and execute precise movements.

The frontal neocortex of early placental mammals was organized into a hierarchy. At the top of the hierarchy was the agranular prefrontal cortex, where high-level goals are constructed based on amygdala and hypothalamus activation.

The aPFC then propagates these goals to a nearby frontal region (the premotor cortex), which constructs subgoals and propagates these subgoals further until they reach the motor cortex, which then constructs sub-subgoals.

This hierarchy enables more efficient processing by distributing effort across many different neocortical columns.

There is plenty of evidence for such a motor hierarchy. Recordings have shown that neurons in the aPFC are sensitive to high-level goals, whereas those in the premotor and motor cortex are sensitive to progressively lower-level subgoals. Learning a new behavior activates all levels of the motor hierarchy at first, but as the behavior becomes automatic, it activates only lower levels in the hierarchy.

There is also plenty of evidence for the idea that the frontal neocortex is the locus of simulation, while the basal ganglia is the locus of automation. Damaging an animal’s motor cortex impairs movement planning and learning new movements but not the execution of well-trained movements (because the back part of the basal ganglia already learned them).

Indeed, why primates have such big brains—and specifically such large neocortices—is a question that has perplexed scientists since the days of Darwin. What was it about the lifestyle of early primates that necessitated such a big brain?

It was Robin Dunbar who did this, and what he found shook the field. This correlation has been confirmed across many primates: the bigger the neocortex of a primate, the bigger its social group. But monkeys and apes are far from the only mammals, let alone animals, that live in groups. And interestingly, this correlation does not hold for most other animals. The brain of a buffalo living in a thousand-member herd is not meaningfully bigger than the brain of a solitary moose. It isn’t group size in general but the specific type of group that early primates created that seemed to have required larger brains. There was something unique about primate groups relative to those of most other mammals, something we can understand only by understanding the general drive for grouping itself.

In these early mammals, this collaborative period between mother and child was relatively short-lived. After a period of childhood development, the bond tends to fade, and the children and mothers go their separate ways. This is how it is for many mammals that spend most of their lives on their own, like tigers and bears. But not all animals separate in adulthood like this. In fact, the simplest, most widely used, and likely first collective behavior in animals was group living, whereby animals of the same species simply clustered together.

The key benefit of group living is that it helps stave off predators.

Since Menzel’s work, numerous other experiments have similarly found that apes can, in fact, understand the intentions of others.

This act of inferring someone’s intent and knowledge is called “theory of mind”—so named because it requires us to have a theory about the minds of others. It is a cognitive feat that evidence suggests emerged in early primates. And as we will see, theory of mind might explain why primates have such big brains and why their brain size correlates with group size.