Chapter 10: The Neural Dark Ages

Core Idea

For more than a hundred million years after vertebrates came onto land, bodies changed a great deal but brains hardly changed at all. Bennett calls this the neural dark age. The spark came only with the first mammals, small warm-blooded burrowers hiding from dinosaurs. One region of their cortex became the neocortex, which gave them the ability to simulate actions before taking them. If early vertebrates learned by doing, early mammals could learn “learning before doing—of learning by imagining.” This is Breakthrough #3: Simulating.

Frameworks Introduced

The two speculated requirements for simulating:
- When to use: to explain why simulation appeared in mammals (and birds) but not in fish, reptiles, or other animals that also had the first move.
- Requirement 1, far-ranging vision: planning a route only pays off when you can see much of your surroundings. On land, even at night, you can see up to one hundred times farther than underwater. Fish instead chose fast reactions, which is why they have a large midbrain and hindbrain and a comparatively smaller cortex.
- Requirement 2, warm-bloodedness: simulating is far more costly and slow than the reinforcement learning of the cortex and basal ganglia. Neurons fire much more slowly when cold, so warm brains can run much more complex computations. Reptiles had long-range vision on land but stayed cold-blooded, and never simulated.
- Check: birds are the only nonmammals with evidence of simulating and planning, and they are the only other living group that independently evolved warm-bloodedness.
The gift of the first move: animals that hide in a burrow or behind a branch can look around before acting. They can spot a far-off bird and a tasty insect and decide whether to run. Many animals had this position (crabs under sand, small fish in coral), but only mammals evolved a brain that used it.
The four-way split of the cortex (figure 10.3). In early mammals, the cortex of early vertebrates split into four structures. Three kept their old jobs, and one was truly new:

Early vertebrate cortex	Early mammal structure	What it does
Ventral cortex	Associative amygdala	Recognizes patterns across senses, especially those that predict good or bad outcomes
Lateral cortex	Olfactory cortex	Detects smell patterns with auto-associative networks
Medial cortex	Hippocampus	Learns spatial maps with similar circuitry
(a fourth region)	Neocortex	Completely different circuitry; simulates

The figure draws this as two cross-sections. In the early vertebrate, the medial cortex sits on top, the lateral cortex at the side and the ventral cortex below. In the early mammal, the neocortex is a thin band wrapping over the top and sides, with the hippocampus tucked inside it, the olfactory cortex low on the side, and a small amygdala at the bottom.

Key Concepts

Neural dark age: the long stretch from early vertebrates through tetrapods, reptiles and therapsids when evolution changed bodies (jaws, armor, lungs, scales, fur, warm blood) but not brains. A modern fish brain and a modern reptile brain are still remarkably similar.
Tetrapods and amniotes: the first four-footed animals, and then the lineage that laid leathery eggs that could survive out of water.
Therapsids: our ancestors among the amniotes. They were the first vertebrates to generate their own internal heat, which let them hunt reptiles that lay immobile on cold nights.
Cynodonts: small, burrowing, plant-eating therapsids that survived the Permian-Triassic extinction. Their descendants shrank to about four inches long and became the first mammals.
Neocortex: the new region of cortex in early mammals (neo means new). In early mammals it was small, and the olfactory cortex took most of the brain’s volume. In humans it takes up 70 percent of brain volume.
Simulating: playing out possible actions and their results inside the brain, then picking the path where you end up “both alive and well fed.”
What stayed the same: the basal ganglia still learned to take actions that maximize dopamine release, now with input from the neocortex too. The hypothalamus still triggered valence responses. The midbrain and hindbrain still ran reflexive movement, now for walking rather than swimming.

Mental Models

Think of evolution as switching its effort between bodies and brains. For hundreds of millions of years it settled for the reinforcement-learning brain and changed body parts instead.
Think of extinction events as moments when a small niche can become the winning strategy. Lungs, warm blood and burrowing were all small niches before a mass death made them dominant.
Use two questions to judge whether planning would pay off for an animal or an agent: who gets the first move, and can they see far enough to use it?
Think of warm-bloodedness as a gamble. It costs far more food, but it buys activity at any hour and faster neurons.

Anti-patterns

Assuming brains grew steadily more complex over time: from early vertebrates to the first mammals they barely changed.
Assuming the first move is enough: crabs and coral fish had it too. Without far-ranging vision and fast, warm neurons, planning does not pay.
Treating warm-bloodedness as a pure win: the large therapsids died first in the Permian-Triassic extinction because they needed so many calories. Reptiles, with their scant diets, survived better.

Rules of Thumb (author’s wording)

“all reigns come to an end.”
“Extinction events create opportunities for small niches to transform into dominant strategies.”
The early mammal could not outfight or outrun dinosaurs, but “they got to make the first move.”

Worked Example

The evolutionary path to the first mammals (figure 10.2 shows this as a tree with a timeline).

Devonian period, 420 to 375 million years ago: predatory fish fill the oceans. Arthropods move onto land among new plants, which grow from 30 centimeters tall to 30 meters tall over the period.
The spread of plants lowers carbon dioxide, the climate cools and the oceans freeze: the Late Devonian Extinction. Our ancestors had already evolved lungs for short trips between shore puddles, so they survived. They lost their gills and became tetrapods.
One tetrapod line stayed by the water (amphibians). The other went inland, evolved leathery eggs and became the amniotes, which spread during the Carboniferous and Permian (350 to 250 million years ago).
Amniotes split into reptiles and therapsids. Therapsids evolved warm-bloodedness, then whiskers and fur, and from 300 to 250 million years ago they were the most successful land animals, some as big as a tiger.
The Permian-Triassic extinction, around 250 million years ago, kills 96 percent of marine life and 70 percent of land life within five to ten million years. The large therapsids die. Only small burrowing cynodonts survive.
Reptiles then rule for about 150 million years, including archosaurs, dinosaurs and the first flying vertebrate, the pterosaur. Cynodonts shrink to about four inches, hide by day and hunt insects at night. They become the first mammals.
During this hundred-million-year reign of dinosaurs, the mammals evolve the neocortex and with it simulation.

The tree also marks the dinosaur branch toward theropods with a possible independent evolution of warm-bloodedness, the line that leads to birds. The part opener shows a small, round-bodied early mammal with a long tail looking out from its ground, and figure 10.1 shows a large, lizard-like first therapsid with tusk-like teeth.

Key Takeaways

Brains were stuck for hundreds of millions of years. The neocortex was the biggest neural innovation since the Cambrian fish.
The neocortex let early mammals simulate actions before taking them, which is learning by imagining instead of only by doing.
Simulating probably needed both far-ranging vision and warm-bloodedness. That explains why only mammals and birds show it.
Of the four cortical regions in early mammals, three (amygdala, olfactory cortex, hippocampus) kept old jobs, and only the neocortex was new.
The rest of the brain (basal ganglia, hypothalamus, midbrain, hindbrain) stayed much as it was in early vertebrates.

Connects To

Ch 5–9: the reinforcement-learning brain of early vertebrates, with its cortex, basal ganglia, and TD learning, is what the neocortex builds on.
Ch 11: what the neocortex computes (a generative model).
Ch 12–13: the abilities simulation gave mammals and how the brain chooses what to simulate.
Ch 15: the neocortex keeps expanding in primates.