Roadmap
This page says where the book is going. The first chapter is written and the second has begun. Everything after that is a plan, and the plan changes as the experiments come in: stages have already been moved, merged and left out, and that will happen again.
The order follows the five breakthroughs of A Brief History of Intelligence. Each is one chapter, each belongs to a group of animals, and each builds on the brain of the chapter before it.
| chapter | animals | what the brain gains | |
|---|---|---|---|
| 1. Steering | the first worms | good and bad, needs, moods, a first kind of learning | written |
| 2. Reinforcing | the first fish | learning from reward what to do | in progress |
| 3. Simulating | the first mammals | trying something in the head before doing it | planned |
| 4. Mentalizing | primates | a model of its own mind, and of the minds of others | planned |
| 5. Speaking | humans | knowledge that passes from one brain to another | far off |
Chapter 1: steering
The brain of a small worm. It turns what it sees into movement (Steering), sorts the world into good and bad (Valence), stops reacting to what is always there (Habituation), learns in one life which new foods hurt (Association), looks after hunger and cold (Drives), and stays in a state for a while after a meal or a poison bite, a state that builds up over several bites (Affect).
This animal learns one thing: what to stay away from. It cannot learn what is worth going to, and it cannot learn what to do.
Chapter 2: reinforcing
The brain of an early fish. The first vertebrates hunted in open water, and nearly all the parts of the vertebrate brain were already there. This chapter adds them in the following order.
- The pallium (page). The first region whose wiring is random and not shaped by evolution. It spreads the look of a thing out over many cells, so that things that look alike can still be told apart, even when only a combination of features says which is which. In chapter 1, what the animal learned about one food spread to the foods that resembled it.
- Dopamine. A cell that fires when things turn out better than expected, and falls silent when they turn out worse. With it the animal learns what predicts food, and what predicts the thing that predicts food. It also learns to like, which chapter 1 could not do.
- Hands. Until here the animal eats what it grasps. Now it can pick something up, hold it and eat it later, so that there is something to choose between.
- The basal ganglia. The same dopamine signal teaches which action to take. This is learning by trial and error, as in a rat that learns to press a lever.
- Curiosity. New things become a little rewarding in themselves, so that the animal tries what it does not know yet.
Parked until these work: a map of places and the routes between them (the hippocampus), precise timing (the cerebellum), and modulators that set how fast the animal learns when its world is steady or changing.
Two things change in how the stages are done. The circuits are designed and tested before any evolution, and evolution then tunes them, in fewer generations than in chapter 1. And the learning stages get lab experiments: one animal in a scripted situation, as in an animal lab, to see whether it learns the way real animals do.
Chapter 3: simulating
The brain of an early mammal, and its new part, the neocortex. The cortex learns to predict what the animal will sense next. Once it can do that, it can be run without the world: the animal stops at a choice, plays out each option, and picks the one that ends best.
The planned stages: a cortex that predicts, trying actions in the head at a choice point, replaying experience during rest and sleep, holding a goal through distractions, and remembering a single event.
The tests are the classic ones from the rat lab. An animal that explored a maze without reward goes straight to food that appears there later. An animal whose path is blocked takes a detour. An animal that has had enough of one food stops working for it at once.
Chapter 4: mentalizing
The brain of a primate, living in a group in which the others matter. The brain first gets a model of itself: what it knows, and how sure it is. Applied to another animal, the same model says what that animal wants and what it has seen.
The planned stages: knowing when it does not know, reading the goals of others, learning a skill by watching, and preparing for a need it does not feel yet, such as storing food before the winter.
Chapter 5: speaking
The human step. Calls with a fixed meaning come first, then signals whose meaning is agreed on within a group, then naming what two animals are both looking at. The aim at the end is knowledge that no single life could find again and that still grows from one generation to the next.
This chapter is the least certain, both in what is known about evolution and in whether a small grid world can carry it.
The world grows too
A brain can only be tested on what its world asks of it, so the world grows with the brain. It takes more and more of the rules of One Hour One Life: tools, fire and cooking, and chains of steps that are too long to find by accident. Animals that move by themselves come in as prey and as predators.
For the last two chapters, the way lives are run has to change as well. A generation now starts and ends together. Animals that watch, imitate and teach each other need parents, children and lives that overlap.