Glossary

Active inference — Friston’s idea that the brain explains its own behavior and then acts to make its predictions true (Ch 13, 14)

Actor and critic — the two parts of temporal difference learning: the critic predicts future reward, and the actor is reinforced by changes in that prediction (Ch 6)

Adaptation — neurons shift how stimulus strength maps to firing rate, so they report change relative to a recent baseline (Ch 1, 2)

Affect — a behavioral state described by valence (good or bad) and arousal (high or low) (Ch 3)

Agranular prefrontal cortex (aPFC) — the frontal neocortex of early mammals, which lacks layer 4; it models the animal’s own behavior as intent and triggers simulation (Ch 13)

Auto-association — neurons that connect to each other, so a partial pattern brings back the whole one (Ch 7)

Basal ganglia — the vertebrate structure that learns by trial and error; it acts as the actor (and possibly the critic) and ungates actions that led to dopamine (Ch 6)

Bilaterian — an animal with a front, a back and left-right symmetry; the first brains appeared in this group (Ch 2)

Bischof-Kohler hypothesis — the claim that only humans can plan for future needs; the book revises it to include primates (Ch 18)

Catastrophic forgetting — a network trained on new patterns loses the old ones (Ch 7)

Continual learning problem — learning new things as data arrives while the rules of the world change (Ch 4)

Counterfactual learning — learning from what would have happened after a different choice (Ch 12)

Credit assignment problem — deciding which earlier cue or action gets credit for an outcome; the temporal version concerns things separated in time (Ch 4, 6)

Curiosity — surprise or novelty treated as reinforcing in itself; needed to solve the exploration-exploitation dilemma (Ch 8)

Declarative label — an arbitrary symbol for an object or action, with no reward attached; only humans use them naturally (Ch 19)

Ecological-brain hypothesis — the fruit diet of early primates drove their brain growth (Ch 18)

Eligibility traces / overshadowing / latent inhibition / blocking — the four original credit assignment tricks, which favor the recent, the strong, the novel and the already-learned cue (Ch 4)

Episodic memory — re-simulating specific past events (Ch 12)

Expansion recoding — sparse projection of a few inputs onto many neurons, so overlapping patterns produce different outputs (Ch 7)

Exploration-exploitation dilemma — the conflict between repeating reinforced behavior and trying new behavior (Ch 8)

False-belief test — a test of whether someone can keep another’s knowledge apart from their own, such as the Sally-Ann test (Ch 16)

Generative model — a model that learns to generate its own data and recognizes things by comparing generated with real data (Ch 11)

Granular prefrontal cortex (gPFC) — new primate frontal cortex with a layer 4; it models the aPFC, giving a model of one’s own mind (Ch 16)

Hierarchy of goals — goals broken into subgoals, each level pairing neocortical simulation with basal ganglia automation (Ch 14)

Human hive mind — humans sharing and accumulating simulations across brains and generations (Ch 19, 21)

Imperative label — a signal that gets a reward, such as “sit” for a treat; any vertebrate can learn one (Ch 19)

Inductive bias — an assumption built into a system’s design, such as translational invariance in CNNs (Ch 7)

Inverse reinforcement learning — inferring the reward an expert is pursuing, then learning against that reward (Ch 17)

Invariance problem — recognizing the same thing despite changes in rotation, distance, location or pitch (Ch 7)

Language curriculum — hardwired instincts for proto-conversation, joint attention and questions that train the neocortex to learn language (Ch 20)

Law of effect — Thorndike’s rule that responses with satisfying effects become more likely and those with discomforting effects less likely (Ch 5)

Mentalizing — simulating one’s own inner simulation, that is, thinking about your own thinking (Ch 16, 18)

Model-based / model-free RL — choosing by simulating outcomes first, versus choosing directly from the current situation (Ch 12, 13)

Mountcastle’s crazy idea — the neocortex is one repeated microcircuit (the column) doing the same computation everywhere (Ch 11)

Myth of layers — Bennett’s name for MacLean’s discredited triune brain of reptile, limbic and neocortical layers (Ch 0)

Neocortex — the six-layered cortex new in mammals; the seat of simulation (Ch 10, 11)

Neuromodulator — a chemical with slow, widespread effects, such as dopamine, serotonin, adrenaline or opioids (Ch 3)

Obstetric dilemma — big-brained babies must pass through hips narrowed by upright walking, so humans are born early (Ch 21)

Paper-clip problem — an obedient optimizer that follows the literal request because it cannot infer what the requester intended (Ch 22)

Predictions, not commands — the motor cortex predicts movements and is wired to make them come true (Ch 14)

Primate sensory cortex (PSC) — Bennett’s name for the new primate sensory areas, the STS and TPJ (Ch 16)

Reinforcement learning — “the ability to learn arbitrary sequences of actions through trial and error” (Ch 5)

Social-brain hypothesis — primate brains grew to meet the demands of social life (Ch 15)

Social projection theory — understanding others by simulating yourself in their situation (Ch 16)

Spontaneous recovery / reacquisition — extinguished associations return with time and are relearned faster than new ones (Ch 4)

Steering — navigating by going forward when things get better and turning when they get worse (Ch 2)

Temporal difference learning — reinforcing behavior by changes in predicted future reward rather than by actual reward (Ch 6)

Theory of mind — the ability to infer another animal’s intent and knowledge (Ch 15, 16)

Transmissibility beats ingenuity — a skill needs inventing only once if others can copy it (Ch 17)

Valence — the goodness or badness of a stimulus, shown by whether an animal approaches or avoids it (Ch 2)

Vicarious trial and error — pausing to play out options in imagination before choosing (Ch 12)

Wallace’s problem — the difficulty of explaining how language evolved (Ch 21)

Wanting vs liking — dopamine drives the anticipation of pleasure, not pleasure itself (Ch 3)

Words without inner worlds — LLMs predict words without a simulation of the world beneath them (Ch 22)

World model — an inner simulation that can be explored to predict the results of actions (Ch 9, 11, 22)