A Brief History of Intelligence: Evolution, AI, and the Five Breakthroughs That Made Our Brains
by Max Solomon Bennett
Rating: 4.0 / 5 · Read · Finished 2024-12-09
Very technical deep dive into intelligence as it evolved from RNA all the way up to humans. It names lots of brain parts, which is boring, but it also is full of super interesting ideas. Also cool to see parallels between organic intelligence and artificial intelligence.
Most evolved intelligence operates on simple heuristics.
Precursors to emotions: Arousal and Affect. Arousal: Do I want to expend energy? Affect: Do I want to remain here or go somewhere else? High arousal + Positive affect: excitement, expectation. Try hard to get the reward. High arousal + Negative affect: anxiety, anger, fear. Fight or flight. Low arousal + Positive affect: contentment. enjoy the good times and don’t mess things up. Low arousal + Negative affect: depression. save energy and hope the situation changes.
Dopamine = high arousal = wanting = food is nearby Serotonin = low arousal = liking = food is inside
Temporal Difference Learning - Assume that later predictions are more accurate than earlier predictions, and use that to improve predictions. Then reinforce behaviors that maximize predicted reward.
Dopamine spikes when you are given stimulus that predicts reward. It drops when expected reward is not given. Dopamine started as a signal to trigger exploitation behavior, but it evolved into a reinforcement signal. Disappointment and relief are as reinforcing as reward and punishment.
Pattern recognition came in the form of smell. It’s the first sense that recognizes a pattern of neurons rather than particular neurons.
Predicting, imagining, remembering and dreaming are all basically the same thing: simulation.
Constant passive prediction means you can ignore almost all sensory input like the feeling of your feet on the ground, but get alerted if anything unexpected happens, like stepping in a pothole
Regret is simulating the good thing that didn’t happen.
Episodic memory is a simulation of the past.
Model based AI = goal seeking behavior in animals = “slow” thinking from “Thinking fast and slow” = logic Model-free AI = habitual behavior in animals = “fast” thinking from “Thinking fast and slow” = intuition