Summary of A Brief History of Intelligence by Max Bennett
#books 3 min.
The fact that Brief History of Intelligence is written not by a researcher or academic but a CEO of an AI startup, might give you a pause - it certainly did it for me. He may have hired a great ghostwriter, his editor did a lot of the heavy lifting, or it’s all just raw talent, but you could never tell it’s a hobby project - the book is well-written, engaging, and informative.
The book does a good job summarizing the major evolutionary leaps that led to human-like intelligence, explaining the ingenious techniques and tricks evolution came up with. The author then draws parallels to the current AI tools, which are based on lessons from the natural world. The book, however, focuses only on large language models (LLMs), the AI branch that most readers are best familiar with.
It was so good that I decided to do something I’ve never done before—summarize the book in this blog for when I want to revisit it later.
Breakthrough #1 - steering
Steering allowed early organisms to navigate their environment by categorizing stimuli into “good” and “bad.” That kind of intelligence emerged six hundred million years ago from coral-like life. Hormones such as dopamine and serotonin were also a product of this breakthrough, which still guide many of our behaviors today. Pleasure, pain, satiation, and stress were also direct results.
Breakthrough #2 - reinforcing
Five hundred million years ago, life acquired the capability to learn to repeat behaviors that lead to good or bad outcomes. It happened within the first vertebrates, which today we would call fish. This breakthrough also gave us time perception, curiosity, fear, excitement, disappointment, and relief.
Breakthrough #3 - simulating
One hundred million years ago, in small mammals, a newly formed brain structure called the neocortex allowed for the simulation of possible outcomes of actions before taking them. It allowed mammals to plan or recall past events, leading to more complex behaviors and more efficient learning by imagination.
Breakthrough #4 - mentalizing
Ten to thirty million years ago, in early primates, the neocortex grew in size and complexity, allowing for mentalizing—predicting the behavior of themselves and others based on their mental states. Now, these animals could anticipate their own future needs, understand the intents and knowledge of others, and learn skills through observation! This, in turn, allowed for even more complex social structures, cooperation, and politics.
Breakthrough #5 - speech
The speech was the breakthrough that enabled the accumulation of thoughts and knowledge across generations. You no longer needed to see someone get poisoned by red berries to learn they are inedible - your mom telling “red berries bad” was enough. There is no unique brain structure responsible for speech in humans. What’s unique to humans is the learning “program” - a set of behaviors - that teaches you speech. That is why teaching children languages gets increasingly harder as they age.
Summary
Each breakthrough was built on top of the previous one. Who knows what breakthrough #6 will be, but it seems increasingly likely to be artificial superintelligence. Max argues that we are already seeing the first signs of it with large language models (LLMs) such as OpenAI’s GPTs. While still very different and inferior to humans, biology will not constrain their evolution. Considering it took only a few billion years to go from single-cell organisms to a human intellect, it’s hard to imagine what will become of us in another few billion years. Whatever it is, it’s sure to be exciting.