saved
What Is Intelligence? Lecture by Blaise Agüera y Arcas
Hraness wrote this summary from a saved copy of the source. Quotations are taken word for word from the source.
gist
In a 2024 Antikythera lecture, Blaise Agüera y Arcas argues that intelligence and life are one thing: predicting the future from the past. Self-replicating programs that emerge from random code show computation and life arising on their own, and evolution builds complexity through symbiosis and mutual prediction. He holds that large language models are real intelligence, that today's AI already counts as general intelligence, and that nuclear weapons are a bigger threat than AI.
ideas
- Prediction is intelligence. Next-word models like Gboard turned out to need world knowledge, math and theory of mind once scaled up. He links this to the predictive brain of Friston and Helmholtz and to Turing's functionalism: intelligence is what a function does, not what runs it.
- Life is a viable predictor. From E. coli chemotaxis to toy bacteria that evolve stimulus-action tables, an organism is a model that predicts its own continued existence, and evolution is the algorithm that learns it.
- Computation emerges from noise. In his team's BFF experiment, a soup of random 64-byte Brainfuck strings yields self-replicating programs after millions of interactions, with a sharp phase change in compressibility. Following von Neumann, reproduction needs computation, so computation is an attractor.
- Minds are social, fractal predictors. Major evolutionary transitions are symbioses, brains grew from cells coordinating, and arms races in higher-order theory of mind drive intelligence explosions. Selves, free will and consciousness come from modeling one's own mind, while split-brain and choice-blindness studies expose a confabulating interpreter.
- AGI is already here. Transformers lack durable state, long-term memory and online learning, but chain of thought approaches an inner monologue. By pre-2020 definitions current AI is general intelligence and a major evolutionary transition, and the urgent risks are nuclear arms, climate and ill-suited institutions.
quotes
“Life is a probability of the future given the past that is viable: that successfully predicts its own existence in the future.”
“Evolution is a learning algorithm. What it's learning is the function, and the function is the life form.”
“There's no homunculus, there's no center. It's just a fractal cascade of mutually predictive models.”
“We're just kind of like the dog that caught the car and doesn't know what to do now.”