scripod.com

Adam Marblestone — AI is missing something fundamental about the brain

Overview

Shownote

Highlights

Transcript

Chapters

Pins

Adam Marblestone — AI is missing something fundamental about the brain

Dwarkesh Podcast

2025/12/30

Adam Marblestone — AI is missing something fundamental about the brain

Adam Marblestone — AI is missing something fundamental about the brain

Dwarkesh Podcast

Dwarkesh Podcast

2025/12/30

Overview Shownote Highlights Transcript Chapters Pins

Shownote

Adam Marblestone is CEO of Convergent Research. He’s had a very interesting past life: he was a research scientist at Google Deepmind on their neuroscience team and has worked on everything from brain-computer interfaces to quantum computing to nanotech an...

Highlights

In this deep exploration of neuroscience and artificial intelligence, Adam Marblestone examines the fundamental differences between how human brains and AI systems learn, focusing on the unique mechanisms that give biological intelligence its remarkable efficiency and adaptability.

09:12

The cortex learns to predict the steering subsystem's innate responses to guide learning.

41:36

Social reward functions in the steering subsystem require vision and audio to understand cues for learning

48:01

Human cultural evolution operates like model-free reinforcement learning over generations.

1:01:42

TD learning may be implemented in the brain through dopamine signaling

1:17:31

Gwern proposed using neural activity patterns as an auxiliary prediction task to improve model generalization

Chapters

The brain’s secret sauce is the reward functions, not the architecture

00:00

Amortized inference and what the genome actually stores

22:20

Model-based vs model-free RL in the brain

42:42

Is biological hardware a limitation or an advantage?

50:31

Why a map of the human brain is important

1:03:59

What value will automating math have?

1:23:28

Architecture of the brain

1:38:18

Transcript

Dwarkesh Patel: The big million-dollar question that I have, that I've been trying to get the answer to through all these interviews with AI researchers, how does the brain do it, right? Like, we're throwing way more data at these LLMs, and they still have...