The Good Fight Geoffrey Hinton on Artificial Intelligence
Oct 30, 2025
In a compelling discussion, Geoffrey Hinton, the 'godfather of AI' and 2024 Nobel Prize winner, dives into the intricacies of AI and its risks with Yascha Mounk. They explore how neural networks surpass traditional AI by focusing on perception, the significance of intuitive versus deliberate reasoning, and the critical role of self-supervised learning in language models. Hinton also addresses concerns about local optima in gradient methods and articulates how AI can evolve through scaling and innovative architectures, challenging traditional symbolic approaches.
AI Snips
Chapters
Transcript
Episode notes
Intuition Before Logic
- Human thinking mixes fast intuition and slow logical reasoning, so starting with perception and intuition is more fruitful for AI than beginning with logic.
- Geoffrey Hinton argues building systems that mirror animal perception yields stronger foundations for later reasoning.
How Vision Learns Hierarchies
- Vision networks learn hierarchical features: edges, motifs, object parts, then whole objects.
- Hinton explains how layers of detectors combine low-level edges into high-level bird features.
Use Gradients Not One-By-One Mutations
- Use backpropagation to compute gradients for all weights simultaneously instead of changing one weight at a time.
- Change all weights together using the gradient to learn efficiently.