On Point with Meghna Chakrabarti The ‘Godfather of AI’ says we can’t afford to get it wrong
19 snips
Dec 29, 2025 
Geoffrey Hinton, dubbed the 'Godfather of AI,' shares insights as a pioneering AI researcher and recent Nobel Prize winner. He discusses the evolution of neural networks, emphasizing their similarities to biological learning. Hinton delves into the risks of superintelligent AI, expressing concerns about extinction and control. He challenges misconceptions about AI's cognition and connects human errors to AI confabulations. Finally, he highlights the importance of prioritizing safety in AI development as we navigate this complex landscape.
AI Snips
Chapters
Transcript
Episode notes
Learning Lives In Connections
- Neural networks simulate many neurons and learn by changing connection strengths rather than millions of lines of code.
- Geoffrey Hinton emphasizes that the code is small; the learned connection weights are the large, data-driven part of the system.
Childhood Doubt Fueled Persistence
- Hinton credits a childhood experience of being the only skeptic at a Christian school for resilience in pursuing neural nets.
- That sustained contrarian experience helped him persist when the field was unpopular.
Knowledge Is Feature Interactions
- Language models convert words into feature activations and learn interactions that predict next words, not memorized strings.
- Hinton says knowledge resides in feature interactions, which is why models resynthesize and sometimes hallucinate.