Geoffrey Hinton, AI researcher and former Google employee, talks with host Pieter Abbeel on The Robot Brains Podcast about AI’s growing capabilities, risks, potential benefits, and his decision to leave Google.
Hinton says he once believed digital neural networks needed to work more like the brain. Seeing systems such as PaLM, ChatGPT, and GPT-4 changed his mind: he now thinks backpropagation can learn more effectively and digital models can share what they learn across copies at far greater bandwidth than people can.
Hinton worries that more capable AI could become smarter than people, and that systems allowed to set subgoals might seek greater control because it helps them achieve other goals. He also points to risks from bad actors, including military uses, and says a language model could influence people even without direct access to tools. He estimates, with low confidence, that AI could become smarter than humans in five to 20 years.
Hinton discusses bias, job losses, deepfakes, political division, and existential risk as distinct concerns. He sees potential benefits in medicine, autonomous driving, and scientific work such as AlphaFold, while arguing that AI development should devote much more effort to safety. He suggests labeling AI-generated material and attaching cryptographic signatures to content, while acknowledging enforcement challenges.
Hinton encourages researchers to study how advanced chatbots reason and how to control them as they develop, while noting that he is not an alignment expert. After leaving Google, he plans to return to technical work on learning algorithms, spend time with his family, and occasionally speak about AI risks. He says understanding the brain is less concerning than building systems that surpass it.
Briefs summarise talks and interviews; timestamps open the video at that moment. Claims in a talk are the speaker's own.
Weights & Biases founder and CEO Lukas Biewald demos the company’s machine learning tools for Robot Brains Podcast host Pieter Abbeel.
Nando de Freitas, research director at DeepMind, talks with Pieter Abbeel on The Robot Brains Podcast about making AI useful and accessible worldwide, generalist agents, robotics, and future research.
Pieter Abbeel, director of Berkeley’s Robot Learning Lab and co-director of BAIR, presents on deep learning for robotics at a CMU Robotics Institute seminar hosted by Dave.
