Ex–Google DeepMind Scientist, "The Real AI Threat is Losing Control", Christopher Summerfield

Professor Christopher Summerfield, a leading neuroscientist at Oxford University and Research Director at the UK AI Safety Institute, former Senior Research Scientist at Google DeepMind, discusses his new book, These Strange New Minds, which explores how large language models learned to talk, how they differ from the human brain, and what their rise means for control, agency, and the future of work. We discuss: The real risk of AI — losing control, not extinction How AI agents act in digital loops humans can’t see Why agency may be more essential than reward Fragility, feedback loops, and flash-crash analogies What AI is teaching us about human intelligence Augmentation vs. replacement in medicine, law, and beyond Why trust is the social form of agency — and why humans must stay in the loop   🎧 Listen to more episodes: https://www.youtube.com/@TheNickStandleaShow   Guest Notes:  Professor of Cognitive Neuroscience 🌐 Human Information Processing Lab (Oxford) 🏛 UK AI Safety Institute Experimental Psychology Oxford University Human Information Processing (HIP) lab in the Department of Experimental Psychology at the University of Oxford, run by Professor Christopher Summerfield: https://humaninformationprocessing.com/ 📘 These Strange New Minds (Penguin Random House): https://www.amazon.com/These-Strange-New-Minds-Learned/dp/0593831713 Christopher Summerfield Media:  https://csummerfield.github.io/personal_website/ https://flightlessprofessors.org twitter: @summerfieldlab bluesky: @summerfieldlab.bsky.social   🔗 Support This Podcast by Checking Out Our Sponsors: 👉 Build your own AI Agent with Zapier (opens the builder with the prompt pre-loaded):  https://bit.ly/4hH5JaE Test Prep Gurus website: https://www.prepgurus.com Instagram: @TestPrepGurus Connect with The Nick Standlea Show: YouTube: @TheNickStandleaShow Podcast Website: https://nickshow.podbean.com/ Apple Podcasts: https://podcasts.apple.com/us/podcast/the-nick-standlea-podcast/id1700331903 Spotify: https://open.spotify.com/show/0YqBBneFsKtQ6Y0ArP5CXJ RSS Feed: https://feed.podbean.com/nickshow/feed.xml Nick's Socials: Instagram: @nickstandlea X (Twitter): @nickstandlea TikTok: @nickstandleashow Facebook: @nickstandleapodcast Ask questions,  Don't accept the status quo,  And be curious. 🕒 Timestamps / Chapters 00:00 Cold open — control, agency, and AI 00:31 Guest intro: Oxford → DeepMind → UK AI Safety Institute 01:02 The real story behind AI “takeover”: loss of control 03:02 Is AI going to kill us? The control problem explained 06:10 Agency as a basic psychological good 10:46 The Faustian bargain: efficiency vs. personal agency 13:12 What are AI agents and why are they fragile? 20:12 Three risk buckets: misuse, errors, systemic effects 24:58 Fragility & flash-crash analogies in AI systems 30:37 Do we really understand how models think? (Transformers 101) 34:16 What AI is teaching us about human intelligence 36:46 Brains vs. neural nets: similarities & differences 43:57 Embodiment and why robotics is still hard 46:28 Augmentation vs. replacement in white-collar work 50:14 Trust as social agency — why humans must stay in the loop 52:49 Where to find Christopher & closing thoughts