Super Data Science: ML & AI Podcast with Jon Krohn · Jon Krohn

949: Why AI Keeps Failing Society, with Stanford professor Alex “Sandy” Pentland

·55 min·3 clips

Alex Pentland links the Soviet Union’s collapse to linear AI models that failed on bad data and rare events.

1. Super Data Science: ML & AI Podcast episode 949 centers on Alex Sandy Pentland’s book Shared Wisdom and its argument about AI and society. 2. Jon Krohn hosts, and Alex Sandy Pentland is a Stanford and MIT professor who pioneered computational social science. 3. The episode asks why AI systems fail when they ignore human context, social behavior, and institutional design. 4. Pentland starts with the late 1950s and early 1960s origins of AI, naming John McCarthy and Marvin Minsky. 5. He says the Soviet Union’s use of linear constraints and optimal resource allocation worked until bad data and rare events broke the model. 6. Pentland says expert systems are still everywhere in corporate approvals, audits, and the tax system, but they are optimized for the average. 7. He argues that average-based systems can fit neither people who are slightly off-average nor people who are far from average. 8. Pentland links centralized systems to the loss of local clinics, local banks, and the “bowling alone” dynamic. 9. He describes machine learning as backward-looking and says it can produce echo chambers when social media is built around feature vectors instead of communities. 10. He says loud voices on social media drive polarization, while removing those voices reveals that most people agree on many issues. 11. Pentland cites experiments with currency traders moving $10 million a day and says they do better when they pay attention to other traders. 12. He uses the 2008 financial crisis to show how tail risk and unusual events can topple equations that ignore human behavior. 13. Pentland says the book’s practical focus is on making AI systems robust to weird edge cases rather than replacing people. 14. He describes loyalagents.org as a project to build AI agents that serve the user, stay legal, and handle edge conditions more reliably. 15. Pentland describes an AI buddy that reads manuals, newsletters, and company activity to keep employees in the loop without making decisions for them. 16. He says deliberation.io works like Twitter or X without big voices, using visualization and Socratic dialogue to summarize group views. 17. Pentland says people in Washington, D.C. wanted personal AI to defend them against city bureaucracy, fill forms, and surface opportunities. 18. The interview format is energetic and idea-dense, with Pentland moving quickly from theory to projects, examples, and policy. 19. Listeners interested in AI governance, social science, and organizational design will get the most from it. 20. Listeners wanting a lightweight AI product tour with no policy or systems discussion may skip it.

As heard by us

A practical warning about AI that leans into mediation and civic design.

Jon Krohn and Alex Sandy Pentland treat AI less as a product pitch than as a social systems problem, and that choice gives the discussion real shape. They focus on polarization, information environments, and the way extreme voices can bend the apparent center of opinion.

Read the full review in PlayNext →

Why you'd press play

Hear why a Stanford and MIT professor says AI could put society on a crash course.

Read the full recommendation in PlayNext →

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from Super Data Science: ML & AI Podcast with Jon Krohn