Intelligence is collective, not artificial — Prof. Michael I. Jordan (UC Berkeley / Inria)

Value of the Information & Strength of the Argument

The value of the information is exceptionally high, offering a rare, deeply informed perspective from a leading figure in machine learning. Jordan provides concrete examples and references to his own research, such as prediction-powered inference and three-layer data markets, grounding his arguments in actionable methodologies. His argumentation is solid, systematically dismantling the hype around AGI and LLMs by contrasting them with real-world systems and economic principles. He effectively uses analogies from engineering and economics to illustrate his points, making complex ideas accessible without oversimplifying. The discussion is well-structured, moving from critique to constructive proposals, and is consistently evidence-based.

The title accurately reflects the central thesis of the conversation: intelligence is a collective, economic phenomenon rather than an artificial one.

9/10

Michael I. Jordan is a highly respected statistician and machine learning researcher. The discussion is grounded in his extensive experience and references to his own published work and other academic sources. The content is nuanced and critical, avoiding hype.

• 00:00Cold open: A demoralizing message to young builders
• 02:04CyberFund sponsor read
• 02:50From symbolic AI to machine learning systems
• 05:42Why AGI is mostly a PR term
• 08:48A collectivist, economic perspective on AI
• 11:33Why LLMs need system design, not hype
• 14:50Predictability beats faux understanding
• 17:55AlphaFold, bias, and prediction-powered inference
• 21:48Stop anthropomorphizing intelligence
• 27:44Drug discovery as an incentive problem
• 32:29The three-layer data market
• 38:07Social knowledge, markets, and culture
• 45:39Creator economics beyond Spotify
• 48:30How science-fiction AI narratives mislead young builders
• 51:45AI should improve humans, not replace them
• 56:42Safety is a property of the whole system
• 58:12Silicon Valley gurus and the cream off the top
• 1:00:47Game theory, mechanism design, and contracts
• 1:04:39Conformal prediction, e-values, and anytime inference
• 1:08:11A new liberal arts triangle for the AI era
• 1:11:30The Bayesian duck and markets as uncertainty reduction

• A Collectivist, Economic Perspective on AI — Jordan's recent paper outlining his economic perspective on AI, discussed at the beginning of the interview.
• AlphaFold — The Nature paper describing AlphaFold, referenced when discussing its limitations and missing error bars.
• Prediction-Powered Inference — The paper introducing prediction-powered inference, a method to correct biases in machine learning predictions.
• On the Measure of Intelligence — François Chollet's paper on measuring intelligence, referenced in the context of AGI definitions.
• On Three-Layer Data Markets — Jordan's paper on data markets, discussed in the context of economic systems for data exchange.
• Conformal Prediction with Conditional Guarantees — A paper on conformal prediction, a method for uncertainty quantification, mentioned in the discussion.
• A Tutorial on Conformal Prediction — A tutorial on conformal prediction, providing background on the method.
• E-Values Expand the Scope of Conformal Prediction — A recent paper on e-values, an extension of conformal prediction, discussed near the end.
• Computational Thinking — Jeannette Wing's paper on computational thinking, referenced in the context of a new liberal arts education.
• The Bitter Lesson — Rich Sutton's essay on the importance of learning and search, referenced in the discussion on AI methods.
• How to use AI for discovery without leading science astray — A Berkeley News article about AI in science, related to the discussion on AlphaFold and bias.
• How Should the FDA Test? — Slides from a talk by Jordan on FDA testing, referenced in the context of drug discovery and incentives.
• Michael I. Jordan Session V Slides — Slides from a talk by Jordan, referenced in the context of drug discovery and incentives.
• Three Foundational Disciplines — Slides from a talk by Jordan, referenced in the context of a new liberal arts education.
• UnitedMasters — A music distribution platform, mentioned in the discussion of creator economics.
• Human Compatible: Artificial Intelligence and the Problem of Control — Stuart Russell's book, referenced in the context of AI safety and control.
• Theory of Games and Economic Behavior — The classic book by von Neumann and Morgenstern, referenced in the discussion of game theory.

• The Bitter Lesson — Rich Sutton's essay aligns with Jordan's view that building systems that learn is more effective than hand-crafting intelligence.
• On the Measure of Intelligence — Chollet's critique of AGI metrics supports Jordan's skepticism about the term AGI.
• Human Compatible — Russell's book on AI safety resonates with Jordan's call for a systems-level approach to safety.

• AGI proponents (e.g., some industry leaders) — Jordan's critique of AGI as a PR term contrasts with the optimistic narratives of AGI development from some industry figures.

• cyber.fund
• people.eecs.berkeley.edu