Key Moments
Michael I. Jordan: Machine Learning, Recommender Systems, and Future of AI | Lex Fridman Podcast #74
Lex FridmanKey Moments
Michael I. Jordan discusses AI's evolution, from engineering to a human-centric discipline, emphasizing value creation beyond mere prediction.
Key Insights
AI is evolving into a new branch of engineering, analogous to chemical or electrical engineering, focused on creating real-world value.
True understanding of the human brain is centuries away; current AI inspiration is metaphorical, not based on deep neural principles.
"AI" as a field is broad, encompassing systems making decisions under uncertainty at scale, distinct from the philosophical dream of mimicking human intelligence.
Current AI successes are engineering achievements, not necessarily scientific breakthroughs or indicators of approaching human-level intelligence.
Effective AI systems require a blend of prediction and decision-making, with a critical need to consider risks, externalities, and human impact.
The advertising-driven business model of many online platforms limits their potential for creating genuine producer-consumer economic relationships and value.
Future AI development should focus on human-centric engineering, empowering individuals and creating new markets rather than replacing human roles.
AI AS ENGINEERING, NOT MERE SCIENCE FICTION
Michael I. Jordan likens the current stage of AI development to the emergence of chemical and electrical engineering from their foundational sciences. He argues that AI is becoming a practical engineering discipline focused on building valuable systems using statistical and computational principles. This is distinct from the historical philosophical aspirations of AI, such as mimicking human thought, which he believes are still centuries away due to our limited understanding of the brain.
THE DISTANCE BETWEEN MAN AND MACHINE INTELLIGENCE
Jordan emphasizes that our understanding of the human brain is fundamentally limited, comparing current metaphors about AI to ancient Greeks speculating about space travel. Neuroscientists agree that comprehending even a single synapse is a monumental task. He cautions against the overselling of "brain-inspired" AI, particularly by those seeking funding, and distinguishes between the scientific understanding of intelligence and the engineering feat of creating useful, albeit not truly intelligent, systems.
REDEFINING ARTIFICIAL INTELLIGENCE BEYOND PREDICTION
Jordan clarifies that the current wave of AI success is not about replicating human intelligence or achieving perfect prediction. Instead, it's about building scalable systems that make consequential decisions under uncertainty. He contrasts this with the emphasis on prediction by some in the field, such as Yann LeCun, arguing that decision-making, risk assessment, and understanding consequences are equally, if not more, critical, especially in real-world applications.
CREATING NEW MARKETS AND EMPOWERING CREATORS
A key focus of Jordan's work is the creation of new markets enabled by AI, moving beyond the limitations of current advertising-driven models. He uses the example of a music market where creators could directly connect with consumers, foster careers, and generate income. This human-centric approach aims to empower individuals by building transparent, valuable systems, creating jobs rather than replacing them, and enhancing human happiness through direct connections.
THE LIMITATIONS AND POTENTIAL OF RECOMENDER SYSTEMS
Recommender systems are crucial for connecting consumers with creators and content. While algorithms are not magic, effective systems can significantly improve user experience, as seen with early Amazon recommendations. However, Jordan notes that systems primarily driven by advertising or simple click-through rates can be problematic, leading to issues like clickbait and the spread of misinformation. He advocates for models that facilitate genuine producer-consumer economic value exchange.
THE CRITICAL ROLE OF BUSINESS MODELS AND ETHICS
Jordan critiques the dominance of advertising-based business models in platforms like Google and Facebook for incentivizing clickbait and potentially harmful content. He suggests that a shift towards direct consumer-producer payment models, even micro-payments, could foster more genuine value creation and ethical practices. Such models require understanding economic principles, respecting culture, and building trust, rather than solely maximizing ad revenue.
PRIVACY, CONTROL, AND THE FUTURE OF HUMAN-TECHNOLOGY INTERACTION
The conversation explores privacy not as a binary issue but as a complex interplay of control and trust. Jordan argues that technology should empower users, providing transparency and agency over their data and interactions. He contrasts the exploitative data practices of some companies with the potential for trusted relationships, likening it to a reliable "old uncle" rather than an intrusive "big brother." Developing these new structures and ethical frameworks is a long-term engineering challenge.
STOCHASTICITY, OPTIMIZATION, AND THE NATURE OF REALITY
Jordan delves into the mathematical underpinnings of AI, highlighting the importance of stochasticity (randomness) in optimization problems. He explains how randomness can help algorithms navigate complex, non-convex landscapes more effectively than purely deterministic methods. This reflects his view that the world is inherently stochastic, and understanding these principles is crucial for building robust AI systems that can operate in the real world.
THE DEBATE BETWEEN BAYESIAN AND FREQUENTIST STATISTICS
He unpacks the fundamental differences between Bayesian and frequentist statistical approaches, linking them to decision theory. Frequentist methods offer guarantees based on hypothetical repeated experiments, suitable for software reliability. Bayesian methods incorporate prior beliefs and focus on the observed data, useful for incorporating human expertise. He advocates for a blend, like empirical Bayes, to leverage the strengths of both in practical applications.
INTELLIGENCE BEYOND HUMANITY: MARKETS AS SYSTEMS
Jordan posits that intelligence is not limited to human cognition. He considers robust, adaptive systems like markets or even decentralized economic networks as forms of intelligence. Studying these systems, he argues, offers more relevant principles for future AI, such as matching and auctions, than trying to solely understand human intelligence, which remains a distant and complex goal.
ADVICE FOR ASPIRING AI PROFESSIONALS
Jordan advises aspiring AI professionals to embrace a journey of lifelong learning, focusing on hard problems, broad exploration, and collaboration. He stresses the importance of apprenticeship, human connections, and a cooperative spirit, likening the path to becoming an artist or musician. He encourages study across disciplines like math, poetry, history, and languages to foster a well-rounded, critically thinking individual.
Mentioned in This Episode
●Software & Apps
●Companies
●Organizations
●Books
●Concepts
●People Referenced
Common Questions
Michael I. Jordan views the current developments not as 'AI' in the sense of creating human-like intelligence, but rather as the emergence of a new engineering discipline, akin to chemical or electrical engineering. This field builds viable systems that use human data and decisions at scale, valuing practical, understandable, and robust solutions over attempts to mimic human thought.
Topics
Mentioned in this video
Professor at Berkeley and influential figure in machine learning, statistics, and AI, cited over 170,000 times. Lex Fridman's guest.
World-class researcher in AI, mentored by Michael I. Jordan.
Mathematician from whom ideas like empirical Bayes and FDR emerged around 1960.
Provided a Bayesian interpretation of the False Discovery Rate.
World-class researcher in AI, mentored by Michael I. Jordan.
World-class researcher in AI, mentored by Michael I. Jordan.
Basketball player mentioned for comparison to Michael I. Jordan's influence.
Statistician who has written beautifully about empirical Bayes and related topics.
Jazz musician used as a metaphor for Michael I. Jordan's explorative and reinventive career.
Colleague at the University of Chicago who wrote a beautiful paper on James-Stein estimation.
Fellow AI researcher who calls Michael I. Jordan the 'Miles Davis of machine learning'.
Pioneer in the development of the False Discovery Rate concept.
World-class researcher in AI, mentored by Michael I. Jordan.
CEO of Neuralink and SpaceX, mentioned for his views on AI and brain-computer interfaces.
Key figure in cybernetics, mentioned as a contrast to John McCarthy's coining of AI.
Coined the term 'Artificial Intelligence' as a philosophical aspiration, distinct from cybernetics.
CEO of Microsoft, credited with changing the company's approach to user trust and privacy.
Streaming service that created content but is viewed as top-down, not fostering cultural ecosystems.
Praised as a breakthrough for its early recommender systems and its focus on creating genuine markets, unlike advertising-first models.
Praised for pivoting under Satya Nadella to prioritize user trust and control over data and privacy.
Praised as a breakthrough for its search engine and also criticized for its advertising-dependent business model.
Music distribution company that Michael I. Jordan is on the board of, praised for creating a viable music market for artists.
Heavily criticized for its advertising-based business model, lack of transparency, and 'creepy' use of personal data.
Platform mentioned for its potential for market creation and issues with comment sections.
Elon Musk's company working on brain-computer interfaces with electrodes.
Streaming service that monetizes through subscriptions or advertising, with an aspiration to help 1 million creators make a living.
Cited as a breakthrough for its innovative service delivery and market creation using data flows.
A surprising and deep idea in optimization that improves gradient descent's convergence rate, often by introducing momentum.
A surprising and counter-intuitive idea in statistics that provides better estimates than classical methods, often seen as a paradox.
A function that a particle in a potential well is said to optimize, illustrating how optimization describes natural phenomena.
A criterion for evaluating multiple hypothesis tests, focusing on the fraction of false discoveries among all discoveries made.
A statistical approach that blends Bayesian framework with frequentist estimation to produce robust, justifiable results.
Finance app that allows sending money, buying Bitcoin, and investing in the stock market with fractional shares.
Platform where creators upload music, but which doesn't facilitate direct economic value for creators.
Open-source machine learning framework, mentioned in the context of young researchers thinking problems are solved.
Product by Facebook designed to recommend news, but tied to an advertising model.
Amazon's voice assistant, discussed as a research platform with potential for useful, non-intrusive AI interactions.
Blogging platform where Lex Fridman published a post, mentioned as a site where micro-payments could be useful.
Podcast by The New York Times, praised for its quality journalism.
More from Lex Fridman
View all 505 summaries
154 minRick Beato: Greatest Guitarists of All Time, History & Future of Music | Lex Fridman Podcast #492
23 minKhabib vs Lex: Training with Khabib | FULL EXCLUSIVE FOOTAGE
196 minOpenClaw: The Viral AI Agent that Broke the Internet - Peter Steinberger | Lex Fridman Podcast #491
266 minState of AI in 2026: LLMs, Coding, Scaling Laws, China, Agents, GPUs, AGI | Lex Fridman Podcast #490
Found this useful? Build your knowledge library
Get AI-powered summaries of any YouTube video, podcast, or article in seconds. Save them to your personal pods and access them anytime.
Try Summify free