Part 1: Eric Schmidt and Fei-Fei Li: Human Life After Artificial Superintelligence | EP #206

Summary

• Schmidt puts true ASI beyond the industry’s “San Francisco consensus” of three to four years, even as compounding gains could pull the date forward. He defines it as intelligence equal to “the sum of everyone” or better than all humans, but today’s systems cannot quickly feed newly learned reasoning back into themselves; creative superintelligence may require changing objectives while operating. Real ASI “probably” needs “another algorithmic breakthrough.”
• Li says AI is already superhuman in translation, calculation and knowledge breadth, but not yet in creative abstraction. Given celestial observations, today’s algorithms would not deduce Newtonian motion; she asks whether AI can “ever be Newton,” Einstein or Picasso. That gap—and robotics’ still-distant human dexterity—makes her “not as bullish” on post-scarcity and more confident in human-AI collaboration.
• The episode’s economic split is democratized services versus concentrated surplus. Diamandis cites as much as 15 trillion in AI value by 2030, autonomous transportation four times cheaper than car ownership and free top-tier healthcare; Schmidt says network effects favor early adopters, well-run countries and perhaps capital, with 10%-20% efficiency gains in Saudi oil systems as the concrete prize. Li’s distinction: higher productivity “does not necessarily translate to shared prosperity,” which also depends on policy, geopolitics and distribution.
• Compute sovereignty is constrained by capital, chips and energy, so partnership may matter more than owning a national data center. Schmidt says U.S. capital markets plus TSMC’s chips give America a “huge lead,” puts China second, and points to Saudi/UAE hyperscalers and France’s Abu Dhabi partnership. Li says countries should invest in human capital, partnerships, their technology stacks and business ecosystems, but requiring every country to build data centers is too sweeping.
• Math and software should move first because their outputs are verifiable and can scale without waiting on physical reality. Schmidt expects the greatest gains there “in the next few years,” with cyberattacks in the same class; Diamandis offers a five-year horizon for reaching a position to solve everything and seeing super-exponential discovery. Li responds, “I actually want to respectfully disagree”: humans will keep inventing new questions, and the panel agrees to bet on the forecast.
• World Labs is Li’s work on large world models that supply spatial intelligence missing from large language models. She says the company has created “the first large world model” for understanding, imagining, reasoning about and interacting with 3D worlds, opening a physical-virtual hybrid across medicine, education, productivity, communication and entertainment.
• Human judgment remains central even if machines become far more capable. Schmidt says humanlike machine intelligence is unlikely and human exclusion is “highly unlikely”; the win is “teaming” between human judgment and supercomputer capability. He notes that supercomputers and superintelligence need energy. Diamandis imagines systems designing more chips or energy and accelerating fusion, labels that “science fiction,” and Schmidt agrees. Li closes by insisting that human dignity, agency and well-being remain central.

Deep dive

Not yet available upstream; scheduled sync will retry.