We scan new podcasts and send you the top 5 insights daily.
Schmidt uses a thought experiment: train a model on all math known by 1902 and ask it to discover general relativity. Current evidence says it cannot. He describes today’s AI as knowing every fact and formula but real discovery still five to ten years away.
Observed AI mathematical breakthroughs operate largely within the convex hull of existing knowledge, recombining recognized results rather than creating novel frameworks. Generating genuinely surprising hypotheses or proposing the next foundational open question remains an unresolved challenge for automated systems.
An OpenAI model, without any specific mathematical training, solved a famous 80-year-old math problem. This proves general-purpose AI can autonomously produce landmark scientific results, not just accelerate human research. It signals a new era for discovery where AI is a primary research agent.
AI models advanced in pure mathematics faster than in theoretical physics because math offers self-contained verification methods. This immediate, objective feedback loop makes mathematical proof search ideal for reinforcement learning, whereas evaluating theoretical physics requires external, less formalized criteria.
An AI model disproved a mathematical conjecture not through a flash of creative genius, but by methodically applying a known technique from a different math subfield. This highlights AI's current strength: synthesizing vast, disparate human knowledge rather than generating truly novel, alien ideas. It's an exhaustive librarian, not an intuitive genius.
AI's capabilities are highly uneven. Models are already superhuman in specific domains like speaking 150 languages or possessing encyclopedic knowledge. However, they still fail at tasks typical humans find easy, such as continual learning or nuanced visual reasoning like understanding perspective in a photo.
Asked whether AI can achieve zero-shot innovation, Socher says most breakthroughs combine smaller past innovations. Einstein's relativity sat atop calculus and other prior work. He argues vast underexplored combinations remain, so AI can generate novel-seeming ideas, though people underestimate how much builds on existing knowledge, like new proteins from known amino acids.
AGI won't be achieved by pattern-matching existing knowledge. A real benchmark is whether a model can synthesize anomalous data (like Mercury's orbit) and create a fundamentally new representation of the universe, as Einstein did, moving beyond correlation to a new causal model.
AI is not just a digital tool; it is solving complex math problems that underpin our understanding of physics. This could unlock new, powerful capabilities, similar to how discovering nuclear physics changed the world, by revealing more of the universe's fundamental "code."
Even if AI could instantly prove any mathematical claim, it wouldn't end the field. The truly creative and valuable work in mathematics lies in higher-level tasks AI can't do: asking interesting questions, identifying fruitful problems, and inventing entirely new branches of mathematics like calculus or information theory.
The AI's breakthrough on this notoriously difficult math problem demonstrates novel, elegant problem-solving, not just pattern regurgitation. This refutes the common argument that AI is fundamentally limited by the human knowledge it was trained on.