Terence Tao, one of the world's leading mathematicians, discusses the intersection of artificial intelligence and mathematical research with the host.

The conversation begins with Kepler's discovery of planetary motion laws as an analogy for how AI might approach mathematical problem-solving - trying many random relationships until finding empirical regularities that can be verified against data.

Tao explores how AI tools are already transforming mathematical research, from solving dozens of previously unsolved problems to changing how mathematicians write papers and conduct literature searches.

The discussion covers the current limitations of AI in mathematics, the complementary relationship between human depth and AI breadth, and predictions for how mathematical research will evolve as AI capabilities advance.

Kepler as a High-Temperature LLM: Pattern Discovery Through Trial and Error

Kepler spent decades trying random geometric relationships - from Platonic solids to musical harmonies - before discovering his three laws of planetary motion through systematic data analysis of Tycho Brahe's observations.

"The take I want to try on you is that Kepler was a high-temperature LLM" - the host argues that AI could excel at this kind of exhaustive hypothesis testing when combined with verifiable datasets.

Kepler's third law appeared almost as an aside in a book about planetary harmonies and astrology, demonstrating how breakthrough discoveries can emerge from seemingly random exploration.

The key insight is that empirical regularities discovered through trial-and-error can drive deep scientific progress, as Newton later explained Kepler's laws through the inverse square law of gravity.

The Bottleneck Has Shifted from Ideas to Verification

"AI has basically driven the cost of idea generation down to almost zero" - Tao explains that the scientific bottleneck has shifted from hypothesis generation to validation and evaluation.

Traditional peer review systems are being overwhelmed by AI-generated submissions, requiring new structures to filter high-signal ideas from "AI slop."

Modern science increasingly starts with big data analysis rather than the classical method of forming hypotheses first, then collecting data to test them.

The challenge is developing systems that can assess which ideas represent genuine progress versus dead ends when generating thousands of theories daily.

AI's Current Mathematical Capabilities and Limitations

Over 50 Erdős problems have been solved with AI assistance in recent months, but "there's been a pause because the low-hanging fruit had been picked."

AI tools have approximately a 1-2% success rate on any given mathematical problem, but their ability to operate at scale makes them effective for systematic exploration.

"They excel at breadth and humans excel at depth" - AI can try standard techniques across many problems while humans are better at developing new approaches for resistant cases.

Current AI lacks the ability to build cumulative progress from partial results, operating more like "jumping robots" that either succeed or fail rather than learning from intermediate steps.

The Productivity Revolution in Mathematical Research

Tao reports being potentially 5x more productive on certain tasks: "The type of papers that I would write today, if I had to do them without AI assistance, they would definitely take five times longer."

AI has transformed secondary tasks like literature searches, formatting, generating plots and numerical data, allowing mathematicians to create richer, more comprehensive papers.

The core mathematical work - solving the most difficult parts of problems - remains largely unchanged, still requiring pen-and-paper human insight.

Writing and refactoring papers has become much easier, enabling mathematicians to create multiple versions and explore different presentations of their work.

The Need for Formal Languages Beyond Proofs

Tao advocates for developing "formal or semi-formal language for mathematical strategies as opposed to just mathematical proofs," which current systems like Lean cannot handle.

Mathematical progress often involves assessing plausibility, building narratives, and constructing arguments that combine data with subjective elements - areas where formal frameworks don't yet exist.

The prime number theorem exemplifies how mathematical conjectures emerge from statistical patterns in data, leading to conceptual frameworks that guide research despite being largely non-rigorous.

"We don't really have a way of measuring" scientific progress objectively, partly because we lack sufficient data on how mathematics and science develop across different contexts.

Comparing Scientific Breakthroughs: Newton vs. Darwin

The Clockwork Universe by Edward Dolnick illustrates how Darwin's On the Origin of Species (1859) came two centuries after Newton's Principia Mathematica (1687), despite being conceptually simpler.

"How stupid not to have thought of that" - Thomas Huxley's reaction to Darwin's theory contrasts sharply with the mathematical complexity that prevented anyone from anticipating Newton's insights.

The key difference lies in verification methods: Newton could test his equations against precise astronomical data, while Darwin's theory required cumulative, retrospective evidence.

This suggests AI may make faster progress in domains with tight data loops for verification, even when the underlying concepts are more mathematically sophisticated.

Future Predictions and Career Advice

Tao predicts that by 2026, AI will function as "a trustworthy co-author, if used correctly" in mathematical research.

"Hybrid human plus AIs will dominate mathematics for a lot longer" than purely autonomous AI solutions, requiring new collaborative paradigms.

For aspiring mathematicians: "You need a very adaptable mindset" as traditional educational paths may become less relevant while new opportunities emerge at earlier stages.

The field will become "unrecognizable" as AI enables exploration of entire new areas through broad, systematic mapping before human experts tackle the most difficult problems.