The historical process of scientific discovery, exemplified by Kepler using Brahe's data, is shifting.
AI has made hypothesis generation nearly free, making rigorous data analysis and verification the new bottleneck.
Terence Tao characterizes current AI as a powerful tool for mathematical 'breadth'—applying known techniques at scale—but it lacks the 'depth' for novel conceptual breakthroughs, with a success rate on truly hard problems remaining very low (1-2%).
Tao predicts a future dominated by hybrid human-AI collaboration.
AI will automate routine work and enable a new 'experimental' side of mathematics, while humans will focus on strategy, intuition, and creating new frameworks.
He stands by his 2026 prediction of AI as a 'trustworthy co-author' but believes a full replacement of human mathematicians is not imminent, citing AI's fundamental inability to build cumulatively on partial progress.
11 quotes
Concerns Raised
Current AI models cannot build on partial progress, limiting their ability to solve complex, multi-step problems.
The flood of AI-generated ideas and papers makes verification and quality control the new scientific bottleneck.
Over-reliance on AI for answers could inhibit the development of deep intuition and problem-solving skills in researchers.
Opportunities Identified
AI will function as a 'trustworthy co-author' by 2026, handling routine tasks and checking for errors.
AI will enable a new paradigm of 'experimental mathematics,' allowing for large-scale data analysis of mathematical structures and techniques.
Hybrid human-AI teams will dominate research by combining AI's breadth and speed with human depth and intuition.