Terence Tao on AI in mathematics, verification and inequality

Summary

In an interview conducted in July 2026, mathematician Terence Tao argues that AI should be discussed with nuance rather than as an unqualified good or bad. He discloses that he has received premium model access and has academic and nonprofit ties involving AI companies and funding, while saying he is not directly paid by them and does not feel pressured; he acknowledges that these relationships create incentives. Tao says mathematics is unusually suited to AI because proofs can sometimes be checked mechanically, but that success cannot be transferred uncritically to fields without comparable safeguards. He calls for verification and input from disciplines including law, economics, and the humanities. On education, he warns that convenient AI assistance can bypass intellectual effort and says schools will need to teach students to manage their use of these tools. He also recognizes inequality risks from proprietary frontier models, but argues that open models, local deployment, cheaper systems, and public or philanthropic support could broaden access. Tao says AI-generated proofs create a need to value clear explanation and human understanding, not just correctness; he expects journals and educators to develop standards for assessing whether work is well explained, while acknowledging that such judgments are more subjective. He cites his tracking of AI-assisted solutions to Erdős problems as a case where scrutiny was followed by fewer misleading company claims, though he cannot establish causation. On AI originating fruitful mathematical concepts without human help, he sees the possibility but offers no reliable timeline, saying forecasting has become highly uncertain. He remains willing to use frontier models but is content with current assistance and envisages delegating tasks to less powerful models or scripts. Overall, he presents AI’s effects as contingent on safeguards, access, education, and community choices.