3 Questions: What is the best path forward for AI in academia?

Sasha Rakhlin, director of the MIT Statistics and Data Science Center and the Distinguished Professor in Data, Systems, and Society, published a new essay on how AI is changing academia, concentrating on mathematics, statistics, machine learning, and engineering but framed around graduate research and education. It synthesizes discussions and readings on the topic and follows an MIT report on AI and education released this summer.

Rakhlin points to mathematics to show the pace: a model reached gold-medal level at the International Mathematical Olympiad last year, and a year later models are producing new research results, including a proposed solution to one of the Millennium Prize Problems. He identifies verification speed and reliability as the factor governing progress in a field, since formalized proofs can be checked automatically and programs can be run and tested. Where evaluation is fast and reliable, he writes, systems can generate candidates, learn from outcomes, and improve, including in AI research itself, a compounding process that accelerates development.

For evaluation, Rakhlin argues a polished paper is becoming a weaker signal of individual expertise in a growing number of fields, and departments should reconsider what they reward without defining valuable work as whatever AI cannot yet do. Asking good questions, replication, synthesis, informative negative results, and shared datasets may deserve more recognition, and evaluation should establish what a researcher contributed and takes responsibility for, including when AI performed substantial parts of the work. Those expectations, he says, should guide hiring, promotion, and funding, and be made explicit for current and incoming PhD students. On training, he notes delegating routine calculations, coding, failed approaches, and small discoveries can remove formative experiences, and that students should learn to formulate problems, audit model outputs, reproduce results, and defend their choices.

For what universities should build, Rakhlin calls for long-horizon questions, open sharing of results, and independent evaluation of claims, with industry partnerships that do not assume commercial priorities cover the breadth of science. He proposes a future in which MIT laboratories act like a single scientific organism through shared AI research infrastructure, with an AI agent connecting researchers across groups, proposing benchmarks, and surfacing unresolved questions. That requires workflows capturing hypotheses, interventions, outcomes, failures, and interpretations, plus investment in compute, secure data systems, and expertise in adapting and post-training models.

Why it matters: departments, hiring bodies, and PhD programs are asked to revise credit and evaluation rules now, while universities would need new investment in compute, data systems, and model expertise to pursue the shared-infrastructure plan.