MIT essay maps how AI is reshaping academic research and credit
MIT’s Sasha Rakhlin outlines how fast AI is advancing in mathematics, why verification speed matters, and what universities should change in training and credit.
The story in focus
An MIT essay by Sasha Rakhlin examines how AI is changing graduate research and education, especially in mathematics, statistics, machine learning, and engineering. It urges departments to rethink credit and training, and to invest in shared AI infrastructure that captures hypotheses, failures, and interpretations.
- Rakhlin says fast, reliable verification helps explain rapid AI progress in some fields.
- A polished paper is becoming a weaker signal of individual expertise in a growing number of fields.
- Rakhlin proposes shared AI research infrastructure to connect laboratories and capture tacit knowledge.
Why progress accelerates in some fields
Rakhlin writes that mathematics shows how quickly AI capabilities are advancing: a model reached gold-medal level at the International Mathematical Olympiad last year, and only a year later models are producing new research results, including a recently proposed solution to one of the Millennium Prize Problems.
He identifies verification speed and reliability as a central factor in how fast AI progresses within a discipline. Formalized proofs can be checked automatically, and programs can be run and tested. Where evaluation is fast and reliable, systems can generate candidates, learn from outcomes, and improve, creating a compounding process.
Rethinking credit and training
In a growing number of fields, Rakhlin argues, a polished paper is becoming a weaker signal of individual expertise. Departments should reconsider what they reward without simply defining valuable work as whatever AI cannot yet do. Asking good questions, replication, synthesis of ideas, informative negative results, and shared datasets may deserve greater recognition.
Training poses a harder problem. Routine calculations, coding, failed approaches, and small discoveries have long helped students acquire intuition and judgment, and delegating this work can remove formative experiences. Students should learn to formulate problems, audit model outputs, reproduce results, and defend their choices.
What universities should build
Rakhlin says universities should be able to pursue questions over long horizons, share results openly, and evaluate claims independently. Partnerships with industry will be essential, but commercial priorities may not cover the breadth of science or remain aligned with it over time, so some technological independence will be necessary.
He points to academia’s accumulated knowledge and experience, including unpublished failed experiments and scientific interpretations, as a strategic asset. He imagines MIT laboratories connected through shared AI research infrastructure, with workflows capturing hypotheses, interventions, outcomes, failures, and interpretations. Making this possible would require investment in compute, secure data systems, and expertise in adapting and post-training AI models.
What could universities gain from recording failed experiments and researchers’ interpretations?
Rakhlin suggests such records capture tacit knowledge that published papers omit, possibly helping models anticipate consequences experts find obvious, and supporting shared infrastructure for scientific exploration.
Sources & further reading
Help keep the record accurate. Send a correction.
