Opinion: MIT professor outlines path for AI in academia

Rakhlin contends that a polished paper is becoming a weaker signal of individual expertise. Departments should shift rewards toward asking good questions, replication, and synthesis, while explicitly defining what researchers take intellectual responsibility for, even when AI performs substantial parts of the work. He warns that delegating routine calculations and coding to AI may remove formative experiences necessary for developing intuition and judgment. Students must still learn to formulate problems, audit outputs, and defend their choices.

To address these challenges, Rakhlin proposes that universities build shared AI research infrastructure to connect laboratories. This system would capture hypotheses, outcomes, and failures, allowing AI agents to surface relevant advances across departments. He emphasizes that academia’s tacit knowledge, including unpublished negative results, is a strategic asset that could help improve AI models in empirical sciences. Achieving this requires substantial investment in compute, secure data systems, and post-training expertise to ensure universities maintain technological independence and long-horizon research capabilities.