Reporting on a new Google, Google DeepMind and MIT FutureTech study of 637 scientists, the piece finds that AI is accelerating theoretical scientific work such as mathematics but has not proportionally sped up experimental research in biology and chemistry. Of those surveyed, 44% reported research bottlenecks shifting toward physical experimentation and 41% reported growing backlogs of untested hypotheses, attributed to the difficulty and expense of automating lab work. Economist Daron Acemoglu is quoted noting that, unlike mathematical proofs, experimental biology has ‘no ground truth’ against which AI predictions can be readily verified.
