AI for science needs reasoning, not just data. In 2024, Google Deep Mind scientists shared the Nobel Prize in Chemistry for a neural network, Alpha Fold, that predicts protein structures. It relied on a dataset of roughly 170,000 experimentally validated protein structures that took 53 years and about $21 billion to assemble. Comparable datasets will be difficult or impossible to create in many fields.

Instead, another approach may hold the key: AI agents. Alpha Fold applied a powerful approach to a limited question, but agents are inherently generalists. They model the iterative, highly contingent process of actual research, digitally mirroring the human process of discovery. This is not a new way to do science—it is a new way to *represent* it.

The “censorship‑industrial complex” idea, once a fringe right‑wing theory, has now entered the Trump administration. MIT Technology Review traced its origins and will discuss the findings in a virtual Roundtables session with senior reporter Eileen Guo and executive editor Amy Nordrum.

Meanwhile, a new Amazon data center in Texas could become the U.S.’s biggest polluter, with a permit to release up to 33 million tons of CO₂ and generate up to 7.65 gigawatts of power.