AI Agents Could Accelerate Scientific Discovery

AI agents offer a promising new paradigm for accelerating scientific discovery, moving beyond the limitations of purely data-driven models. Eric Schmidt, former CEO of Google and cofounder of Schmidt Sciences, and Suhas Mahesh, who leads AI for science at Schmidt Sciences, argue that while AI has demonstrated its potential for groundbreaking discoveries, such as Google DeepMind's AlphaFold which predicted protein structures, this approach may not be the most effective template for broad scientific acceleration. AlphaFold, for instance, relied on a substantial dataset of approximately 170,000 experimentally validated protein structures, the assembly of which took 53 years and an estimated $21 billion in experimental work. The authors contend that creating comparable datasets will be exceedingly difficult, if not impossible, across many scientific fields.

Instead, AI agents are presented as a more adaptable and powerful solution. These agents are designed to model the iterative and highly contingent nature of actual scientific research, mirroring the human process of discovery. Unlike tools like AlphaFold, which apply a specific, powerful approach to a narrowly defined question, AI agents are characterized as generalists. They do not represent a fundamentally new method of conducting science but rather a digital simulation of the human-driven process of inquiry and innovation. This approach holds the potential to unlock advancements in areas where large, curated datasets are not readily available or feasible to create.

Separately, the concept of a "censorship-industrial complex" has emerged as a significant influence on US policy discussions. This theory, which posits the suppression of conservative and populist speech, has gained traction in right-wing online communities and has now entered the political discourse of the Trump administration. MIT Technology Review has conducted an extensive nine-month investigation into the origins and proliferation of this theory. Senior reporter Eileen Guo and executive editor Amy Nordrum are scheduled to discuss their findings, the theory's trajectory, and its potential implications for democracy and the internet in a virtual roundtable session on Thursday, August 13, at 19:00 GMT. Registration for this session is available, indicating a growing interest in understanding the societal impact of such narratives on policy and public discourse.