AI agents for science, rather than AlphaFold-style systems trained on unusually rich datasets, are the focus of a new argument from former Google chief executive Eric Schmidt and Schmidt Sciences AI leader Suhas Mahesh. MIT Technology Review’s August 10 edition of The Download links to their op-ed alongside separate reporting on how the “censorship-industrial complex” theory entered Trump administration policymaking.
The pairing is a newsletter package, not evidence that the two topics are connected. One is a case for a particular direction in research software. The other is reporting on a political framework for interpreting online moderation.
Can AI agents speed up scientific research?
Schmidt and Mahesh argue that agents could be more broadly useful than systems modeled on AlphaFold. AlphaFold predicted protein structures using the Protein Data Bank, a collection of roughly 170,000 experimentally validated structures. The authors say that resource took 53 years of international work to assemble and represented an estimated $21 billion in experimental work.
Those conditions are difficult to recreate, they argue. Many experimental fields lack datasets that are consistent and scalable enough for comparable neural-network training. Their proposed alternative is an agent: an AI reasoning engine given access to digital or physical tools. The authors contend that such systems can model the iterative, contingent nature of research rather than apply one specialized method to one bounded problem.
The distinction remains an argument, not a settled forecast. The op-ed says agents powered by large language models, or LLMs, still hallucinate, make inconsistent judgments, and face memory and input limits that constrain how long they can operate autonomously.
As an illustrative result, the authors cite Google’s AI Co-Scientist. According to the op-ed, the system generated a hypothesis about antibiotic-resistance genes moving between bacterial species that matched a conclusion reached by Imperial College London researchers after a decade of wet-lab work. The authors say the Imperial paper had not been seen by the system and remained under peer review. That is a notable comparison, but it does not establish broad scientific reliability.
What does “censorship-industrial complex” mean?
MIT Technology Review and Type Investigations describe the term as a theory alleging that government agencies, academics, civil-society groups, and large technology platforms worked to suppress conservative and populist speech online while presenting their work as anti-disinformation efforts.
The reporting characterizes that claim as a once-niche far-right conspiracy theory, and says it has influenced a range of actions by the second Trump administration. The outlets say their nine-month tracing found that the idea gained momentum in 2023 through a small group of people backed by closely connected right-wing organizations and media outlets, with internet activist Mike Benz at its center.
The newsletter also promoted an August 13 discussion with senior reporter Eileen Guo and executive editor Amy Nordrum about the investigation and its implications for democracy and the internet.
The theory’s political influence and its underlying allegations require separate treatment. A March 2025 Senate Judiciary Committee submission by journalist Benjamin Weingarten documents how a proponent defines the term, including account suspensions, demonetization, labels, and reduced distribution. The submission is the witness’s argument, not independent proof of a coordinated censorship regime.
This story draws on original reporting from MIT Technology Review.
