The Download: AI agents for science, and the “censorship-industrial complex”
| Source: MIT Technology Review AI
Tags: AI agents, Eric Schmidt, Schmidt Sciences, AlphaFold, OpenAI, Astra, cyberattacks, data centers
MIT Technology Review's digest covers two major stories: Eric Schmidt arguing AI agents outperform AlphaFold-style tools in data-scarce scientific domains, and OpenAI pausing its Astra model after tests showed it could autonomously launch cyberattacks — with Amazon's Texas data center also drawing scrutiny for a 33M-ton CO2 permit.
Details
Eric Schmidt, former Google CEO and Schmidt Sciences co-founder, argues in an op-ed that AI agents — not tools like AlphaFold — are the better template for scientific discovery. AlphaFold required 53 years of experimental work and roughly $21 billion in validated protein structures. Comparable datasets do not exist in most fields. Agents, as generalists that model the iterative, contingent process of research, can substitute for data-rich approaches in data-sparse domains. The digest also covers OpenAI pausing development on its Astra AI model after internal testing found it could launch cyberattacks autonomously. The FT report surfaces the tension between capability disclosures (transparency) and critics who argue such announcements are designed to generate hype. Additional items: Amazon's new Texas data center received a permit to emit up to 33 million tons of CO2 annually, which would make it the most polluting power plant in the US. North Korean state-linked group Kimsuky has built AI tools for spear-phishing campaigns that can automate attacks and analyze stolen data. For practitioners, the Schmidt piece articulates a model-level case for why agentic science pipelines warrant investment even when training data is scarce.