The Computational Assembly Line
AI laboratories are increasingly packaging routine automated workflows as autonomous scientific breakthroughs, confusing computational data processing with actual discovery. Earlier this year, Anthropic launched a molecular biology lab where Claude agents generated hypotheses about complex biological problems, leaving human scientists to run the physical experiments. As Anthropic reported, a system of 950 agents flagged a repeating pattern surrounding a known enzyme—a cluster Anthropic claimed had never been catalogued before.
Anthropic leaned heavily into the marketing, describing the pattern as "reminiscent" of the early days of CRISPR. Working researchers immediately pushed back. As pointed out in a viral critique by a biologist—subsequently endorsed by Eli Lilly CEO David Ricks—finding an unusual cluster of genes and repeats is the easy part. The actual science happens when you figure out what the biological system does in practice.
"Finding a weird cluster of genes and repeats is often the easy part. The hard part, and where the real discoveries come from, is figuring out what the system actually does."
This wide gulf between automated data processing and functional biological discovery illustrates a broader pattern: tech providers are selling computational labor as self-directed scientific milestones to inflate R&D expectations.
Verification and Attribution Conflicts
Methodological confusion only intensifies when AI systems surface findings that conveniently overlap with ongoing academic research. As reported by The New York Times, Mario Rodríguez Mestre, a biologist at the University of Copenhagen, noted that his team had already discovered this exact pattern. Because Rodríguez Mestre regularly chatted with Claude during his research, he questioned whether Anthropic's team had scraped or learned from his prior conversations. While Anthropic denies this, Rodríguez Mestre stated he is halting all use of Claude anyway.
This friction exposes the core economic incentive at play: tech giants profit immensely from blurring the line between algorithmic pattern recognition and fundamental breakthroughs. Without clear methodological criteria to evaluate an LLM's actual contribution to the scientific method, corporate R&D budgets remain vulnerable to an expensive bubble of overhyped capabilities.
Labeling hypothesis-filtering algorithms as autonomous discoverers does nothing to advance science. It merely demonstrates that marketing departments have gotten exceptionally good at dressing up basic data sorting as a revolution.