Artificial scientists

The buried tension here isn't just hype versus reality; it's that 'scientific AI' has become a fundraising narrative that lets companies defer accountability indefinitely. A cure that's always five years away is unfalsifiable, which is precisely what makes it useful marketing.

This connects directly to two threads already on the site. The Verge's 'AI is inevitable' trap piece from mid-April documented how vague AI promises drive stock surges regardless of product reality, and this story is essentially the research-sector version of that same dynamic. More structurally, MIT Technology Review's 'agent orchestration' piece from April 21 positioned agents as the bridge between today's LLMs and real-world impact in drug discovery specifically. Read together, the two MIT pieces are in quiet tension: one treats scientific application as a near-term agent use case, while this one questions whether the underlying capability justifies that framing at all. That gap is worth holding onto.

Watch whether any of the major AI-in-drug-discovery players (Isomorphic Labs, Recursion, Insilico) announce a compound reaching Phase II trials within the next 18 months. That would be the first concrete, externally verifiable checkpoint against which the broader scientific AI narrative can actually be measured.