我们何时才能说 AI 做出了科学发现？

This story originally appeared in The Algorithm, our weekly newsletter on AI. To get stories like this in your inbox first, sign up here. Last Wednesday, Anthropic announced that earlier this year it had launched a molecular biology lab, where Claude agents read and conjecture about hard biology problems and human scientists run experiments on what they report. And this AI-powered lab, the company said, had made its first discovery. To understand what Anthropic says its system did, imagine you’re flipping through a library of millions of DNA sequences, amassed as scientists sequence more and more of the living world. One step toward a breakthrough might be finding a peculiar sequence that encodes an interesting enzyme, perhaps. Then you’d need to figure out what that enzyme does and, eventually, how to manipulate it to do something useful. What Anthropic says its system of 950 agents found after 21 hours was not a brand-new sequence. The agents instead flagged a repeating pattern surrounding a known enzyme, a particular pattern Anthropic said hadn’t been catalogued before. But if you read through Anthropic’s announcement, which calls this pattern “reminiscent” of what led to the gene-editing technology CRISPR that “has already transformed science and medicine,” it sounds as if this army of agents really found something of note. These claims have angered some biologists. A viral post from one, subsequently endorsed by the chair and CEO of the drugmaker Eli Lilly, said that “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.” The agents helped with some laboratory grunt work, in other words. But a discovery it is not. It’s a reminder that even if AI does something impressive—like finding a pattern in a mass of biological data that would be difficult to perceive with human eyes alone—the result itself may not constitute a breakthrough for science. What is novel for AI may be routine, unsurprising, or simply not that consequential to a biologist. Muddying the issue further, Mario Rodríguez Mestre, a biologist at the University of Copenhagen, said over the weekend that his team had already discovered this particular pattern, the New York Times reported. Mestre, who regularly chatted with Claude in his work, wondered whether Anthropic’s team had learned from his conversations. Anthropic denies this, but Mestre says he’s stopping all use of Claude anyway. Part of the problem here is that AI companies aren’t presenting their systems simply as tools scientists can use, like microscopes or supercomputers. They’re insisting that the AI systems are making discoveries themselves. To some, that approach is incompatible with how science actually works, with new knowledge more typically emerging from collaboration and an ever-growing arsenal of tools. It’s also making people more skeptical of genuine progress when it happens. Whittling 200,000 candidates down to a few worth exploring is no small feat; it is legitimate scientific work. The fact that a general-purpose chatbot could do that work is notable, even if humans helped steer it and ultimately ran the experiments. But once the standard is whether Claude itself made a discovery, all that becomes evidence for one side or the other in a debate that has only two answers: breakthrough or bust. Once we’re judging AI by whether it has made a discovery, it’s also tempting to shift the goalposts even after it really does seem to notch a win. Earlier this month, OpenAI said its own team agents had cracked a million-dollar problem in mathematics. But a couple of weeks later, nearly every AI skeptic in my feed was sharing an article asking whether it was the math problem that really mattered. To be clear, the piece did not argue that OpenAI’s solution was wrong. Instead, it argued that the particular result may not be the one mathematicians care most about. Throw in the accusation by a mathematician that the models may have used some of his work without credit, and people are left thinking either OpenAI cheated or the solution wasn’t important anyway. Or both. That’s part of what concerns Lucas Harrington, the biologist who wrote the post critiquing Anthropic’s announcement. He closed with a suggestion: AI companies, he said, should “set the bar high now, so that when an AI actually discovers a fundamentally new biological mechanism, everyone appreciates how big a deal it is.” But as OpenAI’s Sam Altman and Anthropic’s Dario Amodei race to one-up each other, raising the bar for scientific breakthroughs by AI might be the last thing on their minds.