Nature reported this week that Anthropic's AI biolab has identified what appears to be CRISPR-like DNA editing machinery in viruses, a finding that, if confirmed, would represent a genuinely new class of biological tool. The discovery was made by an AI system scanning biological sequence data. It was not designed to find this. It found it anyway. This is the part that matters culturally, not just scientifically.
Alchemy, Curiosity, and the Authorship Problem
Nature ran a second piece this week, a books essay on why science would not exist without alchemy, curiosity, and great writing. The argument is that scientific discovery has always been entangled with the personality and obsession of the discoverer. The alchemist who burned papyrus scrolls to understand combustion (the subject of another Nature briefing this week) was doing something that looked like destruction and was actually investigation. The AI system scanning viral genomes for functional patterns is doing something that looks like indexing and may be investigation. The question the Anthropic result forces is whether curiosity, as a generative condition for discovery, is actually necessary, or whether it is merely sufficient. We assumed the former. The virus finding suggests we should revisit that assumption.
Green Chemistry's Lesson for AI Biotech
The other academic story that brackets this one is chemists still struggling to eliminate hazardous solvents after decades of green chemistry effort. The lesson from that field: discovering a better method and actually replacing incumbent practice are separated by enormous institutional inertia. Anthropic's AI found something potentially significant in biology. Whether that translates into a deployed tool, a widely adopted technique, or a footnote in a future review paper depends almost entirely on factors outside the AI's knowledge or control, including funding, regulatory pathway, and whether the humans in the loop are curious enough to follow the lead. Brewster Kahle's argument about the global brain applies here: the discovery infrastructure is accelerating, but the integration infrastructure, the part where findings become knowledge become practice, is still running on the old timetable.