Is Biology Necessary to Advance Biology?

John T Van Stan, II , Theodore Bach, Olivier Dangles, Robert Guralnick, Sarah Huebner, Roland Kays, Justin Kitzes, Robert A Krebs, Benjamin Noren, Jarrad H. Van Stan, Michael O Wiitala, Roman Yampolskiy

AI is transforming the workflow of biological science. But what role remains for human cognition when the priority is deep theory? We propose a division of labor based on a “coverage asymmetry” between artificial and biological agents. AI excels at extending the encounterable: scaling known mechanisms and exploring complex parameter regimes. However, AI cannot (yet) perform kind formation – the minting of new variables and constraints – because it is bounded by its training ontology. Current systems cannot register that their own vocabulary is incomplete, because recognizing missing variables requires material participation in the system under study. We leverage phenomenological philosophy to argue that new kinds originate in Wild Being: the embodied friction that occurs when an organism encounters something its current categories cannot name (an Out-of-Ontology encounter). We outline a curriculum to train human scientists as “detectors of the unparameterized,” ensuring that the growth of knowledge remains coupled to the growth of care. The question is not whether AI can do biology, but how we design systems to maximize discovery efficiency, producing both knowledge and knowers whose embodied engagement guides what that knowledge is for.