Biology needs to become prospective

Biology’s first big data era — defined by the opportunistic, random accumulation of sequences and structures — is due for an overhaul. While biological foundation models (BFMs) represent the zenith of this era, their utility will always be hamstrung by the inherent non-independence of biological data. Massive increases in data volume don’t yield proportional increases in unique information. We argue that the next era of biology must be prospective rather than retrospective. We show that even a simple Bayesian framework can move us beyond the "bitter lesson" of brute-force scaling and instead treat biological measurement as a strategic act of inference. There’s no excuse to gather data blindly; biology must become prospective. This pub is intended for computational biologists, machine learning researchers, and anyone modelling the complexities of life. It’s especially relevant for funders, decision makers, and researchers invested in large-scale biological data collection efforts and the development of biological foundation models.