Superintelligence won't remove the main brake on longevity: biology needs data, experiments, and clinical feedback, writes Bo Wang.
On March 19, Bo Wang, a professor at the University of Toronto and chief AI scientist at University Health Network, supported Jeffrey Miller's thesis that the success of language models doesn't automatically translate to aging. For text and code, models had internet-scale corpora and fast response verification. For the human body, data is sparse, scattered, and returns results too slowly.
ASI, or hypothetical superintelligence, is often described as a universal accelerator for biomedicine. The logic is clear: if a model becomes smarter than the best researchers, it will find targets, molecules, and treatment regimens faster. Bo Wang points to the next link in this chain. For longevity, it's not enough to come up with a hypothesis; it also needs to be tested on the human body.
Language models have grown on billions of existing examples. Biomedicine works with closed medical records, small cohorts, and various protocols and outcomes that cannot be summarized into a single, clean dataset. In longevity research, the challenge is even greater: the central question sounds simpleâwho lived longer and whyâbut the answer unfolds over years. Therefore, the field requires surrogate metrics, that is, indicators that allow one to assess the risk of aging before the patient's death. Even such metrics are slowly validated: in osteoporosis, the SABRE program sought recognition of bone mineral density as a surrogate indicator over 12 years, even though the data itself already existed.
Feedback is a similar story. The chatbot receives an assessment of the response almost immediately. A drug candidate first undergoes testing in cells and animals, then the first stage of human trials, where safety is assessed, followed by patient recruitment and months of observation. Biology, logistics, and regulatory processes dictate the calendar time here. Asimov Press recently described this clinical ceiling for AI: even a very good drug still faces limitations in terms of patient recruitment, side effects, logistics, and regulatory requirements.
"To cure cancer, we don't need a magic oracle. We need more experiments, more data, more clinical trials. These are bought with money, not conjured," wrote Bo Wang.
Bo Wang added a financial argument: he estimates that the current hundreds of billions going to ASI would increase funding for longevity research by approximately 100 times if it were directed directly to this field. This is a rough estimate. The field needs a machine for discovering the truth about humans: long-term biobanks, uniform measurement protocols, surrogate aging indicators, and clinical pathways where a signal is visible before death or severe complications occur. Without this infrastructure, AI will become increasingly adept at proposing hypotheses and will continue to wait for slow biology to respond.
