The researcher dismantles the promise of the head of Anthropic by relying on a series of specialists: neither the body of knowledge nor the regulatory deadlines allow for this timetable.
The formula regularly returns to the mouths of the leaders of the sector: artificial intelligence would cure "most human diseases" in five to ten years. Gary Marcus devotes a text to methodically dismantling this deadline, recalling that after more than a decade of efforts, no drugs designed by AI have yet passed into routine clinical use.
Its central argument is less about the power of models than about the nature of the problem. There are three gaps: the lack of causal understanding in current systems, a very incomplete body of medical knowledge, and an underestimation of the incompressible time taken by clinical validation and regulatory authorization. A clinical trial does not shorten because the molecule was found faster.
The text is based on several specialists. cardiologist and researcher Eric Topol judges the projection out of reach; Catharine Young recalls that the word "cancer" covers more than two hundred distinct diseases with different causes; Keith Robison points out that the safety demonstration alone consumes the bulk of the announced schedule; Lior Pachter notes that there is not even a complete inventory of human diseases. Marcus concludes by inviting his contradictor to spend a few months in hospital.
Read the full article on the original source
Go to the source →