An essay published on writingruxandrabio.com challenges the dominant belief in Silicon Valley that artificial intelligence, or future AGI, will be the main driver of real-world change. The author, who prefers to identify as a ‘naive hamster’, recounts a dinner in San Francisco where an employee of an AI lab questioned him with disdain about his work focused on regulatory bottlenecks in medicine.
According to the essay, the interlocutor stated that AGI will be ‘hyperpersuasive’ soon, citing benchmarks where it already outperforms professional debaters. But the author counters: even with unlimited intelligence, problems like housing and medicine remain stuck due to political will, regulation, and bureaucracy.
The text cites Eroom’s Law: the number of new drugs approved per dollar of R&D has fallen for decades, despite more powerful scientific tools. It also mentions Adaptimmune, which brought two innovative therapies to market for rare cancers but struggles to survive due to development costs.
Another example is the case of ‘baby KJ’, saved by a personalized gene editing therapy: scientists have the knowledge, but cannot easily repeat the feat due to manufacturing costs and regulatory requirements.
The author highlights that clinical trials consume about seven years and over a billion dollars per drug. He argues that these trials generate irreplaceable human data, essential for training better AI models.
There is robust evidence that faster trials boost biomedical innovation, both directly and indirectly. The author cites the case of China: in a decade, the share of Chinese biopharmaceuticals in licensing deals with big Western pharma jumped from zero to more than half, driven by regulatory reforms that allow faster iterative learning with human data.
The essay advocates for the use of surrogate endpoints and biomarkers, where AI could transform discrete readings into continuous ones, optimizing trials. For some indications, improvement could be up to 10 times in speed and cost, if the right surrogates are found.
But the author stresses that the bottleneck is not just intelligence, but governance. Companies trying to build biomarkers face difficulties accessing data: some have waited a year for the NIH to release image datasets. Validation of bone mineral density as a surrogate endpoint in osteoporosis trials took 12 years, even though the data already existed and the analyses were basically regressions.
The author criticizes the superficiality with which the AI sector treats regulation and governance in medicine. Many argue that since most drugs fail due to lack of efficacy, regulation does not explain the slowdown in medical progress. But the author rebuts: the problem is not the final approval decision, but the entire upstream process — how human data is collected and used.
The essay concludes that, although it is difficult to maintain this conviction in the face of smarter people with privileged information, the author remains firm: intelligence is not the main bottleneck for real-world change. He warns that blind faith in the words of AI lab leaders is a mistake, as it is not guaranteed that AI will solve the problems people actually care about, like medicine, unless those problems are thought through clearly.