digest
Short Summary
Main recommendation: Drug development should be made faster and cheaper by relying on reliable biomarkers, such as bone density for osteoporosis, rather than requiring every new drug to prove its effect through slow, expensive outcome trials, so that AI-discovered drugs can actually reach patients. Epstein argues that "Eroom's law" (Moore's law in reverse) explains why science has become faster and cheaper while medicine has become slower and more expensive. He presents this as a fixable policy problem, with the 2025 FDA decision to again accept bone density for osteoporosis drugs as an early sign of progress.
Key Points
- Eroom's law is the reverse of Moore's law: with $2 billion (inflation-adjusted) in 1960, one could expect about 40 FDA-approved drugs, while today the same spending yields about 0.8.
- AI can now design a potential drug to bind a discovered protein in under an hour, but getting one drug approved takes on average 13 years and more than $100 million.
- The speaker describes the FDA's 1990s shift from accepting biomarkers to requiring direct fracture outcomes as making trials far larger and slower, which he presents as a well-intentioned policy choice that added delay.
- The Prolia case shows the cost: its fracture trial needed 7,393 patients followed for three years, versus the roughly 100 patients a bone density trial would need to show an effect.
- Sixty years ago the average drug took about five years from patent to patient, and Operation Warp Speed delivered COVID vaccines in under a year, showing faster development is possible.
Details Worth Noting
- Concrete instructions: The speaker gives no step-by-step instructions for individuals or companies. The actionable recommendations are directed at drug companies, AI companies, regulators, and investors: work together to make clinical development faster and cheaper, collect the right data so AI can help develop reliable biomarkers for each form of aging damage, and rely on biomarkers where they are trustworthy enough to replace long outcome trials.
- The speaker names Dario Amodei (Anthropic CEO), who said AI could cure most major diseases in 5 to 10 years, and the speaker calls that unlikely under current drug development practices.
- Prolia (developed by Amgen, with development beginning around 2000) reduced spinal fractures by 68% in its trial and was approved in 2010, about nine years after patenting; the speaker cites 13 years as the average development time.
- The speaker frames the need for biomarkers as a way to test drugs for aging damage, such as a measurement that could show within weeks whether a drug would prevent dementia years later; this is presented as a goal, not an existing tool.
- The speaker's claims about costs and timelines come from his own presentation and cited figures, and are not independently verified in the transcript.
Bottom Line
The speaker's core argument is that drug approval rules, not scientific capability, are the main bottleneck, and reforming how drugs are proven safe and effective is the step that would let AI-driven discoveries help aging patients in time.
