AI and the FDA (marginalrevolution.com)

Dean Ball’s survey—and the analysis it sparked—argues that AI will dramatically accelerate drug discovery within a few years, producing many more candidate therapies far faster and cheaper than today. That surge will collide with an FDA approval process designed for a pre-AI era, creating backlogs that delay patient access and could push startups to relocate to friendlier jurisdictions. The nearer-term, real-world benefit is drug repurposing: because physicians can prescribe approved drugs off‑label, AI-driven identification of new uses (as with dexamethasone in COVID) can rapidly save lives without new trials. Technically, AI shifts drug development toward rational, molecularly targeted design and personalized medicine, exposing limits of large randomized controlled trials (RCTs), which report average effects across heterogeneous patient subgroups. As patient-level molecular data proliferates, treatments will need to be tailored rather than validated only by population averages, and regulatory systems that rely on RCTs may discard useful therapies. The rise in low-cost candidates also invites regulatory competition (e.g., China’s faster approvals). The bottom line: AI can unlock safer, more precise drugs, but realizing that promise requires regulators—especially the FDA—to adapt trial frameworks, approval pathways, and risk assessments to personalized, AI-enabled medicine.