Science Friday AI Was Supposed To Discover New Drugs. Where Are They?
4 snips
Oct 17, 2025 Dr. Peter Coveney, a computational chemistry expert from University College London, discusses the challenges AI faces in drug discovery. Despite its promised speed and cost-efficiency, most AI-designed drugs have yet to reach approval. He reflects on the high-profile failures like Benevolent AI, and emphasizes that while generative AI may create novel molecules, many crucial development stages remain. Advocating for in silico trials and digital twins, he calls for realistic expectations about AI’s role in medicine and the need for addressing data biases.
AI Snips
Chapters
Books
Transcript
Episode notes
Why Drug Discovery Takes Decades
- Drug discovery is slow because we lack full knowledge of human biology and rely on a one-size-fits-all "magic bullet" approach.
- Late-stage failures are common and costly, with far less than 10% of candidates reaching market.
AI Excels At Early Screening
- AI can rapidly search huge molecular libraries and find geometric "hits" that fit protein active sites.
- But binding geometry is only an early step; mechanisms and later-stage properties still require deep understanding and testing.
Benevolent AI's Rise And Fall
- Benevolent AI rose to a multi-billion valuation during the first AI wave but produced no market drugs.
- Its value collapsed and it was taken over around 2023, illustrating overblown expectations.
