AI Will Cure Us All!! Or Not!! - EP 84 Abhi Mahajan
Summary
Host Arne Marogi interviews Abhishek Mahajan from the OpenAI Foundation on whether AI can truly accelerate biology. They debate if intelligence or physical experimentation is the primary bottleneck, citing failed TIGIT trials to illustrate the gap between preclinical data and human outcomes. Mahajan explains his work at Noetic using AI to map mouse data to human clinical responses, arguing that while AI aids narrow tasks like protein folding, broader biological acceleration remains limited by slow feedback loops and complex real-world variables.
Topics discussed
Introduction: AI's potential to revolutionize biology and biotech
Intelligence maximalism vs. physical bottlenecks in drug discovery
Limitations of automation and the failure of Tidget trials
Incentives in pharma and the challenge of cross-species data translation
Perturbation models, generative AI, and mapping mouse to human data
LLMs in chemistry, protein folding, and enzyme engineering
Virtual cells, high-throughput screening, and assay innovation
Organoids as predictive models and the search for gold standard data
Biosecurity: Gene drives, bioterrorism risks, and sci-fi bias
Regulating DNA synthesis and the 'dumbest person' problem
Pascal's Wager in biosecurity and the impact of regulation on innovation
Geopolitics of biosecurity, China, and public health responses
Conclusion: Integrating AI expertise with life science experience
Listen ad-free on Castria