While AI models like AlphaFold 3 and platforms like Noetic AI's show promise in predicting biological interactions and patient responses, the real-world impact is still limited. The discussion highlights that the primary bottleneck in drug development is not a lack of good preclinical assets but the immense challenge and cost of successful clinical trials, an area where AI has yet to prove transformative.
Experts from frontier labs and policy circles struggle to forecast the speed and impact of AI systems that can automate and accelerate AI research itself. This uncertainty, highlighted by a CSET workshop, means that rapid, unexpected capability gains could occur within closed labs, challenging assumptions about manageable progress and creating a high risk of strategic surprise.
The US AI ecosystem is critically dependent on foreign-controlled chokepoints. This includes an overwhelming reliance on Taiwan's TSMC for advanced chips, the presence of Chinese-made components (including potential trojans) in the US power grid, and the fact that a large percentage of top AI talent at US labs are Chinese nationals.
China has evolved from a generics manufacturer to a major hub for CROs and now a source of innovative preclinical drug assets that are being acquired by Western biopharma. In AI, while perhaps scaling existing ideas, Chinese labs are producing powerful models like DeepSeek despite US export controls, indicating a resilient and rapidly advancing ecosystem.
Across domains, the conversation identifies non-obvious bottlenecks. In biology, it's the lack of verifiable ground truth for complex problems, not a lack of data. In US AI development, it's projected to be energy availability by late 2024. For Chinese AI labs, it's a lack of R&D compute due to high inference demand.
Keep pulling the thread on Abhi Mahajan.