Where Is All the A.I.-Driven Scientific Progress?

In this Hard Fork episode, hosts Kevin Russo and Casey Noon sit down with Sam Rodriguez, CEO of Future House and Edison Scientific, to cut through the hype surrounding AI‑driven scientific discovery. Rodriguez introduces Cosmos, an "AI scientist" that aims to scale research productivity by orchestrating dozens of language‑model agents through a structured world model. The conversation frames the broader debate: tech leaders promise cures and climate solutions, while governments launch initiatives like the Genesis mission to harness AI for breakthrough research.

Cosmos operates as a specialized interface rather than a chatbot, running 12‑hour jobs that ingest up to 1,500 research papers and generate roughly 42,000 lines of code. By leveraging models from OpenAI, Google, Anthropic, and proprietary fine‑tuned systems, it builds a persistent knowledge graph that keeps the agents on task. In internal benchmarks, the platform reproduced findings that human researchers took three to six months to discover, achieving an 80% correctness rate and even uncovering four novel contributions, such as a new mechanistic link between a genetic variant and type‑2 diabetes.

While the tool dramatically compresses data‑analysis cycles, Rodriguez stresses that validation remains a human‑led step. The real bottleneck in drug development lies in costly clinical trials and regulatory approval, not hypothesis generation. Nonetheless, AI systems like Cosmos can prioritize experiments, reduce wasted resources, and accelerate the pipeline toward actionable insights. For business leaders eyeing AI investments, the episode underscores a pragmatic view: AI can augment scientific workflows today, but full‑scale therapeutic breakthroughs still depend on downstream human and regulatory processes.