Greg Brockman on Move 37: when will AI achieve a truly novel scientific breakthrough?
Oct 9, 2025
Key Points
- Greg Brockman frames AI's next frontier around achieving genuinely novel breakthroughs—discoveries humans would never have conceived—but the segment's hosts argue current evidence shows acceleration of known approaches rather than true innovation.
- The distinction matters: a researcher solving a three-month problem in twenty minutes represents speed amplification, not the kind of conceptual leap that would constitute Move 37.
- Even without breakthroughs, a 2% speedup in scientific discovery would be enormous, and justifying AI investment does not require true novelty—consistent amplification of human capability suffices.
Summary
Greg Brockman frames the next frontier for AI breakthroughs around "Move 37"—the notion that AI will discover truly novel strategies or insights that humans would not have conceived of, analogous to DeepMind's AlphaGo move that shocked Lee Sedol and shifted how players understood the game. Brockman expects models in the next year to solve hard problems in coding, material science, and medicine, either independently or in collaboration with top researchers.
The distinction between genuine novelty and acceleration is critical. Scott Aronson's experience with GPT-5 clearing a research roadblock that would have taken him another hour or two is framed as helpful assistance, not innovation. A physics professor completing a three-month problem in twenty minutes is similar—speed amplification, not Move 37. Genuine novelty means something a leading expert would never have thought of. If they could have thought of it but just not quickly, that is not Move 37.
The attribution problem cuts deeper. Nobel Prize-caliber breakthroughs already involve computers throughout the workflow. Lab notebooks are digital, pipetting is automated. Adding an LLM that accelerates a known approach is continuous with that trend, not a discontinuous leap.
The practical impact of acceleration is nonetheless massive. A 2% speedup in the rate of scientific discovery would be enormous, and justifying AI investment does not require Move 37. Consistent amplification of human capability is sufficient. Brockman's framing conflates the two, but the distinction matters for assessing what has actually been achieved versus what remains ahead.