A mathematical puzzle that had resisted researchers for more than a decade has finally been solved through an unusual partnership between Nobel Prize-winning physicist Giorgio Parisi, theoretical physicist Francesco Zamponi, and Anthropic’s Claude AI. The breakthrough centered on the physics of jamming, a phenomenon in which densely packed particles become locked in place, preventing movement despite the absence of any orderly structure. The result demonstrates that generative AI can contribute meaningfully to advanced scientific research when paired with human expertise and rigorous verification, tells Live Science.
The problem originated in 2014, when Parisi and Zamponi identified an intriguing mathematical relationship, expressed as a + b = 1, between two parameters describing the forces and microscopic gaps that emerge as particle systems reach the jamming transition. Although numerical evidence consistently supported the relationship, no analytical proof could be found despite years of effort by multiple researchers using different theoretical approaches.
In 2023, Parisi turned to Claude, asking the AI to reformulate the problem and search for possible proofs. After roughly 40 prompts, Claude proposed a novel line of reasoning. While the response contained mathematical errors, its central idea proved valuable. Guided by the researchers, the flawed derivation evolved into a complete and rigorous proof. The experience showed that AI did not independently solve the problem but instead generated an unconventional perspective that researchers had not previously considered.
The researchers emphasized that human judgment remained indispensable throughout the process. Every step required careful evaluation, correction, and mathematical validation before it could be accepted. Zamponi noted that whether Claude reached its insight through sophisticated pattern recognition or a form of reasoning is less important than the fact that it revealed a productive direction that experts had overlooked.
Beyond resolving the jamming conjecture, the work highlights a broader shift in scientific research. Rather than replacing scientists, generative AI may become a powerful collaborator capable of suggesting fresh mathematical ideas, accelerating exploration, and uncovering unexpected connections. For theoretical physics and mathematics, the study offers an early example of AI augmenting human creativity while reinforcing that rigorous proof, interpretation, and scientific responsibility remain firmly in human hands.
