PSI Leveraging AI to Advance Scientific Discovery with Quantum Computing Breakthrough

Image: TechCrunch · Source

Startup PSI used an unreleased Anthropic AI model to resolve a long-standing quantum computing problem, illustrating AI's growing role in scientific research and potential to transform theoretical physics into practical technologies.

Physical Superintelligence (PSI), a startup lab focused on AI-driven scientific research, recently made significant strides by solving an 18-year-old problem in quantum computing using an unreleased version of Anthropic’s Fable 5.2 AI model. David Turturean, an MIT alumnus and PSI technical staff member, utilized early access to this advanced model to address the Aaronson-Ambainis conjecture—a fundamental question regarding quantum computers’ capabilities to outpace classical ones in problem-solving speed.

This problem, posed by researchers including Scott Aaronson, had remained unresolved for nearly two decades. Although Aaronson himself admitted uncertainty about the details of the AI-generated proof, he expressed confidence in its correctness because it was formalized in Lean, a rigorous mathematical coding language, and scrutinized by a team of top researchers collaborating with PSI. The AI’s solution delivered a negative result eliminating one potential pathway quantum computers might have taken to achieve high-speed problem-solving, thereby narrowing the focus of ongoing quantum computing research.

PSI’s approach exemplifies a new paradigm in scientific investigation: integrating specialized AI models with comprehensive research frameworks and tools to tackle complex theoretical physics problems. CEO Matt Pines emphasized the necessity of such systems to provide broad and deep explorations of scientific questions through robust infrastructures including memory systems and simulation capabilities.

Currently, PSI generates revenue by employing its AI tools to optimize physical data center operations, simulating environmental and physical variables that impact both performance and community effects. However, the company’s long-term vision aims at translating theoretical physics advances—facilitated by AI assistance—into tangible technologies.

The startup’s breakthrough reflects AI’s disruptive potential in domains traditionally challenging for machine reasoning, particularly in experimental sciences where natural phenomena must be validated through physical experiments. Pines noted that unlike mathematics, physics involves real-world variables that can confirm or contradict computational models, making experimental validation a crucial component. PSI is thus developing iterative loops between AI-generated hypotheses and empirical testing to harness these discoveries effectively.

With AI accelerating the pace of scientific discovery, PSI and its academic advisors underscore the importance of transparency and collaboration. The company delayed publicizing its results until the underlying AI model approached readiness for release, allowing time to consult experts and refine the presentation for human understanding. This cautious approach highlights emerging best practices in AI-driven research, balancing rapid innovation with methodological rigor.

The PSI case underscores a broader shift as "neolabs for science" emerge, leveraging AI to push frontiers in physics and related disciplines. While challenges remain in standardizing workflows and validation in this new research landscape, the demonstrated results affirm AI’s transformative role in expanding scientific knowledge and ultimately enabling new technologies.