AI Solves 18-Year-Old Quantum Computing Puzzle

Artificial intelligence is quickly pushing past simple text generation and entering the realm of complex mathematical and physical problem-solving. David Turturean, a recent MIT graduate and technical staff member at the startup lab Physical Superintelligence (PSI), recently used early access to Anthropic’s Fable 5.2 model to successfully crack the Aaronson-Ambainis conjecture. This critical open problem in quantum query complexity had stumped researchers for nearly two decades.

This breakthrough highlights a major shift in how academic research happens. Specialized machine learning tools are beginning to outpace traditional human paper publication cycles. By autoformalizing the proof in Lean—a specialized programming language for mathematics—and collaborating with leading human experts, PSI has demonstrated a practical blueprint for commercializing automated scientific discovery.

Understanding the Aaronson-Ambainis Conjecture

The solved problem directly addresses whether quantum computers can consistently outperform classical architectures under specific conditions. Scott Aaronson, one of the original researchers who formulated the problem, admitted mixed emotions upon reviewing the solution, noting feelings of satisfaction, awe, disappointment, and fatigue. Because the proof was autoformalized in Lean and rigorously vetted by external researchers, experts verified its correctness despite its machine-generated origins.

Rather than opening up a new pathway for quantum speedups, the proof serves as a definitive negative result. It effectively closes off a specific avenue that scientists were exploring, allowing researchers to redirect their efforts toward more viable domains. Turturean noted that publishing traditional papers may soon become obsolete as machines generate rigorous mathematical proofs instantaneously.

The Technical Architecture Behind AI Proof Generation

The success of PSI relies on a carefully engineered technical stack that extends far beyond the raw weights of a foundational large language model. At its core, the system utilizes advanced model harnesses, structured agentic scaffolding, and specialized tool integrations designed to maintain logical consistency across thousands of reasoning steps. According to PSI CEO Matt Pines, an effective AI research infrastructure requires comprehensive simulation environments, persistent memory systems, and dedicated tooling that can handle both the breadth of exploratory mathematical search and the depth of formal verification.

In this workflow, the foundational model acts as a hypothesis generator, proposing mathematical lemmas and structural steps. These proposals are instantly translated into the Lean theorem prover. Lean acts as an uncompromising programmatic gatekeeper: if a machine-generated proof contains a logical fallacy or an unproven assertion, the compiler rejects it. This closed-loop feedback mechanism allows the AI system to iteratively correct its own errors without human intervention during the initial derivation phase.

Commercializing Theoretical Science

While the long-term vision of PSI involves translating theoretical physics into cutting-edge technology, the company currently sustains its operations through practical commercial applications. PSI leverages its proprietary physics and simulation tools to optimize data center performance, balancing complex variables such as heat, power, water consumption, and acoustic impact. This pragmatic revenue model bridges the gap between theoretical exploration and immediate enterprise utility.

When evaluated against traditional research laboratories and enterprise AI startups, PSI occupies a distinct operational niche. Traditional tech giants rely on brute-force computational scaling, while boutique engineering firms focus strictly on applied software. PSI operates at the intersection, monetizing auxiliary simulation capabilities while funding fundamental theoretical exploration. As the tech industry continues to expand massive computational infrastructures, specialized physics optimization is becoming critical for sustainable resource management.

The Timeline of Automated Mathematical Discovery

The trajectory of AI-driven scientific discovery has accelerated dramatically over the past several years, shifting from simple pattern recognition to active hypothesis generation and formal verification. The timeline of the Aaronson-Ambainis breakthrough illustrates this rapid maturation:

• 2008: Scott Aaronson and Andris Ambainis formulate the quantum query complexity conjecture, highlighting fundamental limits on quantum speedups.
• 2023–2024: Early automated theorem provers emerge, successfully handling undergraduate-level mathematics but struggling with open, frontier research problems.
• Late 2025: PSI begins utilizing advanced agentic frameworks and early access models, pairing generative text capabilities with formal verification languages like Lean.
• November 2026: David Turturean successfully uses Anthropic’s Fable 5.2 model harness to crack the 18-year-old conjecture, coordinating validation with academic advisors like Jim Halverson at Northeastern University.
• Projected 2027–2030: Industry watchers anticipate the widespread adoption of AI-generated pre-prints, forcing academic institutions to completely overhaul peer-review frameworks.

Frequently Asked Questions

What is the Aaronson-Ambainis conjecture?

It is a fundamental problem in quantum query complexity formulated nearly two decades ago regarding the specific conditions under which quantum computers can outperform classical computers.

How did PSI solve this quantum problem?

PSI staff member David Turturean used early access to an unreleased Anthropic AI model (Fable 5.2) alongside specialized software harnesses and the Lean math coding language to generate and verify the proof.

How does PSI make money right now?

While the long-term goal is commercializing theoretical physics, PSI currently generates revenue by applying its simulation tools to optimize data centers for heat, water, power, and acoustic performance.

Why is formal verification in Lean necessary?

Large language models are prone to hallucinations and subtle logical errors. Lean is a rigorous programming language for mathematics that acts as a strict compiler, ensuring that every step of a machine-generated proof is logically sound before humans spend time reading it.

Sources and further reading

Read the original reporting on TechCrunch about PSI and its approach to AI-driven scientific discovery.