arXiv paper: Albert neuro-symbolic AI framework autonomously rediscovers Standard Model and infers top quark properties

An arXiv paper presents Albert, a neuro-symbolic artificial intelligence framework for systematic navigation of particle physics theory space. The framework addresses the combinatorial explosion of possible theories beyond the Standard Model. Albert encodes particle physics as a formal language. It generates tokenized sequences representing symmetries, particles, and interactions under a rule-based grammar, eliminating hallucinations common in large language models. A reinforcement learning environment enforces first-principle theoretical constraints. The system computes observables with radiative corrections and evaluates statistical likelihood via chi-squared analysis against experimental data. Researchers trained a 25-million-parameter transformer model on legacy data from the Large Electron-Positron Collider that contains no direct evidence of the top quark. Albert rediscovered the Standard Model and inferred the necessity and properties of the top quark. It predicted the particle mass at 178.9 plus or minus 5.0 GeV, consistent with measurements from the Large Hadron Collider. The findings illustrate the potential for AI-driven theory exploration. They present it as a rigorous, hallucination-free, and scalable paradigm for autonomous discovery of new physics.