String Cosmology in the AI Era

Objective

String theory offers a unified framework for quantum gravity, yet its predictive power is obscured by the vast “landscape” of four-dimensional effective field theories it produces. StringCosmoAI sets out a principled way to recover those theories that are both theoretically controlled and compatible with observations. I will combine sophisticated geometric tools -drawing in particular on tame geometry - with disciplined, auditable machine-learning methods to move beyond ad hoc searches. The programme advances along three complementary strands. First, I will construct and openly release large, validated databases of string-derived effective field theories, complete with provenance, machine-readable schema, and example notebooks for reuse. Second, I will establish when key questions in cosmology are statistically learnable or provably hard within string theory effective theories by drawing on statistical learning theory and computational complexity theory. Third, building on these theoretical foundations, I will deploy machine-learning techniques to chart theories compatible with cosmological observations - particularly slow-roll inflation and small vacuum energy - with rigorous uncertainty quantification and interpretability. The project is anchored in a purposeful collaboration between the University of Bologna and Northeastern University. Northeastern University offers a world-class environment for machine learning in fundamental physics, enabling rapid prototyping, interpretability, and robust evaluation of the AI tools and techniques employed; the University of Bologna brings leading expertise in string phenomenology and cosmology, ensuring theoretical control and meaningful physical interpretation of the results. This partnership will accelerate progress and, in turn, deliver a fresh, groundbreaking perspective on string theory phenomenology.