Information, Dynamics and
Emergent Intelligence with Physics

Intelligence is physical: it is realised in noisy substrates, driven out of equilibrium, and constrained by energy, time and memory. IDEIP asks whether thermodynamics and statistical physics can yield testable limits, practical design rules and universal laws for learning systems.

Its two organising ideas are dissipation, connecting generative dynamics, physical computing and the cost of computation; and universality, connecting glassy learning landscapes, renormalisation-group methods, scaling laws and the emergence debate. Topics span physical neural networks and their fundamental limits; thermodynamic, optical and gain-dissipative hardware; the statistical mechanics of learning; diffusion and score-based generation as non-equilibrium dynamics; the stochastic thermodynamics of computation and the energy cost of AI; associative memory and the equivalence between modern Hopfield networks and attention; the free energy principle and active inference; and phase-transition tests for emergence and scaling in large models.

Sessions will be organised around shared observables and trade-offs, bringing theorists, algorithm designers and experimentalists to address the same questions across different models and physical substrates.