Sophisticated Learning (SL) is a plan-and-learn algorithm that integrates active parameter learning within the Sophisticated Inference (SI) tree search framework of Active Inference. Unlike SI, which optimizes beliefs about hidden states, SL also updates beliefs about model parameters within each simulated branch, enabling counterfactual inference about how future observations might improve subsequent plans. We designed experiments using a biologically inspired seasonal foraging task, forcing agents to balance probabilistic reward harvesting with information gathering. SL survived an average of 8.2% longer than SI and 35% longer than Bayes-adaptive Reinforcement Learning (BARL) in early trials, when rapid learning is crucial. While SL and SI showed identical convergence performance, SL converged 40% faster than SI. Furthermore, SL showed robust performance improvements over other algorithms under changing environmental configurations. These results demonstrate that incorporating active learning into multi-stage planning substantially improves decision-making under radical uncertainty and enhances the broad utility of active inference for modeling biologically relevant behaviors.
