Humans have an impressive ability to form action plans in several domains of cognition; for example, planning routes to goals in spatial navigation, or the necessary steps to assemble complex objects, alone or together with other persons. However, the computations that underlie human individual and social planning remain largely unknown. This proposal aims to explain the ways humans face three key forms of uncertainty arising in planning domains; namely, uncertainty about task structure, action sequences, and the contributions of self and others to cooperative plans. To this aim, it advances a radically new theory about human planning, within a Bayesian approach that has been successfully adopted to explain uncertainties arising in perception and control. The theory under scrutiny is that humans plan using probabilistic inference based on hierarchical predictive codes (HPCs): compressed information or task abstractions that afford a powerful form of uncertainty-minimization, by highlighting salient junction points of the problem at hand, analogous to saliency maps for visual search. The methodology will combine empirical and computational modeling methods, to systematically validate the hypotheses of HPC theory about human planning in the face of uncertainties. A cornerstone of the methodology consists in conducting model-based analyses of human participants' behavior while they solve navigation-and-building tasks, alone or in dyads. This approach will permit us to compare the predictions stemming from HPC with those of alternative planning theories and ultimately, to understand the computations that underlie human planning. This ambitious proposal will produce groundbreaking advancements in our understanding of a high-level executive function - planning - while also contextualizing it within the influential theory of predictive processing. Our results will have important implications for psychology, neuroscience, philosophy, AI and robotics.