Hierarchies and networks: Toward heterarchical predictive coding

This paper proposes a reinterpretation of predictive coding (PC) in terms of heterarchical networks of control mechanisms. Standard accounts of PC typically assume a rigid hierarchical organization, in which information flows in fixed directions across levels of abstraction. Such a framework has, however, faced growing theoretical criticism and appears increasingly difficult to reconcile with recent research on predictive routing, which suggests that predictive systems exhibit greater flexibility than hierarchical models allow. In response to these challenges, we propose integrating work on PC with both complex network theory and William Bechtel's conception of control mechanisms. Our central hypothesis is that predictive systems can be understood as emergent patterns arising from the interaction between hierarchical production mechanisms and heterarchical networks of control mechanisms. On this view, local computational hierarchies are dynamically modulated by flexible contextual constraints, thereby preserving the advantages of hierarchical models while avoiding their rigidity. The paper is theoretical in nature and highlights the need for empirical validation of the proposed framework, as well as for a systematic integration of network methodologies with research on predictive processing.