Why, where and how do top-down and bottom-up signals interact in the primate brain?

Here we propose a roadmap for addressing hierarchical predictive coding theory (hPC) in the primate brain, theorizing that perception arises from the integration of top-down (TD) pathways carrying priors and predictions with bottom-up (BU) pathways carrying sensory evidence in the form of precision weighted prediction errors. However, given the conceptual importance of hPC for higher cognitive function in humans there is surprisingly little evidence concerning the location and functional mechanisms underlying such BU-TD integration. This will be overcome by parallel experiments in humans and non-human primates (NHPs). Molecular characterization in NHP of cortico-cortical BU and TD projection cell types should allow tracing these circuits and exerting causal control to explore functional mechanisms in visual paradigms in behaving NHPs. Similar paradigms in human subjects undergoing ultra-high field laminar fMRI scanning will allow investigation of corresponding pathways in the human brain. In NHP and human, the observation of long-distance TD signals carrying predictions suggest that the claustrum may be involved in hPC. Recent connectomic and transcriptomic characterisation of cortical claustral cortical loops may enable a similar approach to explore these pathways. Specifically, we hypothesize that the claustrum is uniquely positioned to mediate the precision-weighting of prediction errors in the sense of selecting salient (bottom-up) signals that drive representational or belief updating in cortical hierarchies. Progress in primate hPC will further emphasise the synergy between neurobiology and artificial intelligence with important consequences for both domains.