CONICET | Buscador de Institutos y Recursos Humanos

In recent years much has been learned about how a single computational processing step is implemented in the brain. By contrast, we still have surprisingly little knowledge of the neuronal mechanisms by which mul- tiple such operations are sequentially assembled into mental algorithms. We outline a theory of how individ- ual neural processing steps might be combined into serial programs. We propose a hybrid neuronal device: each step involves massively parallel computation that feeds a slow and serial production system. Production selection is mediated by a system of competing accu- mulator neurons that extends the role of these neurons beyond the selection of a motor action. Productions change the state of sensory and mnemonic neurons and iteration of such cycles provides a basis for mental programs.