Such Stuff as Habits Are Made On : A Reply to Cooper and Shallice ( 2006 )

The representations and mechanisms guiding everyday routine sequential action remain incompletely understood. In recent work, the authors proposed a computational model of routine sequential behavior that took the form of a recurrent neural network (M. Botvinick & D. C. Plaut, 2004). Subsequently, R. P. Cooper and T. Shallice (2006) put forth a detailed critique of that work, contrasting it with their own account, which assumes a strict hierarchical processing system (R. P. Cooper & T. Shallice, 2000). The authors respond here to the main points of R. P. Cooper and T. Shallice’s (2006) critique. Although careful and constructive, the arguments offered by R. P. Cooper and T. Shallice (2006) mistook several superficial implementational issues for fundamental theoretical ones, underestimated the computational power of recurrent networks as a class, and in some ways mischaracterized the relationship between the accounts they compare. In responding to these points, the authors articulate several key theoretical choices facing models of routine sequential behavior.