An architecturally constrained model of random number generation and its application to modeling the effect of generation rate

Random number generation (RNG) is a complex cognitive task for human subjects, requiring deliberative control to avoid production of habitual, stereotyped sequences. Under various manipulations (e.g., speeded responding, transcranial magnetic stimulation, or neurological damage) the performance of human subjects deteriorates, as reflected in a number of qualitatively distinct, dissociable biases. For example, the intrusion of stereotyped behavior (e.g., counting) increases at faster rates of generation. Theoretical accounts of the task postulate that it requires the integrated operation of multiple, computationally heterogeneous cognitive control (“executive”) processes. We present a computational model of RNG, within the framework of a novel, neuropsychologically-inspired cognitive architecture, ESPro. Manipulating the rate of sequence generation in the model reproduced a number of key effects observed in empirical studies, including increasing sequence stereotypy at faster rates. Within the model, this was due to time limitations on the interaction of supervisory control processes, namely, task setting, proposal of responses, monitoring, and response inhibition. The model thus supports the fractionation of executive function into multiple, computationally heterogeneous processes.