Checking the ACT-R/Brain Mapping Hypothesis with a complex task: Using fMRI and Bayesian identification in a multi-dimensional strategy space

John R. Anderson proposed a correspondence between ACT-R modules and brain regions. In his studies he compared ACT-R-predicted blood-oxygen-level dependent (BOLD) signal curves with BOLD curves obtained from functional Magnetic Resonance Imaging (fMRI) scans. Most of his studies were conducted solving simple algebra tasks. Using different and more complex problems dealing with the interpretation of chemical formulae, we studied the Brain Mapping Hypothesis’ robustness towards a different domain, multidimensional strategy spaces, and modeling errors. The ACT-R architecture tolerates various model implementations of the same task with similar behavior but different BOLD predictions. We repeated the analysis for six different models, each implementing a different strategy for the problem with the result that correlations vary between model-generated and empirical BOLD curves according to the selected problem-solving strategy. As an overall result we could not disconfirm Anderson’s Brain Mapping Hypothesis, but we could not rule out that ACT-R modules are distributed across more brain regions than Anderson suggested.