Figure 1 Control behavior subject vs. ACT-R/PM
As an example, Figure 1 illustrates the prediction of an ACT-R/PM model in comparison to the actual control behavior of a user when operating a simulation of an existing coal-fired power plant. To control the power plant, users had to fix critical variables (supply of coal in kg/s, opening of the turbine valve in %) on a graphical user interface to follow given time-dependent power curves while maintaining a constant pressure in the steam generator. Close analysis of the knowledge the model acquired strongly suggests to maintain a numerical representation of system variables in addition to a graphical display of the system trajectory on the user interface since successful performance of the model presupposes both information sources. For details of the model see Wallach ( 1998).
To be generally useful for practical HCI purposes a cognitive architecture must provide accurate qualitative and quantitative a priori predictions, must be costeffective to use and -- most important -- models developed in a cognitive architecture must be understandable to system designers ( John, 1996). From the viewpoint of cognitive architectures, applications in HCI provide a strict and demanding test bed for the theory underlying an architecture -- HCI and cognitive architectures thus live in a mutual beneficial relationship.
Anderson, J. R. & Lebiere, C. ( 1998). The atomic components of thought. Mahwah, NJ: Erlbaum.