Users as rational interacting agents: formalising assumptions about cognition and interaction

One way of assessing the usability of a computer system is to make reasonable assumptions about users' cognition and to analyse how they can be expected to work with the system, using their knowledge and information from the display to achieve their goals. This is the approach taken in Programmable User Modelling Analysis, a technique for predictive usability evaluation of interactive systems. The technique is based on the premise that an analyst can gain insights into the usability of a computer system by specifying the knowledge that a user needs to be able to use it and drawing inferences on how that knowledge will guide the user's behaviour. This may be done by observing how a cognitive architecture, " programmed " with that knowledge, behaves. An alternative approach is to develop a formal specification of the essential features of the cognitive architecture and to use that specification to reason about likely user behaviour. In this paper, we present the approach and an outline formal specification of the cognitive architecture. This initial specification is derived from an existing implementation. We illustrate how the specification can be used in reasoning by applying it to the task of setting up call diverting on a mobile phone. Successful performance of this task involves a combination of planned and responsive behaviour. The process of doing this analysis highlights what assumptions have been made by the designers about the user's knowledge. We discuss limitations of the current formalisation and identify directions for future work. Introduction One approach to predictive usability evaluation of an interactive computer system is to model how a user is likely to work with it., such user modelling techniques have given analytic leverage largely by providing a constrained language with which the analyst can lay out assumptions about the user and system, and occasionally by allowing the analyst to construct running simulation models of interactive behaviour. Programmable User Modelling Analysis (PUMA) is such a user modelling technique. It involves considering the design of a computer system in terms of the tasks which the system is designed to support, and what the user is expected to do to achieve those tasks. At the heart of the approach is the idea that a PUMA analyst should have access to a Programmable User Model (PUM) which can be programmed with domain and device specific knowledge and run to simulate interactive behaviour. However, rather than always aiming …