Simulating human users in dynamic environments

In the discipline of Human-Computer Interaction, user interface design engineers have tools such as interface tool kits that assist them in development of the computer side of the interaction, but are lacking in tools for the human side. A kind of tool that would be valuable to design engineers is one that would evaluate and test different designs by interacting with them as a human user would, exhibiting the same constraints and variability as human users. As a step towards fulfilling this need on the human side, I developed a Simulated Human User and a Simulated Task Environment, named Argus, for the judgment and decision making domain.; The contribution presented here extends modeling human-computer interaction beyond the typical laboratory domain to a complex and dynamic domain. The domain is complex in the sense that many aspects of the task are unconstrained, thereby allowing users to adopt different interactive strategies (i.e. different combinations of user actions) to accomplish a task goal. The domain is dynamic in the sense that the environment changes over time and there is a time pressure to respond to the changes.; The Simulated Human User is a computational cognitive model implemented under the ACT-R 5.0 embodied cognitive architecture. The model is the most comprehensive and successful ACT-R 5.0 model of human-computer interaction. The model is the first ACT-R 5.0 model to stress the importance of parallelism among cognitive, perceptual, and motor operations in HCI. This dissertation extends the ACT-R 5.0 architectural mechanisms to determine strategies for visual search at the cognitive level. It also implements changes to the visual attention mechanism of the vision module.; Argus is a family of Simulated Task Environments for performing decision making tasks. The design features extensive task environment configuration and control facilities. The system collects and integrates data at a finer grain size than other existing simulated task environments.; The model correctly predicts the overall performance decline on a decision task when workload is increased by adding another task. The results indicate that veridical simulation of human users in complex dynamic environments can be achieved at one level by capturing the range of strategies they employ at the next lower level. Incorporating a Simulated Human User into a design tool will enable user interface designers to make decisions based on quantitative results.