RESEARCH – AI SIMULATION

The project MeMo (Mental Models) focused on the creation of detailed mental models of users and on the development of a workbench capable of evaluating interactive systems (GUI and also speech based) semi-automatically with respect to their usability.

The motivation behind the project was based on the tremendous costs for executing real user tests during the development of new software. To cut down these costs, a technology had to be developed that could simulate the behavior of specific user groups to a certain degree in order to find potential problems with the UX-design.  In a next step, these potentially problematic areas of the software should be evaluated in a real user test.

My task in the project to develop a structure that could represent the mental models and to implement a software capable of interpreting a user interface in order to come up with plausible, simulated user actions (based on the mental model being used). In order to realize this, I developed a complex, rule based algorithm.

The implemented rule system consisted of roughly 3000 UX-related rules which were “fired” during the simulation by combining the current UX-status (screen or speech interaction step) description with the mental model of the simulated user. At the beginning of the project, a real user test was executed with two different systems (the T-Mobile Online-Shop and a speech based Smart-Home system developed at DAI-Labor).

The results of these user tests were the benchmark for the simulated user tests that ran against the same systems. At the end of the project the simulation found very similar, major UX-issues in both systems.

We used Java und JESS (a rule based system implemented in Lisp) .