Heuristica II: Updating a 2011 Game-Based Training Architecture Using Generative AI Tools

In 2011, the authors were part of a team of researchers working on an intelligence analyst project, Heuristica, exploring the use of serious games to teach intelligence analysts to recognize cognitive biases in their own decision-making and in the decisions of those they observed, and to learn to use strategies that would mitigate those biases. In this paper we provide an analysis of the architecture and extend the design to include components built around a large language model (LLM, e.g. ChatGPT). We call the new design Heuristica II. Our analysis consists of envisioning updated components and preliminary explorations of prompt structures that can be inserted into the components of the adaptive instructional system to advance their capabilities. The updated design will take into account lessons learned from the 2011 project and beyond. These explorations reveal the capabilities of using LLMs for adaptive training but also highlight some areas requiring improvement and caution.