DETER’s DASH, Formal Human Systems Laboratory to Explore Collaboration in Computational Representations of Human Mental Models

DETER researcher Jim Blythe recently gave an invited talk on DASH (DETER Agents for Simulating Humans) to the University at Buffalo's Department of Industrial and Systems Engineering. Jim also met with Assistant Professor Matthew L. Bolton and members of his Formal Human Systems Laboratory (FHSL) to explore potential collaboration on computational representations of human mental models for cybersecurity. These models are important for predicting human behavior both in multi-agent simulations using DASH and in verification approaches used by FHSL, whose use of an extended version of the DASH representation would provide a foundation for their new work as well as new use cases for DASH.

The DASH platform for building agent-based models of human behavior is one of the few that includes support for encoding mental models. Bolton's team works on incorporating human behavior models in formal verification of safety-critical systems that depend on human interaction, and Ph.D. student Adam Houser is focusing on mental models. We are exploring how the models used in DASH might also be used in Bolton and Houser's work, allowing us to share models and combine the benefits of model-checking approaches and agent-based simulation that may be more powerful than either approach alone.

Human behavior is a key determining factor in assessing the effectiveness of an organization's cyber defenses, including deployed hardware and software defenses, and also the policies put in place for individuals in the organization. Mental models are encapsulations of the reasoning processes we follow in decision-making. They typically involve simplifications of the decision domain that help us reason efficiently and often make use of analogies to domains that are better understood. However, poor models can lead well-intentioned individuals into suboptimal decisions. Mental models can be elicited through interviews and observation, helping researchers predict likely human responses to choices that affect cybersecurity.