As the pandemic progresses, decision-makers must move from a blanket approach to more precise, sustainable restrictions tailored to geographical regions, social groups, households, and individuals. Consequently, there is a need for more precise and accurate models, and especially models that make predictions about new kinds of behavior change (e.g., to wear masks to restaurants or places of worship). To make progress on these challenges, we propose to research novel computational models of human responses to epidemics and NPIs and integrate them into a state-of-the-art epidemiological model.
Uncertainty in models can arise from low-fidelity computational modeling of specific local conditions and the social-psychological reactions of individuals in their population. For decision-makers, a useful model of the efficacy of implementation of NPIs (such as social distancing) will depend not only on traditional epidemiology, but on a host of factors including locale, social norms, attitudes, intentions, etc. Most importantly, those individual response parameters change dynamically with the evolution of the situation and cannot simply be estimated from past data.
Figure 1. The proposed system will use ACT-R agents (center) to model behavior change (e.g., social distancing; bottom) in response to information dynamics (left), which have dynamical effects on the perceptions, attitudes, beliefs, intentions, and knowledge (top) influencing decisions and actions (right).
Figure 2. A simple SIR model that incorporates the decision to wear a mask as part of the ACT-R cognitive architecture. The three conditions each instantiate a homogeneous population of 100K agents and differ with the degree to which the epidemic dynamics affect mask-wearing behavior. Hope agents discount increases in daily case rate while fear agents are more sensitive to it (neutral is about 50/50). This is a prototype for understanding how person-level decision-making in epidemics might interact with the information environment.