Abstract

The COVID-19 pandemic has required individuals to understand and respond to evolving threats during their daily lives. Our goal is to develop an interactive risk counselor to help people evaluate risk-related implications of participating in daily situations and consider behaviors to mitigate risk. Automated tools have already been developed to help determine whether users may have COVID and should seek care, or to help users evaluate daily risk and provide advice and guidelines. Our tool will use a conversational interface to educate users about risk in a wider range of scenarios and provide recommendations that simplify risk perception. As an initial step, we conducted a survey to elicit people’s concerns about COVID risk in daily situations. Participants were asked to identify situations and for each situation, they were asked about dimensions relevant to risk, similar to the questions we envision the agent asking. Participants responded to our survey in a variety of ways. Some unexpectedly responded with comments, expressing opinions or observations about the pandemic circumstances. Others responded as we expected, describing a variety of objectively risky situations (e.g., indoor activities with little social distancing), and many reported moderate/extreme levels of concern about these situations. The descriptions, whether comments or situations, seemed to cluster into categories. We report several analyses of the survey data that help characterize differences between these two general kinds of responses. We next plan to confirm our initial observations that the data set enables reliable classification of user comments and situation descriptions. The classifiers would enable the agent to respond to a diverse user goals, including recognizing types of situations with known risk attributes related to the pandemic, so the agent can ask questions to help users evaluate situational risk. User responses will in turn allow the agent to educate the users about sources of risk mitigating behaviors.