Researchers from the Pennsylvania State University College of Information Sciences and Technology created an artificial intelligence (AI)-based online symptom checker that explains its recommendations to the user. This helps users avoid unnecessary worry, medical trips or possibly harmful self-medication.
Symptom checkers have been used for years but their use skyrocketed in the last year thanks to COVID-19. But currently, it is not clear to the user why online symptom checkers make the recommendations they do. The team set out to create a more transparent symptom checker and study user perceptions of the given recommendations.
Current symptom checkers are powered by machine learning algorithms which analyze information provided by users to guide the checker toward a possible diagnosis. The lack of transparency and comprehensible language could cause unintended consequences if the user doesn’t fully understand the given recommendations. For example, telling a user that they may have COVID could lead to poor medical decisions like taking a medication on their own over seeking proper medical treatment.
The team interviewed users of online symptom checkers to understand how explanations would improve user experience and their trust in online tools. The interviews showed that users are often confused by questions that chatbots ask and which symptoms or information led to the suggested diagnosis and advice.
For the study, the team reproduced the user’s interaction with the online symptom checker and added explanations for the chatbot questions and how the recommendations were created. The team’s online symptom checker included three types of explanation styles. The first was rationale-based, where it provided an explanation of each question given to the user. The second was feature-based, which offered a personalized summary based on the user’s answers. The final style was example-based, which highlighted an identical example of a patient who received the same clinical recommendation as the user based on identical answers.
When using the new symptom checker, users reported feeling more confident in the checker's recommendations when they received explanations. Results proved that explanations could significantly improve user experience and facilitate medical decision making and improve user trust in the diagnosis.
These findings could be used to inform the future design of online symptom checkers and help users navigate medical issues.
This research will be presented at the 2021 ACM CHI Conference on Human Factors in Computer Systems.