Detail:In this project, we conduct controlled experiments with human subjects using visual interfaces to detect patterns of biased decision-making and reasoning under uncertainty. In the context of visual analytics, confirmation bias can manifest itself as the tendency to look for and interpret data that favors a preconceived hypothesis. Anchoring bias can be seen in the way humans tend to insufficiently adjust initial estimates that have been anchored in spite of evidence suggesting otherwise. In addition, reasoning under uncertainty can be manifested, for example, in the temporal gaps between assumptions and reasoning steps taken by humans as they interact with the visual interface. The experiments are designed with specific tasks given to the subjects to complete using pre-existing visual analytics software. We developed multiple visual analytics software that aim to support decision-making and reasoning in the domains of financial risk analysis, financial fraud detection, event analysis based on social media data, and future event forecasting. Entry and exit surveys are used to determine the human subjects’ tendency towards bias and their critical thinking capabilities. We conduct a series of experiments with each cycle of experiments testing a well-crafted set of hypotheses. Logging software continually tracks the activities of the subjects as they proceed through their task. These activity logs are used conduct post-mortem analysis and detection of generalized patterns of behavior that can be correlated with and validated against the entrance and exit survey responses. Once the generalized patterns of behavior can be robustly modeled, we develop techniques to deliver automatic interventions at opportune moments to the humans. Such interventions can be in the form of alerts or popup messages or other subtle signals in the interface that would help humans become aware of flaws in their reasoning and help mitigate their biases. Our prior work in creating automated agents to influence humans via online dialogues is leveraged to develop such technology.

Publications from this project

C38, C44, C53