Fernanda Viégas: AI Dashboard Design: A User-Centered Approach to Interpretability

A primary goal of interpretability work is to make neural networks safer and more effective. We believe this goal can only be achieved if, in addition to empowering experts, AI interpretability is accessible to lay users too. I will describe an end-to-end application that ties recent advances in interpretability directly to the design of an end-user interface for chatbots. In particular, the application provides a real-time display of the chatbot’s “user model”—that is, an internal representation of the person it is talking with. I will discuss findings from an initial user study that suggest this dashboard can have a significant effect on people’s attitudes, changing their own mental models of AI, and making visible issues ranging from unreliability to underlying biases.