New Frontiers in Explainable AI: Understanding the GI to Interpret the GO

In this paper we focus on the importance of interpreting the quality of the input of predictive models (potentially a GI, i.e., a Garbage In) to make sense of the reliability of their output (potentially a GO, a Garbage Out) in support of human decision making, especially in critical domains, like medicine. To this aim, we propose a framework where we distinguish between the Gold Standard (or Ground Truth) and the set of annotations from which this is derived, and a set of quality dimensions that help to assess and interpret the AI advice: fineness, trueness, representativeness, conformity, dryness. We then discuss implications for obtaining more informative training sets and for the design of more usable Decision Support Systems.