Transparency In Artificial Intelligence

As AI models and services are used in a growing number of high-stakes areas, a consensus is forming around the need for a clearer record of how they are created to increase trust.  Several proposals for higher quality and more consistent AI documentation have emerged, addressing technical details along with ethical concerns and potential social impacts.  This talk will provide an overview of FactSheets, IBM’s approach to AI documentation.  FactSheets will typically contain model purpose, performance, safety, security, and fairness information collected at various points in the AI lifecycle.  We will review some of these common elements and discuss a methodology for creating useful FactSheets.