Data-Driven Innovation for Trustworthy AI

Abstract

Global concerns over the trustworthiness of rapidly proliferating artificial intelligence (AI)-centric artifacts have led to generic institutional recommendations for trustworthy AI, which have yet to be operationalized and integrated with design and innovation processes. We leverage the double hump model of data-driven innovation to propose and illustrate diverse data-driven approaches for identifying and evaluating opportunities, and generating and evaluating concepts for trustworthy AI. These approaches are expected to operationalize the institutional recommendations of trustworthy AI. Building on existing frameworks for classifying and managing risks associated with AI, we advocate for an ontological basis for trustworthy AI to enable fine-grained, computational assessments of AI-centric artifacts, their domains, and the organizations that develop or manage them.