A Requirements-Driven Conceptual Modeling Framework for Responsible AI

The increasing impact and societal relevance of Data Science is accompanied by the onus to meet functional and business goals while simultaneously responding to key challenges of social responsibility such as (but not limited to) bias, fairness, and transparency in a rigorous and systematic manner. Goal oriented (GO) Requirements Engineering and Agent-oriented (AO) are promising techniques that can potentially offer systematic guidance to such issues, as well-established reasoning techniques that have long been used in other areas of system engineering. This doctoral research proposes a requirements driven GO and AO Conceptual Modeling framework to address the unique challenges of Responsible AI. The research will explore the potential inadequacies in current techniques and opportunities to develop novel conceptual modeling constructs, if necessary, with respect to the unique issues of Responsible AI. The framework intends to deal with conflicting goals at design decision points throughout the lifecycle of Machine Learning (ML) development and deployment, to guide the selection of design options that will simultaneously meet strategic business and social responsibility objectives.