Addressing Ethics, Bias, and Equity in AI models: Methodologies and approaches

This is an expertise exchange in the CIRCLS’23 Expertise Exchange session

Session Leaders: Anirban Roy, SRI; Pati Ruiz, Digital Promise; Angela E.B. Stewart, University of Pittsburgh; Collin Lynch, North Carolina State University

In an era where AI models have become integral to learning science applications, it is essential to scrutinize equity in how we create, evaluate, and use such models. In this session, we explore theoretical framings and methodologies to work towards equity in AI learning science systems. This session will be framed around the following topics:

  • What is “fairness”? Here, we explore how we evaluate fairness in AI systems, and discuss different definitions for this concept.
  • Data and Social Construction: Data lies at the heart of AI models, but it often reflects and perpetuates the social inequities present in our world. This segment explores how data can introduce bias and inequity into AI systems. We will discuss real-world examples, highlighting the impact of data on AI outcomes.
  • Theories and Frameworks for Ethics in AI: We will discuss multiple theoretical lenses through which researchers can think about ethics in AI models. We will discuss how critical theories can inform the development of equitable AI systems.
  • Processes and Protocols: We will also consider how processes around AI – how it is designed, how it is introduced, and whom are the stakeholders – can mitigate or exacerbate ethical concerns.