”It’s Everybody’s Role to Speak Up... But Not Everyone Will”: Understanding AI Professionals’ Perceptions of Accountability for AI Bias Mitigation

In this paper, we investigate the perceptions of AI professionals for their accountability for mitigating AI bias. Our work is motivated by calls for socially responsible AI development and governance in the face of societal harm but a lack of accountability across the entire socio-technical system. In particular, we explore a gap in the field stemming from the lack of empirical data needed to conclude how real AI professionals view bias mitigation and why individual AI professionals may be prevented from taking accountability even if they have the technical ability to do so. This gap is concerning as larger responsible AI efforts inherently rely on individuals who contribute to designing, developing, and deploying AI technologies and mitigation solutions. Through semi-structured interviews with AI professionals from diverse roles, organizations, and industries working on development projects, we identify that AI professionals are hindered from mitigating AI bias due to challenges that arise from two key areas: (1) their own technical and connotative understanding of AI bias and (2) internal and external organizational factors that inhibit these individuals. In exploring these factors, we reject previous claims that technical aptitude alone prevents accountability for AI bias. Instead, we point to interpersonal and intra-organizational issues that limit agency, empowerment, and overall participation in responsible computing efforts. Furthermore, to support practical approaches to responsible AI, we propose several high-level principled guidelines that will support the understanding, culpability, and mitigation of AI bias and its harm guided by both socio-technical systems and moral disengagement theories.