Diversity in sociotechnical machine learning systems

There has been a surge of recent interest in sociocultural diversity machine learning research. Currently, however, there is gap between discussions measures and benefits learning, on the one hand, broader research underlying concepts precise mechanisms its functional benefits, other. This problematic because not monolithic concept. Rather, different are based distinct rationales that should inform how we measure given context. Similarly, lack specificity about underpinning diversity’s potential can result uninformative generalities, invalid experimental designs, illicit interpretations findings. In this work, draw philosophy, psychology, social organizational sciences to make three contributions: First, introduce taxonomy from philosophy science, explicate epistemic political these concepts. Second, provide an overview by which benefit group performance. Third, situate taxonomies lifecycle sociotechnical systems case for their usefulness fair accountable learning. We do so illustrating they clarify discourse around context systems, promote formulation more questions impact, conceptual tools further advance practice.