The Social Life of Algorithmic Harms | Data & Society

With artificial intelligence — computational systems that rely on powerful algorithms and vast, interconnected datasets — promising to affect every aspect of our lives, its governance ought to cast an equally wide net. Yet our vocabulary of algorithmic harms is limited to a relatively small proportion of the ways in which these systems negatively impact the world, individuals, communities, societies, and ecosystems: surveillance, bias, and opacity have become watchwords for the harms we anticipate that effective AI governance will protect us from. By extension, we have only a modest set of potential interventions to address these harms. Our capacity to defend ourselves against algorithmic harms is constrained by our collective ability to articulate what they look and feel like.

To expand our vocabulary of algorithmic harms, in 2022 Data & Society convened a workshop that asked researchers and advocates from around the world to consider novel algorithmic harms that are underappreciated by current approaches to AI governance, as well as methods that are emerging to better understand, evaluate, and assess these harms. Rather than start from the problems for which developers can most readily identify technical solutions — like privacy and unfairness, which have received the lion’s share of attention in the world of AI governance — this workshop began by looking at the social life of algorithmic harms.