Seeing Like an Algorithmic Error: What are Algorithmic Mistakes, Why do They Matter, How Might they be Public Problems?

 As digital devices, massive datasets, and machine learning techniques proliferate, computational algorithms increasingly, invisibly, and often inexplicably shape many social, political, and cultural dynamics. In everything from law and politics to commerce and art, algorithms are powerful structuring logics and sociotechnical forces. But what does it mean when algorithms “fail”? What do we learn about the logics and forces that create algorithms when they are seen to have erred or made a mistake? Seeing algorithms as culture, I argue that algorithmic errors are similarly constructs made by intertwined computational, psychological, organizational, infrastructural, discursive, and normative forces. In this talk I tell three stories of algorithmic error, illustrate their sociotechnical dynamics, and examine the institutional and normative forces that define “failure.” Instead of seeing algorithmic errors as unavoidable or self-evident, I instead see them as evidence of how people think systems should work, who has the power to declare failure, which harms trigger which fixes, and how defining and repairing algorithmic mistakes bounds public problems.