MODERATION BY MACHINES:

THE ETHICS OF GOVERNING SPEECH WITH AI

Under preparation for the Oxford Handbook on Hate Speech; please email author to request a copy.

Online platforms enforce a range of rules that disallow various categories ofharmful speech. Enforcement relies on a combination of human moderators and machines, deploying sophisticated AI tools to detect and remove prohibited content. An increasing chorus of commentators argues that it is impermissible to engage in censorship-by-machine – that the normative significance of such decisions requires them to be made by humans. This paper canvasses various versions of this concern, both procedural and substantive. Procedural objections include concerns about the use of prior restraint; the opaqueness of ML decision-making; and disrespect. Substantive objections include concerns over how to infer speaker intent; the problem of algorithmic discrimination; and the risks of false positives and false negatives. I argue that while some of these objections have force, others are overstated or misguided. The upshot is a less pessimistic appraisal of the prospects of deploying machine tools for content moderation

Written by Jeffrey W. Howard