This chapter considers the ethics of deploying machine tools for the task of moderating hate speech on social media. The benefits of deploying AI are substantial: platforms have a moral duty to moderate hate speech on their networks, and AI enables them to satisfy their moderation duties with far greater speed than if left solely to human moderators. Still, the use of AI for hate speech moderation presents considerable ethical risks. First, the opacity of AI systems can compromise our interests in navigating rules successfully and in holding power accountable. Second, the inaccuracy of AI systems can lead to false positives, whereby legitimate speech (such as reclaimed slurs or counter-speech) is erroneously deemed ‘hate speech’ and removed. Third, the inaccuracy of AI systems can produce false negatives, whereby harmful speech that ought to be removed is instead left up (for example, because of coded forms of hate speech or insufficient linguistic training). Finally, the notorious tendency of AI systems to be biased by their training data can lead these errors to be distributed unfairly, with already vulnerable groups subjected to greater burdens. This chapter argues that platforms have a duty to take ongoing steps to minimize each risk, explaining what those steps plausibly involve as part of a cooperative effort between humans and machine tools. Provided that platforms take these steps, the chapter argues that platforms remain justified in deploying AI for content moderation, notwithstanding the likelihood that errors will persist. Users can reasonably be expected to put up with some amount of over-enforcement produced by the system, as a justified collateral cost of the effort to prevent serious harm.
