“…Poorly designed AI may further generate feedback loops that reinforce inequalities, as in the case of predictive policing (Kaufmann, Egbert, & Leese, 2019), for example, or in predictions of creditworthiness that render it difficult for individuals to escape vicious cycles of poverty (O’Neill, 2014). Indirectly harmful outcomes of AIs can arise from the application of AI technologies more generally, often with long-term consequences, such as large-scale technological unemployment (Korinek & Stiglitz, 2018). Such outcomes can also take the form of so-called latent, secondary, and transformative effects (Mittelstadt et al, 2016) that occur when AI outcomes change the ways that people perceive situations, as, for example, in the case of profiling algorithms that powerfully ontologize the world in particular ways and trigger new patterns of behavior (Pasquale, 2015), though these effects are also evident in the ways that content curation and news recommendation algorithms lead to people being unwittingly socialized in “filter bubbles” (Berman & Katona, 2020).…”