When
Title: Compensatory algorithms and (un)ethical manipulation
Abstract: High-consequence decisions are increasingly being informed or guided by AI algorithms, though the final decision stays under human control. That is, AI decision making involves not algorithms in isolation, but instead algorithms interacting with human decision-makers. In this talk, I will focus on compensatory algorithms: ones that compensate (in some sense) for our cognitive, social, and ethical biases. I will first argue that compensatory algorithms are not an unusual edge case, but rather naturally result from some standard practices in AI. I will then present an ethical analysis of such algorithms, concluding that they are sometimes unethical, but there are also conditions in which they can be deployed ethically even without the (explicit) consent of the user. I will close by connecting these issues to broader concerns about the ethics of manipulation via algorithms.