Algorithmic Fairness, Privacy, and Ethics: Bridging Technical Solutions and Philosophical Dilemmas

This episode features Michael Kearns, co-author of "Ethical Algorithm," delving into the complex interplay of algorithmic fairness, privacy, and ethics. The discussion highlights the technical advancements in machine learning and AI, juxtaposed with the growing "buzzkill" of their antisocial behaviors. Kearns emphasizes that while algorithmic approaches can address "low-hanging fruit" issues like egregious unfairness or privacy violations, the deeper, more nuanced problems quickly lead into profound philosophical questions about justice, discrimination, and societal values. He notes a significant rift between how philosophers and computer scientists approach these concepts, with the latter needing concrete, implementable definitions.

The conversation distinguishes between algorithmic privacy, where differential privacy has emerged as a widely accepted definition, and algorithmic fairness, which remains a "mess" due to inherent tensions and trade-offs. Kearns explains that multiple desirable notions of fairness often cannot be simultaneously achieved, making it a much harder problem. He introduces the idea of measuring algorithmic ethics quantitatively, for instance, by assessing the disparity in false rejection rates between different demographic groups in a lending model. However, he acknowledges that current definitions often focus on group-level fairness, which may not align with an individual's subjective sense of being treated fairly.

A key distinction explored is the tension between group-level and individual-level fairness. Kearns introduces the concept of "fairness gerrymandering," where an algorithm might meet fairness criteria for marginal attributes (e.g., race, gender) but still discriminate against specific combinations of these attributes (e.g., disabled Hispanic women over 55). He suggests that achieving individual fairness might involve progressively refining group definitions. The discussion also touches on the challenge of eliciting subjective fairness from ordinary people, rather than relying solely on scholarly definitions, advocating for more behavioral research in AI to understand how people truly perceive fairness and interpretability.

Finally, the podcast explores the broader societal implications, noting that while privacy is generally seen as universally desirable, fairness is a highly politicized topic that inevitably revisits unresolved debates like affirmative action. Kearns highlights that fairness discussions often involve difficult trade-offs and questions of "fair to whom, at what expense to who else," citing examples from criminal justice where fairness for offenders might be prioritized over fairness for victims. The episode underscores that converting ethical concepts into algorithms forces humanity to confront and define its own moral standards in an unprecedented way.

"here are three entirely reasonable desirable notions of fairness and you know here's a proof that you cannot simultaneously have all three of them"

"most people are good and want to do to do right and that deviations from that or you know kind of usually due to circumstance"

"these worlds kind of developed their own social norms and they develop their own rationales for you know behavior for instance that might look unusual to outsiders but when you're in that world it doesn't feel unusual at all"

"for this lending model the false rejection rate on black people and white people is within 3 percent"

"your compensation is the knowledge that we are we are also falsely denying loans to other people you know other groups at the same rate that we're doing it's to you"

"we really actually don't know from a subjective standpoint like what people really think is fair"

"computer scientists kind of don't think of any scientific topic as off limits to them they will like freely wander into areas that others have been thinking about for decades or longer and you know we usually tend to embarrass ourselves"

"asking for fairness at the individual level is to sort of ask for group fairness simultaneously for all possible combinations of groups"

"we call that fairness gerrymandering because like political gerrymandering you know you're giving some guarantee at the aggregate level yes but that when you kind of look in a more granular way at what's going on you realize that you're achieving that aggregate guarantee by sort of favoring some groups in discriminating against other ones"

"fairness is sort of special in that as soon as you start talking about it you inevitably have to participate in debates about fair to whom at what expense to who else"