Starting with pioneering work in the early 2010s that Hardt was involved in, algorithmic fairness has matured into a major research area spanning thousands of papers.

Synthesizing insights from more than a decade of research, Hardt and his co-authors Barocas and Narayanan completed the first dedicated textbook on fairness and machine learning, published with MIT Press in 2023 []. The book takes the reader from the normative foundations of algorithmic fairness to the conceptual and technical tools necessary to engage critically with this important subject.

The book complements Hardt’s general machine learning textbook, Patterns, Predictions, and Actions, co-authored with Benjamin Recht and published by Princeton University Press in 2022 [].

The scale and maturity of the field of algorithmic fairness invites retrospective analysis. Looking back at years of algorithmic fairness methods proposed by computer scientists, PhD student André Cruz led a computationally intensive meta study analyzing the effectiveness of different ways of equalizing the error rates of a predictive model across different groups in a population. What works across numerous datasets is the simplest method that was there first: a group-specific adjustment of decision thresholds post training [].

In 2023, the department welcomed group leader Ana-Andreea Stoica. Aside from diving into a new research direction on machine learning and economics, Stoica continued her work on algorithmic fairness. These include work on fairness in social networks [], the fairness-quality tradeoff in clustering [], and fairness in social influence maximization via optimal transport [].

These works extend the scope of our efforts on algorithmic fairness from prediction to the domain of social networks.