Prof. Talia Gillis (Columbia) on the Fairness of Machine-Assisted Human Decisions

This podcast episode emphasizes the importance of considering human-AI interaction when designing algorithms for high-stake decisions. The optimal design of an algorithm depends on its implementation and how it will interact with human decision-makers. The episode discusses the implications of algorithmic predictions in both automated decision-making systems and human decision-making, exploring how the inclusion or exclusion of protected characteristics in algorithms can affect fairness. Furthermore, the impact of human beliefs on algorithm design is highlighted, with the recognition that algorithms can correct for biases held by human decision-makers. The episode concludes by discussing the need for regulatory approaches that consider the real-world implementation of algorithms through assistance, rather than treating algorithms as the sole decision-making factor.