Overview

The expected economic and societal importance of artificial intelligence creates one of the biggest regulatory challenges of our times: how to implement effective human oversight in automated decision-making (HOADM)?

HOADM is supposed to increase the accuracy, legitimacy, as well as fairness of decisions. However, merely placing a human agent in a hybrid human/machine system will hardly guarantee accurate, fair, or non-discriminatory outcomes, as emerging research on the novel subject shows. At the same time, people tend to trust artificial intelligence systems more if a human being is involved in its decision-making. This trust may be misplaced if the human overseer in the decision loop turns out to have little to no influence on the system’s outputs.

This project therefore takes a currently missing regulatory perspective by asking how norms such as laws and regulations can be improved to promote HOADM’s effectiveness and better protect people from harm. It combines legal analysis with empirical methods to

1. analyze how HOADM is mandated by law,
2. test how actual human overseers perform under the current legal status quo, and
3. make empirically based suggestions for improving HOADM’s implementation.