IS and HCI research on human involvement in AI-augmented decisions has been dominated by accuracy as the primary outcome metric. This framing works for decisions with an established ground truth such as fraud detection, medical triage, or credit scoring, where performance can be benchmarked against known outcomes. However, many consequential organizational decisions have no such baseline. Strategic recommendations, complex customer situations, and judgment-intensive assessments are shaped by context, values, and competing considerations rather than a correct answer. For these decisions, asking whether human involvement improves accuracy is not the adequate question. The relevant goals shift toward concepts such as procedural fairness, comprehensive contextual consideration, or the quality of reasoning. Existing research on these concepts has focused primarily on how affected parties perceive algorithmic decisions rather than on how human involvement in open-ended organizational workflows should be evaluated when no ground truth exists.

The thesis is supposed to answer the research question: "What are the goals of human involvement in AI-augmented decisions without an established ground truth, and how can these goals be operationalized?“

If you are interested in this thesis, feel free to reach out to Dominic Sieber (dominic.sieber@unisg.ch).