Human-Centered Artificial Intelligence: A Field Experiment

Abstract

Humans and artificial intelligence (AI) algorithms increasingly interact on unstructured managerial tasks. We propose that tailoring this human-AI interaction to align with individuals’ cognitive preferences is essential for enhancing performance. This hypothesis is examined through a field experiment in a multinational pharmaceutical firm. In the experiment, we manipulated four contextual parameters of human-AI interaction—work procedures, decision-making authority, training, and incentives—to align with sales experts’ cognitive styles, categorized as either adaptors or innovators. Our results show that tailored interaction significantly improves sales performance, whereas untailored interaction results in negative treatment effects compared with both the tailored and control conditions. Qualitative evidence suggests that this negative outcome arises from role conflicts and ambiguities in untailored interaction. Exploring the mechanisms underlying these outcomes further, a mediation analysis of AI login data reveals that human-AI interaction tailoring leads sales experts to adjust their AI utilization, which contributes to the observed performance outcomes. These findings support a human-centered approach to AI that prioritizes individuals’ information-processing needs and tailors their interaction with AI accordingly. This paper was accepted by Catherine Tucker, Special Issue on the Human-Algorithm Connection. Funding: This work was supported by Erling Persson Family Foundation; Schweizerischer Nationalfonds zur Förderung der Wissenschaftlichen Forschung [Grants 100013M 204670, 181364, 185164]; Jan Wallanders och Tom Hedelius Stiftelse samt Tore Browaldhs Stiftelse [Grant W20-0036]; Marianne and Marcus Wallenberg Foundation [Grants 2021.0074, 2021.0133]. Supplemental Material: The online appendix and data files are available at https://doi.org/10.1287/mnsc.2022.03849 .

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 48 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.