Ensuring Human Agency: A Design Pathway to Human-AI Interaction

The augmentation of human work through Artificial Intelligence (AI) promises to be a panacea to the role of technology in organizations. While frameworks on augmentation theorize how to best divide work between humans and AI, the empirical literature on human-AI interaction offers unexpected and inconclusive findings. Interaction challenges—including overreliance and selected engagement with the algorithmic output—call into question how theorized augmentation benefits can be realized. Rooted in cognitive learning theory, we develop a conceptual model on the design of algorithmic output. We argue that human-AI interaction can lead to multiple beneficial outcomes when algorithmic output is designed in a reciprocal manner. By providing humans with reflection-provoking feedback, reciprocal algorithmic output does not prescribe any actions, and thereby necessitates humans to expend cognitive effort. We identify three crucial augmentation outcomes that reciprocal algorithmic output enables: task performance, human agency, and human learning.