Exploring bounded rationality in human-AI decision-making: From individual choices to team outcomes
EMBARGOED UNTIL 2028-05-30
Loading...
Author / Producer
Date
2025
Publication Type
Doctoral Thesis
ETH Bibliography
yes
Citations
Altmetric
EMBARGOED UNTIL 2028-05-30
Data
Rights / License
Abstract
Despite rapid advancements in Artificial Intelligence (AI), the realization of effective human-AI decision-making faces critical challenges. These include understanding AI integration preferences, improving decision accuracy under conditions of uncertainty, and facilitating the integration of decision-relevant information from both human and AI. At the core of these challenges lies the question of how humans make decisions when influenced by external boundaries (e.g., information accessibility, the behavior of other agents) and internal boundaries (e.g., knowledge, attitudes, beliefs). Although this question is closely related to the study of bounded rationality in human-only judgment and decision-making (JDM) literature, current knowledge exchange between JDM and human-AI interaction research remains limited. Addressing this lack of integration holds significant potential for advancing integrative approaches to human-AI decision-making and enriching both fields with cross-disciplinary perspectives.
The overarching aim of this thesis, spanning three dissertation papers, is to seize this potential to deepen our understanding of how internal boundaries shape human-AI decision-making across contexts characterized by certain external boundaries.
The first paper investigates how internal boundaries (i.e., AI knowledge, and implicit and explicit comparative trust associations with AI vs. physicians) relate to AI integration choices when AI’s risk and benefits are unknown (external boundary). Based on survey data from 378 U.S. citizens, we find that internal boundaries relate to the perceptions of external boundaries, with distinct patterns for risk vs. benefit perceptions. The second paper proposes that ambivalent, i.e., two-dimensional, attitudes toward AI (internal boundary) may benefit decision-making when individuals receive AI recommendations with unknown accuracy (external boundary). We test this assertion through experiments with samples of 109 (study 1) and 148 (study 2) medical students. The findings reveal that ambivalent attitudes—compared to more univalent negative or positive attitudes—foster more balanced, i.e., moderate, intentions to rely on AI (study 1). However, no differences in reliance behavior were observed (study 2). The third paper examines whether AI-based information inquiry (external boundary) enhances information sharing and decision-making performance in teams where decision-relevant information is distributed (external boundary). We also test whether this effect depends on team performance expectations (internal boundary). Experimental data from 98 human-AI teams show that information inquiry improves information sharing and, consequently, team decision-making performance, regardless of performance expectations. A comparison with a human-only team condition (65 teams) revealed that overall decision-making processes are the same, with subtle differences in communication patterns.
Collectively, these findings contribute to a deeper understanding of critical challenges in human-AI decision-making. Specifically, we identify factors that influence individuals’ AI integration preferences, propose a new perspective on the dimensionality of attitudes toward AI in AI-assisted decision-making, and highlight AI behavior that facilitates the integration of decision-relevant information in human-AI teams. In addition, this work offers recommendations for practice regarding the promotion of human-centered AI design and implementation.
Permanent link
Publication status
published
External links
Editor
Contributors
Examiner : Grote, Gudela
Examiner: Schmutz, Jan
Book title
Journal / series
Volume
Pages / Article No.
Publisher
ETH Zurich
Event
Edition / version
Methods
Software
Geographic location
Date collected
Date created
Subject
human-AI interaction; human-AI teams; attitudes; trust; ambivalence; Knowledge; expectations; decision-making; performance
Organisational unit
03356 - Grote, Gudela / Grote, Gudela