Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Implications of AI Involvement for Trust in Expert Advisory Workflows Under Epistemic Dependence

Dennis Kim, Roya Daneshi, Bruce Draper, Sarath Sreedharan

TL;DR

The paper examines how AI involvement in expert advisory workflows shapes trust in the human advisor, the AI system, and the combined team using a simulated academic advising task with five workflow conditions. Employing a workflow-level design and validated trust measures (METI and multi-entity scales) across 77 participants, the study finds that advisory outcomes predominantly drive trust, though the pattern of AI involvement—especially advisor-initiated AI corrections after errors—can reduce perceived advisor expertise and willingness to reuse. The results suggest design implications favoring automatic AI oversight and transparent responsibility signaling, illustrating that how AI participates and is surfaced within the workflow matters as much as outcome accuracy. These findings advance our understanding of trust calibration in human-AI advisory contexts and inform the design of hybrid teams across domains where end users rely on expert guidance augmented by AI.

Abstract

The increasing integration of AI-powered tools into expert workflows, such as medicine, law, and finance, raises a critical question: how does AI involvement influence a user's trust in the human expert, the AI system, and their combination? To investigate this, we conducted a user study (N=77) featuring a simulated course-planning task. We compared various conditions that differed in both the presence of AI and the specific mode of human-AI collaboration. Our results indicate that while the advisor's ability to create a correct schedule is important, the user's perception of expertise and trust is also influenced by how the expert utilized the AI assistant. These findings raise important considerations for the design of human-AI hybrid teams, particularly when the adoption of recommendations depends on the end-user's perception of the recommender's expertise.

Implications of AI Involvement for Trust in Expert Advisory Workflows Under Epistemic Dependence

TL;DR

The paper examines how AI involvement in expert advisory workflows shapes trust in the human advisor, the AI system, and the combined team using a simulated academic advising task with five workflow conditions. Employing a workflow-level design and validated trust measures (METI and multi-entity scales) across 77 participants, the study finds that advisory outcomes predominantly drive trust, though the pattern of AI involvement—especially advisor-initiated AI corrections after errors—can reduce perceived advisor expertise and willingness to reuse. The results suggest design implications favoring automatic AI oversight and transparent responsibility signaling, illustrating that how AI participates and is surfaced within the workflow matters as much as outcome accuracy. These findings advance our understanding of trust calibration in human-AI advisory contexts and inform the design of hybrid teams across domains where end users rely on expert guidance augmented by AI.

Abstract

The increasing integration of AI-powered tools into expert workflows, such as medicine, law, and finance, raises a critical question: how does AI involvement influence a user's trust in the human expert, the AI system, and their combination? To investigate this, we conducted a user study (N=77) featuring a simulated course-planning task. We compared various conditions that differed in both the presence of AI and the specific mode of human-AI collaboration. Our results indicate that while the advisor's ability to create a correct schedule is important, the user's perception of expertise and trust is also influenced by how the expert utilized the AI assistant. These findings raise important considerations for the design of human-AI hybrid teams, particularly when the adoption of recommendations depends on the end-user's perception of the recommender's expertise.
Paper Structure (35 sections, 4 figures, 2 tables)

This paper contains 35 sections, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Contributions
  3. Hypotheses
  4. Hypotheses List
  5. Relevant Work
  6. Trust as Dynamic and Performance-Contingent
  7. Procedural Cues, Epistemic Roles, and Responsibility
  8. Measuring Trust in Multi-Entity Advisory Contexts
  9. Methodology
  10. Study Design
  11. Participants
  12. Measures
  13. Epistemic Trustworthiness (METI)
  14. Trust in the Advisor, AI System, and Team
  15. General Trust
  16. ...and 20 more sections

Figures (4)

  • Figure 1: Illustrative advising workflows examined in this study, highlighting differences in advisor performance and AI intervention combinations.
  • Figure 2: Interaction workflows for the academic advising paradigms examined in this study.
  • Figure 3: Items and dimensional structure of the Muenster Epistemic Trustworthiness Inventory (METI), sourced from hendriks2015meti
  • Figure 4: Screenshots from a mistake simulation shown in chronological order.