Does the Whole Exceed its Parts? The Effect of AI Explanations on Complementary Team Performance

Aug 30, 2026

Upload by Terk

Many studies motivate explainable AI by showing improved human–AI team performance when the AI explains its recommendations. Previous work, however, reported gains mainly when the AI alone outperformed both the individual human and the strongest human–AI team. This study asks whether explanations can produce complementary performance, where team accuracy exceeds either human-only or AI-only performance. Mixed-method user studies across three datasets show complementary improvements from AI augmentation, but explanations do not increase this effect; instead, they increase acceptance of the AI’s recommendation regardless of correctness, raising new challenges for fostering appropriate trust and improving complementary outcomes.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages16

Reading Time40 min

AccessFree to download

Abstract

Introduction

Human-AI teams and complementary performance

Role of explainability and appropriate trust

Keywords and research context

Does the Whole Exceed its Parts? The Effect of AI Explanationson Complementary Team Performance Gagan Bansal∗ Tongshuang Wu∗ [bansalg@cs.washington.edu](bansalg@cs.washington.edu)[ ](bansalg@cs.washington.edu)[wtshuang@cs.washington.edu](wtshuang@cs.washington.edu)[ ](wtshuang@cs.washington.edu)University of Washington Joyce Zhou† Raymond Fok† [jyzhou15@cs.washington.edu](jyzhou15@cs.washington.edu)[ ](jyzhou15@cs.washington.edu)[rayfok@cs.washington.edu](rayfok@cs.washington.edu)[ ](rayfok@cs.washington.edu)University of Washington Besmira Nushi [besmira.nushi@microsoft.com](besmira.nushi@microsoft.com)[ ](besmira.nushi@microsoft.com)Microsoft Research arXiv :2006 . 14779v 3 [ cs .AI] 12 Jan 2021 Ece Kamar [eckamar@microsoft.com](eckamar@microsoft.com)[ ](eckamar@microsoft.com)Microsoft Research Marco Tulio Ribeiro [marcotcr@microsoft.com](marcotcr@microsoft.com)[ ](marcotcr@microsoft.com)Microsoft Research Daniel S. Weld [weld@cs.washington.edu](weld@cs.washington.edu)[ ](weld@cs.washington.edu)[University of Washington &](University of Washington &)[ ](University of Washington &)Allen Institute for Artificial Intelligence ABSTRACT Many researchers motivate explainable AI with studies showing that human-AI team performance on decision-making tasks improves when the AI explains its recommendations. However, prior studies observed improvements from explanations only when the AI, alone, outperformed both the human and the best team. Can explanations help lead to complementary performance, where team accuracy is higher than either the human or the AI working solo? We conduct mixed-method user studies on three datasets, where an AI with accuracy comparable to humans helps participants solve atask (explaining itself in some conditions) . While we observed complementary improvements from AI augmentation, they were not increased by explanations. Rather, explanations increased the chance that humans will accept the AI’s recommendation, regardless of its correctness. Our result poses new challenges for human-centered AI: Can we develop explanatory approaches that encourage appropriate trust in AI, and therefore help generate (or improve) complementary performance? CCS CONCEPTS • Human-centered computing → Empirical studies in HCI; Interactive systems and tools; • Computing methodologies → Machine learning. KEYWORDS Explainable AI, Human-AI teams, Augmented intelligence ∗ Equal contribution. †Made especially large contributions. Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. Copyrights for components of this work owned by others than the author(s) must be honored. Abstracting with credit is permitted. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission [and/or a fee. Request permissions from permissions@acm.org](and/or a fee. Request permissions from permissions@acm.org). CHI’21, May 8–13, 2021, Yokohama, Japan © 2021 Copyright held by the owner/author(s) . Publication rights licensed to ACM. ACM ISBN 978-1-4503-8096-6/21/05. . . $15.00 [https://doi.org/10.1145/3411764.3445717](https://doi.org/10.1145/3411764.3445717) ACM Reference Format: Gagan Bansal, Tongshuang Wu, Joyce Zhou, Raymond Fok, Besmira Nushi, Ece Kamar, Marco Tulio Ribeiro, and Daniel S. Weld. 2021. Does the Whole Exceed its Parts? The Effect of AI Explanations on Complementary Team Performance. In CHI Conference on Human Factors in Computing Systems (CHI’21), May 8–13, 2021, Yokohama, Japan. ACM, New York, NY, USA, 16 pages. [https://doi.org/10.1145/3411764.3445717](https://doi.org/10.1145/3411764.3445717) 1 INTRODUCTION Although the accuracy of Artificial Intelligence (AI) systems is rapidly improving, in many cases, it rem

FAQs

What does the paper investigate about AI explanations?

It examines whether AI explanations can enable complementary team performance, meaning the team outperforms both the human alone and the AI alone.

What did the study find about complementary improvements?

AI augmentation produced complementary accuracy gains, but these gains were not increased by adding explanations.

How did explanations affect human behavior toward the AI?

Explanations increased the likelihood that humans accept the AI’s recommendation, independent of whether the recommendation was correct.

Aug 30, 2026

Upload by Terk