pith. sign in

arxiv: 2604.22319 · v1 · submitted 2026-04-24 · 💻 cs.HC

Rethinking AI-Mediated Minority Support in Power-Imbalanced Group Decision-Making: From Anonymity To Authenticity

Soohwan Lee , Kyungho Lee This is my paper

Pith reviewed 2026-05-08 10:48 UTC · model grok-4.3

classification 💻 cs.HC
keywords AI-mediated communicationminority voicesanonymityauthenticitypsychological safetygroup decision-makingpower imbalanceLLM
0
0 comments X

The pith

In hierarchical groups, AI-anonymized minority input boosts participation yet reduces psychological safety and satisfaction, unlike autonomous counterarguments.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper investigates two ways AI can support minority input during group decisions where some voices hold more power. One approach has AI pass along minority ideas without revealing who said them, which led to more people contributing but also made participants feel less safe and less happy with the process. The other approach has AI produce its own counterarguments without relaying specific minority statements, which raised satisfaction and lowered feelings of exclusion. These results matter because many AI tools assume hiding identity will help minorities, yet the evidence here shows that approach can undermine the sense of agency and authenticity that actually supports fair outcomes.

Core claim

Relaying minority input anonymously through AI increased participation but significantly reduced psychological safety and satisfaction, while generating only autonomous counterarguments improved satisfaction and reduced marginalization. These counterintuitive findings reveal three provocations for AIMC design in hierarchical contexts: the inherent trade-offs among anonymity, authenticity, agency, and accountability; the risk that power asymmetry reverses intended effects; and the need for AI to facilitate group reflection rather than substitute for human responsibility.

What carries the argument

Direct comparison of two LLM-powered minority support strategies in hierarchical group decision-making: anonymous relaying of input versus generation of autonomous counterarguments.

If this is right

  • AI tools meant to protect minorities in uneven groups must weigh anonymity against the loss of perceived authenticity and agency.
  • Power differences can flip the expected benefits of AI anonymity, leading to lower rather than higher inclusion.
  • Effective AI support should prompt groups to reflect on their own dynamics instead of handling minority input on their behalf.
  • Design choices in AIMC need to treat anonymity and authenticity as distinct constructs with separate effects on satisfaction.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Hybrid systems could let minorities choose when to stay anonymous and when to claim their arguments directly.
  • The patterns might appear in other settings with status gaps, such as workplace meetings or online policy forums moderated by AI.
  • Longer-term tracking could check whether the satisfaction gains from autonomous arguments lead to lasting changes in group behavior.
  • Testing the same strategies across cultures or group sizes would clarify how widely the trade-offs apply.

Load-bearing premise

The measured changes in participation, psychological safety, and satisfaction stem directly from the choice of AI mediation strategy rather than from the power hierarchy itself or other unexamined parts of how the study was run.

What would settle it

A controlled follow-up trial that applies the same group tasks and measures but removes the AI component entirely, then finds no corresponding shifts in safety or satisfaction when minority ideas are still kept anonymous.

Figures

Figures reproduced from arXiv: 2604.22319 by Kyungho Lee, Soohwan Lee.

Figure 1
Figure 1. Figure 1: An LLM-powered minority support system mediates majority-minority dynamics through two designs: AIGC, which view at source ↗
Figure 2
Figure 2. Figure 2: Four patterns of AI-mediated group communication[ view at source ↗
read the original abstract

AI-mediated Communication (AIMC) systems increasingly aim to protect minority voices by anonymizing or proxying their input, but anonymity and authenticity are not the same construct. This position paper draws on an ongoing empirical study comparing two LLM-powered minority support strategies in hierarchical group decision-making. We found that relaying minority input anonymously through AI increased participation but significantly reduced psychological safety and satisfaction, while generating only autonomous counterarguments improved satisfaction and reduced marginalization. These counterintuitive findings reveal three provocations for AIMC design in hierarchical contexts: the inherent trade-offs among anonymity, authenticity, agency, and accountability; the risk that power asymmetry reverses intended effects; and the need for AI to facilitate group reflection rather than substitute for human responsibility. These findings and provocations are offered as a contribution to the Restoring Human Authenticity in AI-Mediated Communication workshop.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 0 minor

Summary. This position paper in HCI draws on preliminary results from an ongoing empirical study to examine two LLM-powered strategies for supporting minority voices in hierarchical group decision-making. It reports that anonymously relaying minority input via AI increased participation but significantly reduced psychological safety and satisfaction, while generating only autonomous counterarguments improved satisfaction and reduced marginalization. The paper uses these observations to articulate three provocations for AIMC design: inherent trade-offs among anonymity, authenticity, agency, and accountability; the risk that power asymmetry can reverse intended effects; and the need for AI to facilitate group reflection rather than substitute for human responsibility. The work is positioned as a contribution to the Restoring Human Authenticity in AI-Mediated Communication workshop.

Significance. If the reported differences prove robust under controlled conditions, the findings would carry moderate significance for the CSCW and HCI communities by challenging the assumption that anonymity reliably protects minorities in power-imbalanced settings. The three provocations offer a useful conceptual frame for future AIMC systems that prioritize authenticity and shared accountability over proxy mechanisms. The paper earns credit for surfacing counterintuitive empirical patterns and for framing them as design provocations rather than prescriptive guidelines.

major comments (2)
  1. [Abstract] Abstract and the description of the ongoing empirical study: the claims that anonymous AI relaying 'significantly reduced psychological safety and satisfaction' and that autonomous counterarguments 'improved satisfaction and reduced marginalization' are presented without any reported sample size, randomization procedure, statistical tests, effect sizes, or control conditions. These details are load-bearing because the three provocations rest directly on the attribution of the observed differences to the specific AI mediation strategies rather than to unmeasured factors in the hierarchical group setting.
  2. [Empirical Study] The empirical study description: no baseline measures of group dynamics, order effects, or AI presence as a potential confound are mentioned, leaving open the possibility that the participation increase and safety decrease are driven by the power-imbalanced context itself rather than the anonymity versus authenticity manipulation. This directly weakens the central claim that the two strategies produce the reported divergent outcomes.

Simulated Author's Rebuttal

2 responses · 1 unresolved

We thank the referee for their insightful comments, which highlight important areas for clarifying the evidential basis of our position paper. We respond to each major comment below, proposing revisions to address the concerns while preserving the paper's focus on design provocations.

read point-by-point responses

  1. Referee: [Abstract] Abstract and the description of the ongoing empirical study: the claims that anonymous AI relaying 'significantly reduced psychological safety and satisfaction' and that autonomous counterarguments 'improved satisfaction and reduced marginalization' are presented without any reported sample size, randomization procedure, statistical tests, effect sizes, or control conditions. These details are load-bearing because the three provocations rest directly on the attribution of the observed differences to the specific AI mediation strategies rather than to unmeasured factors in the hierarchical group setting.

    Authors: We agree that the current presentation risks overstating preliminary observations as robust findings. In the revised manuscript, we will modify the abstract and study description to explicitly state that the reported patterns are initial observations from an ongoing pilot study, without claiming statistical significance. We will qualify the language (e.g., replacing 'significantly reduced' with 'appeared to reduce') and note that the provocations are derived from these early insights to stimulate discussion, with full empirical validation planned for future work. This revision ensures the claims are appropriately scoped to the available data. revision: yes

  2. Referee: [Empirical Study] The empirical study description: no baseline measures of group dynamics, order effects, or AI presence as a potential confound are mentioned, leaving open the possibility that the participation increase and safety decrease are driven by the power-imbalanced context itself rather than the anonymity versus authenticity manipulation. This directly weakens the central claim that the two strategies produce the reported divergent outcomes.

    Authors: This is a valid point regarding the level of detail provided. The study employs a within-subjects design with counterbalanced conditions and includes baseline assessments of group dynamics prior to AI intervention, as well as a control condition without AI mediation. However, these elements were omitted from the position paper to maintain brevity. We will revise the manuscript to include a more detailed description of the experimental protocol, including measures for order effects, baseline group dynamics, and controls for AI presence. This will strengthen the attribution to the specific strategies. Full results addressing these factors will be available upon study completion. revision: partial

standing simulated objections not resolved
  • Detailed statistical results including sample sizes, p-values, and effect sizes, since the empirical study is ongoing and data analysis is not yet complete.

Circularity Check

0 steps flagged

No circularity: empirical position paper with observational claims only

full rationale

The paper is a position piece that reports preliminary empirical observations from an ongoing study (increased participation from anonymous AI relaying, reduced safety/satisfaction, and benefits from autonomous counterarguments) and derives three design provocations from them. No equations, derivations, fitted parameters, or self-citation chains appear in the provided text or abstract. Claims rest on reported differences in participation, psychological safety, and satisfaction rather than reducing to inputs by construction. This is the expected non-finding for an empirical/design provocation paper without mathematical or self-referential logic.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The paper introduces no mathematical models, fitted parameters, formal axioms, or new postulated entities; it is a position paper grounded in preliminary observations.

pith-pipeline@v0.9.0 · 10458 in / 1021 out tokens · 92909 ms · 2026-05-08T10:48:04.735843+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

12 extracted references · 12 canonical work pages

  1. [1]

    Chun-Wei Chiang, Zhuoran Lu, Zhuoyan Li, and Ming Yin. 2024. Enhancing AI-Assisted Group Decision Making through LLM-Powered Devil’s Advocate. InProceedings of the 29th International Conference on Intelligent User Interfaces (IUI ’24). Association for Computing Machinery, New York, NY, USA, 103–119. doi:10.1145/3640543.3645199

    work page doi:10.1145/3640543.3645199 2024

  2. [2]

    Amy Edmondson. 1999. Psychological Safety and Learning Behavior in Work Teams.Administrative Science Quarterly44, 2 (June 1999), 350–383. doi:10.2307/ 2666999

    work page 1999

  3. [3]

    Angel Hsing-Chi Hwang and Andrea Stevenson Won. 2024. The Sound of Support: Gendered Voice Agent as Support to Minority Teammates in Gender-Imbalanced Team. InProceedings of the CHI Conference on Human Factors in Computing Systems (CHI ’24). Association for Computing Machinery, New York, NY, USA, 1–22. doi:10.1145/3613904.3642202

    work page doi:10.1145/3613904.3642202 2024

  4. [4]

    (Irving Lester) Janis

    Irving L. (Irving Lester) Janis. 1982.Groupthink : Psychological Studies of Policy Decisions and Fiascoes. Boston : Houghton Mifflin

    work page 1982

  5. [5]

    Herbert C. Kelman. 1958. Compliance, Identification, and Internalization Three Processes of Attitude Change.Journal of Conflict Resolution2, 1 (March 1958), 51–60. doi:10.1177/002200275800200106

    work page doi:10.1177/002200275800200106 1958

  6. [6]

    Soohwan Lee, Seoyeong Hwang, Dajung Kim, and Kyungho Lee. 2025. Conver- sational Agents as Catalysts for Critical Thinking: Challenging Social Influence in Group Decision-making. InProceedings of the Extended Abstracts of the CHI Conference on Human Factors in Computing Systems (CHI EA ’25). Association for Computing Machinery, New York, NY, USA, 1–12. do...

    work page doi:10.1145/3706599.3719792 2025

  7. [7]

    SooHwan Lee, Mingyu Kim, Seoyeong Hwang, Dajung Kim, and Kyungho Lee

    work page

  8. [8]

    InCompanion Proceedings of the 30th Interna- tional Conference on Intelligent User Interfaces (IUI ’25 Companion)

    Amplifying Minority Voices: AI-Mediated Devil’s Advocate System for Inclusive Group Decision-Making. InCompanion Proceedings of the 30th Interna- tional Conference on Intelligent User Interfaces (IUI ’25 Companion). Association for Computing Machinery, New York, NY, USA, 17–21. doi:10.1145/3708557.3716334

    work page doi:10.1145/3708557.3716334

  9. [9]

    Xiaoyan Li, Naomi Yamashita, Wen Duan, Yoshinari Shirai, and Susan R. Fussell

    work page

  10. [10]

    ACM Hum.-Comput

    Improving Non-Native Speakers’ Participation with an Automatic Agent in Multilingual Groups.Proc. ACM Hum.-Comput. Interact.7, GROUP (Dec. 2022), 12:1–12:28. doi:10.1145/3567562

    work page doi:10.1145/3567562 2022

  11. [11]

    Serge Moscovici and Elisabeth Lage. 1976. Studies in Social Influence III: Majority versus Minority Influence in a Group.European Journal of Social Psychology6, 2 (1976), 149–174. doi:10.1002/ejsp.2420060202

    work page doi:10.1002/ejsp.2420060202 1976

  12. [12]

    Dimitri Staufer, Frank Pallas, and Bettina Berendt. 2024. Silencing the Risk, Not the Whistle: A Semi-automated Text Sanitization Tool for Mitigating the Risk of Whistleblower Re-Identification. InProceedings of the 2024 ACM Conference on Fairness, Accountability, and Transparency (FAccT ’24). Association for Computing Machinery, New York, NY, USA, 733–74...

    work page doi:10.1145/3630106.3658936 2024