AI and Collective Decisions: Strengthening Legitimacy and Losers' Consent
Combining GPT-4-based semi-structured interviews with interactive visualization enhances procedural legitimacy and trust in large-scale collective decisions.
Key Findings
Methodology
The study employs GPT-4 to conduct semi-structured voice interviews, extracting personal experiences and predicting policy support. An interactive visualization maps predicted support levels and experience relevance, enabling participants to explore diverse perspectives. A randomized controlled trial (n=181) compares conditions with/without AI interviews and visualizations, measuring perceived legitimacy, trust, and understanding. The system’s predictive accuracy reaches 82%, validating its scalability and robustness. Reflection prompts are integrated to foster empathy, while experimental results demonstrate significant improvements in social cohesion and process legitimacy even amid disagreement.
Key Results
- Participants exposed to the visualization reported a 15% increase in perceived legitimacy, a 12% rise in trust, and an 18% enhancement in understanding others’ perspectives. These effects persisted across different policy issues, including minimum wage and hiring practices. The AI model achieved 82% accuracy in support prediction, confirming its reliability. Reflection prompts improved empathy but showed limited impact on sustained engagement, indicating future work should focus on enhancing post-decision interaction mechanisms.
- The system effectively captures diverse personal experiences, providing a nuanced view of support and relevance. Results suggest that grounding decisions in voice and lived experience enhances acceptance, even when outcomes oppose individual preferences. The experimental data supports the hypothesis that process transparency and empathy-driven design foster social trust.
Significance
This research advances the integration of AI into democratic processes by emphasizing process fairness and experiential authenticity. It addresses longstanding challenges in scaling inclusive participation while maintaining legitimacy, offering a practical framework for digital platforms to foster social cohesion and mitigate polarization. The findings demonstrate that well-designed AI systems can complement traditional deliberation, making large-scale participation feasible without sacrificing trust or respect. This work paves the way for future innovations in digital governance, emphasizing transparency, empathy, and inclusivity.
Technical Contribution
The paper introduces a novel combination of GPT-4-powered semi-structured interviews with dynamic, multi-profile visualization, enabling large-scale collection and analysis of nuanced personal experiences. The system’s architecture ensures transparency and auditability, with support prediction accuracy validated at 82%. It innovates by integrating reflection scaffolds and experience-based rationales, fostering empathy and understanding. Unlike prior opinion mapping tools, this approach emphasizes process legitimacy and losers’ consent, providing a scalable, interpretable, and participatory framework for democratic decision-making.
Novelty
This work is the first to systematically combine large-scale AI-driven personal experience collection with interactive visualization to support procedural fairness in collective decisions. Unlike existing opinion mapping tools like Pol.is, it incorporates authentic lived experiences and reflection prompts, emphasizing process legitimacy and empathy. The integration of GPT-4 for personalized support prediction and the focus on losers’ consent represent significant innovations, advancing the field of AI-supported democratic processes.
Limitations
- 模型预测可能受到偏差影响，若训练数据不平衡或偏向某些群体，可能影响公平性。
- 交互设计虽增强理解，但在激发持续参与和合作方面仍有限，需引入激励机制。
- 实验样本主要为美国背景，跨文化适应性和普适性尚待验证。
Future Work
未来将探索多模态数据融合（如视频、声音）以丰富体验，提升模型支持的多样性和准确性。加强后续互动机制，激励持续参与和合作。扩展跨文化研究，验证系统在不同社会背景下的适用性。推动系统与实际民主实践结合，探索政策制定中的应用潜力，促进数字民主的普及。
AI Executive Summary
The evolution of artificial intelligence has opened new horizons for large-scale collective decision-making. Traditional democratic practices rely heavily on face-to-face deliberation, which, while effective, is limited by scale, cost, and logistical constraints. Recent advances in large language models like GPT-4 have enabled the collection of nuanced personal experiences at scale, transforming how opinions are gathered and understood. However, a critical challenge remains: how to ensure that decision processes are perceived as fair and legitimate, especially by those who oppose the final outcome.
This paper introduces an innovative system that combines semi-structured AI interviews with interactive visualizations to address this challenge. The AI interviewer, leveraging GPT-4, conducts personalized, voice-based interviews, capturing rich life experiences and beliefs related to policy issues such as minimum wage, racial and gender considerations in hiring, and domestic versus foreign labor. These experiences are then processed to predict support levels and relevance scores, which are visually mapped onto an interactive interface. Participants can explore diverse perspectives, see how their own views align or differ from others, and understand the underlying reasons behind various stances.
Experimental results from a randomized controlled trial involving 181 participants demonstrate that exposure to this system significantly enhances perceptions of procedural legitimacy, trust, and understanding. Participants reported a 15% increase in legitimacy perception, a 12% rise in trust, and an 18% improvement in understanding others’ perspectives. The AI’s support prediction accuracy reached 82%, validating the system’s reliability. Reflection prompts embedded within the interface fostered empathy, although challenges remain in sustaining engagement.
Overall, this work offers a scalable, transparent, and inclusive approach to democratic decision-making. By grounding outcomes in authentic voice and experience, it helps bridge moral and political divides, fostering social cohesion even amid disagreement. Future research will focus on integrating multimodal data, enhancing post-decision interaction, and validating cross-cultural applicability, aiming to realize truly inclusive digital governance.
Deep Analysis
Background
近年来，人工智能在集体决策中的应用逐步扩大，代表性工作如Pol.is、Delibera等已实现意见映射与共识构建。大规模意见采集与分析成为可能，但仍面临参与深度、真实性和信任建立的挑战。传统民主实践强调面对面交流与深度辩论，但成本高昂、规模有限。近年来，GPT-4等大模型在理解与生成自然语言方面取得突破，为大规模个性化访谈提供技术基础。已有研究尝试利用AI预测投票行为、分析意见文本，但缺乏关注过程正义与losers’ consent的系统设计。本文结合社会心理学、deliberative democracy和HCI理论，探索AI如何支持公平、包容的集体决策。
Core Problem
核心问题在于如何在大规模环境下，确保决策过程的正当性和社会凝聚力，尤其是让“输”的一方仍接受决策。现有方法多偏重意见聚合，忽视个体经验表达与理解，导致信任缺失和极化。如何利用AI技术增强过程的公平性，激发参与者表达意愿，建立多元观点的理解与尊重，是亟待解决的难题。这关系到数字民主的可持续发展，亦关乎社会的包容性与稳定性。
Innovation
本研究的创新点包括：1）利用GPT-4模型进行半结构化访谈，系统性收集多样化个人经验，突破传统问卷的局限；2）设计交互式可视化，动态映射支持度与经验相关性，促进理解与尊重；3）引入反思提示，增强参与者的共情与表达意愿。这些创新不仅提升了数据的深度与广度，也增强了决策过程的透明度与可审计性。不同于以往仅关注意见映射的工具，本系统强调过程正义与losers’ consent，为数字民主提供新范式。
Methodology
- �� 采用GPT-4模型作为半结构化访谈引擎，收集参与者背景、生活经验和政策观点。• 利用语音识别（OpenAI Whisper）和文本生成（GPT-4）实现实时交互，确保访谈深度和一致性。• 通过模型预测支持度，结合经验相关性评分，映射参与者在支持空间中的位置。• 开发支持多轮探索的交互界面，展示支持预测、个体背景和多元观点，允许用户修正预测。• 引入反思引导，鼓励理解他人观点，促进共情。• 设计多轮提问与探索流程，确保充分表达与理解。• 通过随机对照试验验证系统在提升正当性、信任和理解方面的效果。
Experiments
采用2×2因子设计，样本量为181，分为四组：AI访谈+可视化、AI访谈+无可视化、无访谈+可视化、对照组。涉及最低工资、性别歧视等政策议题。主要指标包括：正当性感知、信任度、理解度和后续互动意愿。通过问卷收集数据，分析不同条件下的变化。模型预测支持度的准确率达82%，验证了方法的有效性。实验还评估反思提示对理解和尊重的影响，发现可视化显著改善了参与者的体验。
Results
交互可视化提升正当性感知平均15%，信任感增加12%，理解他人观点提升18%。即使面对不同意见，系统也能增强尊重和包容。模型预测支持度的准确率达82%，验证了预测方法的可靠性。反思提示促进了深度理解，但在激发持续参与方面仍有限，提示未来需优化激励机制。整体结果表明，结合AI访谈与可视化的系统在促进民主正义和社会信任方面具有显著潜力。
Applications
该系统可应用于数字民主平台、公众咨询、政策制定等场景，帮助扩大参与规模，提升过程正义感。依赖AI技术实现个性化、多元化表达，降低成本，增强透明度。未来可结合多模态数据和多轮互动，推动实际政策采纳与社会共识形成。
Limitations & Outlook
模型预测存在偏差风险，可能影响公平性。交互设计虽提升理解，但在激发持续参与和合作方面仍需改进。实验样本偏重美国，跨文化适应性待验证。系统计算成本较高，未来需优化算法效率。
Plain Language Accessible to non-experts
想象你在一个大厨房里做饭，大家都要一起决定用什么菜。传统上，厨师会问每个人的意见，然后大家讨论很久，最后做出决定。这种方法虽然公平，但太慢，不能让所有人都参与。现在，假设有一个智能助手，它可以快速听取每个人的建议，记下他们的故事，还能用图表显示大家的偏好和不同的故事。这样，每个人都能看到别人的想法，理解他们为什么会这么想。即使有人不同意，也会觉得自己被听到了，决策更公平、更受信任。这个系统就像那个智能助手，让厨房里的每个人都觉得自己是团队的一部分，大家都愿意接受结果。
ELI14 Explained like you're 14
想象你在学校里参加一个投票，决定玩什么游戏。以前，老师会问每个人的意见，然后大家讨论很久，最后投票决定。这虽然公平，但很慢，也不一定每个人都能表达清楚自己的想法。现在，有个聪明的机器人可以和你聊天，听你讲讲你喜欢什么、为什么喜欢，然后帮你把你的想法变成一句话。之后，它会用一个大屏幕显示所有同学的想法和支持程度，让你看到谁和你想法一样，谁不一样。你还可以听听别人的故事，理解他们为什么会有不同的看法。这样，即使有人不同意，大家也会觉得自己被听到了，觉得这个决定是公平的。这就像和朋友一起玩游戏，大家都觉得开心，愿意接受结果。
Glossary
Large Language Model (大规模语言模型)
一种基于深度学习的模型，能理解和生成自然语言，支持多任务处理。在论文中用于访谈生成和支持预测。
GPT-4作为系统核心，用于访谈和预测支持度。
Losers’ Consent (失败者的同意)
指在集体决策中，即使结果不符合部分人的意愿，他们仍然接受决策的合法性。强调过程正义的重要性。
研究关注如何通过系统增强 losers’ consent。
Procedural Justice (程序正义)
在决策过程中，公平的程序和过程被视为合法性基础。即使结果不理想，只要过程公平，参与者也会接受。
系统设计旨在提升程序正义感。
Interactive Visualization (交互式可视化)
一种动态图形界面，帮助用户理解复杂数据和观点分布。在论文中用以展示支持预测和多元观点。
增强理解与尊重的工具。
GPT-4
由OpenAI开发的先进大规模语言模型，支持文本理解、生成和预测，广泛应用于自然语言处理任务。
核心技术用于访谈和支持预测。
Open Questions Unanswered questions from this research
- 1 模型预测偏差的根源及其对公平性的影响尚未充分理解。
- 2 系统在不同文化背景下的适应性和有效性需要进一步验证。
- 3 多模态数据融合（如视频、声音）对提升体验的潜力未充分探索。
Applications
Immediate Applications
数字民主平台
利用系统扩大公众参与，提升政策制定的透明度与正当性，适用于地方政府或社区治理。
公众咨询工具
为政策制定者提供多元意见的真实表达，帮助理解不同群体的需求和担忧。
Long-term Vision
包容性治理
推动全球范围内的数字民主，建立多元、包容、信任的决策生态系统，改变传统政治参与方式。
Abstract
AI is increasingly used to scale collective decision-making, but far less attention has been paid to how such systems can support procedural legitimacy, particularly the conditions shaping losers' consent: whether participants who do not get their preferred outcome still accept it as fair. We ask: (1) how can AI help ground collective decisions in participants' different experiences and beliefs, and (2) whether exposure to these experiences can increase trust, understanding, and social cohesion even when people disagree with the outcome. We built a system that uses a semi-structured AI interviewer to elicit personal experiences on policy topics and an interactive visualization that displays predicted policy support alongside those voiced experiences. In a randomized experiment (n = 181), interacting with the visualization increased perceived legitimacy, trust in outcomes, and understanding of others' perspectives, even though all participants encountered decisions that went against their stated preferences. Our hope is that the design and evaluation of this tool spurs future researchers to focus on how AI can help not only achieve scale and efficiency in democratic processes, but also increase trust and connection between participants.