Research Article
ARTIFICIAL INTELLIGENCE
Share on
AI can help humans find common ground in democratic deliberation
Michael Henry Tessler https://orcid.org/0000-0003-3945-0239, Michiel A. Bakker https://orcid.org/0000-0003-4474-7109, Daniel Jarrett, Hannah Sheahan https://orcid.org/0000-0003-0443-9400, Martin J. Chadwick, Raphael Koster https://orcid.org/0000-0001-5264-3137, Georgina Evans, Lucy Campbell-Gillingham https://orcid.org/0009-0000-5476-5402, Tantum Collins https://orcid.org/0009-0004-8461-6028, David C. Parkes https://orcid.org/0000-0002-2701-3464, Matthew Botvinick https://orcid.org/0000-0001-7758-6896, and Christopher Summerfield https://orcid.org/0000-0002-2941-2653 fewerAuthors Info & Affiliations
Science
18 Oct 2024
Vol 386, Issue 6719
DOI: 10.1126/science.adq2852
47 141
81
Metrics
Total Downloads47 141

Last 6 Months6 196
Last 12 Months13 515

Total Citations81

Last 6 Months50
Last 12 Months73

Editor’s summary
To act collectively, groups must reach agreement; however, this can be challenging when discussants present very different but valid opinions. Tessler et al. investigated whether artificial intelligence (AI) can help groups reach a consensus during democratic debate (see the Policy Forum by Nyhan and Titiunik). The authors trained a large language model called the Habermas Machine to serve as an AI mediator that helped small UK groups find common ground while discussing divisive political issues such as Brexit, immigration, the minimum wage, climate change, and universal childcare. Compared with human mediators, AI mediators produced more palatable statements that generated wide agreement and left groups less divided. The AI’s statements were more clear, logical, and informative without alienating minority perspectives. This work carries policy implications for AI’s potential to unify deeply divided groups. —Ekeoma Uzogara

Structured Abstract
INTRODUCTION
Democracy, at its best, rests upon the free and equal exchange of views among people with diverse perspectives. Collective deliberation can be effectively supported by structured events, such as citizens’ assemblies, but such events are expensive, are difficult to scale, and can result in voices being heard unequally. This study investigates the potential of artificial intelligence (AI) to overcome these limitations, using AI mediation to help people find common ground on complex social and political issues.
RATIONALE
We asked whether an AI system based on large language models (LLMs) could successfully capture the underlying shared perspectives of a group of human discussants by writing a “group statement” that the discussants would collectively endorse. Inspired by Jürgen Habermas’s theory of communicative action, we designed the “Habermas Machine” to iteratively generate group statements that were based on the personal opinions and critiques from individual users, with the goal of maximizing group approval ratings. Through successive rounds of human data collection, we used supervised fine-tuning and reward modeling to progressively enhance the Habermas Machine’s ability to capture shared perspectives. To evaluate the efficacy of AI-mediated deliberation, we conducted a series of experiments with over 5000 participants from the United Kingdom. These experiments investigated the impact of AI mediation on finding common ground, how the views of discussants changed across the process, the balance between minority and majority perspectives in group statements, and potential biases present in those statements. Lastly, we used the Habermas Machine for a virtual citizens’ assembly, assessing its ability to support deliberation on controversial issues within a demographically representative sample of UK residents.
RESULTS
Group opinion statements generated by the Habermas Machine were consistently preferred by group members over those written by human mediators and received higher ratings from external judges for quality, clarity, informativeness, and perceived fairness. AI-mediated deliberation also reduced division within groups, with participants’ reported stances converging toward a common position on the issue after deliberation; this result did not occur when discussants directly exchanged views, unmediated. Although support for the majority position increased after deliberation, the Habermas Machine demonstrably incorporated minority critiques into revised statements. We replicated these results in a virtual citizens’ assembly, additionally finding that during AI-mediated deliberation, the views of groups of discussants tended to move in a similar direction on controversial issues. These shifts were not attributable to biases in the AI, suggesting that the deliberation process genuinely aided the emergence of shared perspectives on potentially polarizing social and political issues.
CONCLUSION
This research demonstrates the potential of AI to enhance collective deliberation by finding common ground among discussants with diverse views. The AI-mediated approach is time-efficient, fair, scalable, and outperforms human mediators on key dimensions. Rather than simply appealing to the majority, the Habermas Machine prominently incorporated dissenting voices into the group statements. AI-assisted deliberation is not without its risks, however; to ensure fair and inclusive debate, steps must be taken to ensure users are representative of the target population and are prepared to contribute in good faith. Under such conditions, AI may be leveraged to improve collective decision-making across various domains, from contract negotiations and conflict resolution to political discussions and citizens’ assemblies. The Habermas Machine offers a promising tool for finding agreement and promoting collective action in an increasingly divided world.
AI helps people find common ground in collective deliberation.
(Left) The AI mediator uses participants’ opinions to generate group statements and iteratively refines those statements through participants’ critiques. (Middle) Statements from the AI mediator (purple) garner stronger endorsement than those written by a human mediator (orange). (Right) AI mediation leaves groups less divided after deliberation, whereas simply sharing opinions with others does not.
Abstract
Finding agreement through a free exchange of views is often difficult. Collective deliberation can be slow, difficult to scale, and unequally attentive to different voices. In this study, we trained an artificial intelligence (AI) to mediate human deliberation. Using participants’ personal opinions and critiques, the AI mediator iteratively generates and refines statements that express common ground among the group on social or political issues. Participants (N = 5734) preferred AI-generated statements to those written by human mediators, rating them as more informative, clear, and unbiased. Discussants often updated their views after the deliberation, converging on a shared perspective. Text embeddings revealed that successful group statements incorporated dissenting voices while respecting the majority position. These findings were replicated in a virtual citizens’ assembly involving a demographically representative sample of the UK population.

Affiliations
TOUS de Google DeepMind, London, UK. !!

• Michael Henry Tessler*,† https://orcid.org/0000-0003-3945-0239 mhtessler@google.com
  Roles: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Project administration, Resources, Software, Supervision, Validation, Visualization, Writing - original draft, and Writing - review & editing.
• Michiel A. Bakker*,† https://orcid.org/0000-0003-4474-7109 mhtessler@google.com
  Roles: Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Software, Supervision, Validation, Visualization, Writing - original draft, and Writing - review & editing.
• Daniel Jarrett
  Roles: Conceptualization, Data curation, Formal analysis, Methodology, Software, Validation, Visualization, Writing - original draft, and Writing - review & editing.
• Hannah Sheahan https://orcid.org/0000-0003-0443-9400
  Roles: Conceptualization, Data curation, Methodology, and Software.
• Martin J. Chadwick
  Roles: Conceptualization, Methodology, Resources, Software, and Writing - review & editing.
• Raphael Koster https://orcid.org/0000-0001-5264-3137
  Roles: Investigation and Software.
• Georgina Evans
  Roles: Formal analysis, Methodology, Software, Validation, Visualization, and Writing - review & editing.
• Lucy Campbell-Gillingham https://orcid.org/0009-0000-5476-5402
  Roles: Funding acquisition, Investigation, and Project administration.
• Tantum Collins https://orcid.org/0009-0004-8461-6028
  Roles: Conceptualization, Methodology, and Writing - review & editing.
• David C. Parkes https://orcid.org/0000-0002-2701-3464
  Department of Computer Science, Harvard University, Cambridge, MA, USA.
  Roles: Conceptualization and Methodology.
• Matthew Botvinick* https://orcid.org/0000-0001-7758-6896 mhtessler@google.com
  Yale Law School, New Haven, CT, USA.
  Roles: Conceptualization, Methodology, Project administration, Supervision, Visualization, Writing - original draft, and Writing - review & editing.
• Christopher Summerfield* https://orcid.org/0000-0002-2941-2653 mhtessler@google.com
  Department of Experimental Psychology, University of Oxford, Oxford, UK.
  Roles: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Project administration, Resources, Software, Supervision, Validation, Visualization, Writing - original draft, and Writing - review & editing.

Notes
Corresponding author. Email: mhtessler@google.com (M.H.T.); miba@google.com (M.A.B.); botvinick@google.com (M.B.); christopher.summerfield@psy.ox.ac.uk (C.S.)
† These authors contributed equally to this work.
Connu / https://tools.immae.eu/Shaarli/mediationpourtous?anEhcQ
Ndlr :

• Première observation : tous les auteurs sont de GOOGLE ! N'et-ce pas un ÉNORME BIAIS ?!
• APPROFONDIR ACT

Billet de blog 17 février 2025 / Stephanie Lamy, Féministe, chargée d'enseignement, chercheuse

L’attentat déjoué à Annecy a reçu peu de couverture médiatique, une discrétion qui ne surprend plus lorsqu’il s’agit de violences perpétrées contre les femmes et les minorités de genre. Pourtant, cet événement semble marquer un précédent historique : un passage à l’acte inspiré par la radicalisation masculiniste est qualifié sous des chefs d’accusation relevant du terrorisme.
...
Connu / mel
"
Sujet : Pour une mémoire inclusive : rendre hommage aux victimes de tous les terrorismes
Date : Tue, 25 Mar 2025 10:32:42 +0000
De : Entre les lignes entre les mots
"

minoritaire & populaire
entretien avec Annie Collovald & Olivier Schwartz

« Populaire » : ce qui en politique est un espoir (le renversement de l’ordre injuste, de la domination), la sociologie semble le défaire (le peuple est trop démuni pour rompre la domination). Comprendre les batailles politiques qui se forment aujourd’hui (la consécration des « suffrages populaires » en Saint Graal de la compétition politique et, dans le même temps, la conjuration sans cesse répétée du « populisme ») oblige alors à un exercice très exact de repérage de la stratification sociale actuelle, et des forces politiques qu’elle libère.

Annie Collovald est professeure de sociologie politique à l’Université de Nantes. Olivier Schwartz est professeur de sociologie à l’Université de Paris V.
...
Connu / TG le 23/04/23 à 19:01

Connu / https://tools.immae.eu/Shaarli/ind1ju?1lMvQA
...
un ensemble de pratiques, ou plutôt un ensemble d’attentes et de demandes concernant les logiciels de médias sociaux, dans lesquels les efforts disparates de projets de médias sociaux alternatifs convergent en un réseau partagé avec des objectifs plus ou moins similaires. Les différents modèles d’utilisation, partagés entre les serveurs, vont de plateformes d’extrême-droite financées par le capital-risque à des systèmes de publication d’images japonaises, en passant par des collectifs anarcho-communistes, des groupes politiques, des amateurs d’algorithmes de codage en direct, des « espaces sûrs » pour les travailleur⋅euse⋅s du sexe, des forums de jardinage, des blogs personnels et des coopératives d’auto-hébergement. Ces pratiques se forment parallèlement au problème du partage des données et du travail gratuit, et font partie de la transformation
...
contribuer à façonner la culture des plateformes par leurs échanges et le contenu partagés. Twitter en est un exemple bien connu, où des fonctionnalités essentielles, telles que les noms d’utilisateur préfixés par @ et les hashtags par #, ont d’abord été suggérés par les utilisateurs. Les forums comme Reddit permettent également aux utilisateurs de définir et modérer des pages, créant ainsi des communautés distinctes et spécifiques.

Sur les plateformes alternatives de médias sociaux comme le Fédiverse, en particulier à ses débuts, ces formes de participation vont plus loin continuelle de ce que cela implique d’être utilisateur ou utilisatrice de logiciel.
...
La manière dont les facilitateurs de services sont soutenus sur le Fédiverse, par exemple, est analogue à la manière dont les créateurs de contenu sur les plateformes de streaming au sein des communautés de jeux sont soutenus par leur public. Des appels à une meilleure gouvernance des projets logiciels sont également en cours dans les communautés FLOSS plus largement. L’élaboration de codes de conduite (un document clé pour les instances de la Fédiverse mis en place pour exposer leur vision de la communauté et de la politique) a été introduit dans diverses communautés FLOSS au début des années 2010, en réponse à la misogynie systématique et à la l’exclusion des minorités des espaces FLOSS à la fois en ligne et hors ligne29. Les codes de conduite répondent également au besoin de formes génératrices de résolution des conflits par-delà les barrières culturelles et linguistiques.
...
La nature évolutive de l’utilisation du Fédiverse montre à quel point l’écart est grand entre les extrêmes stéréotypés du modèle capitaliste de surveillance et du martyre auto-infligé des plateformes de bénévolat. Ce qui a des implications sur le rôle des utilisateurs en relation avec les médias sociaux alternatifs, ainsi que pour le développement de la culture des FLOSS30.