Uploaded January 2025 | Updated September 2026, 3 weeks ago
How can #AI help bridge differing opinions in discussions? Finding common ground in group discussions can be slow and challenging. In this talk Michael Henry Tessler & Michiel Bakker from Google DeepMind share how they trained an AI to mediate human deliberation by generating and refining group statements based on participants' opinions and critiques.
Their study of 5,734 participants showed that AI-generated statements were rated as more informative, clear, and less biased than those from human mediators. Participants often adjusted their views after the deliberation, moving towards shared perspectives that respected both dissenting voices and majority opinions.
These findings were replicated in a virtual citizens’ assembly involving a demographically representative sample of the UK population.
Timestamps:
0:00 Introduction
2:22 LLMs for finding common ground
3:27 Habermas machine: a simulated election
8:40 Example deliberation round
10:25 Does AI-mediated deliberation help people find common ground? Human mediator comparison, impact of iterative critiquing
15:54 Does AI-mediated deliberation leave groups less divided? Analyzing belief change, comparison to unmediated exchange of opinions
18:30 Exploratory analysis: belief change under Human Mediator vs. Habermas Machine
19:35 Does the AI mediator represent all viewpoints equally? Embedding geometry
25:10 Can AI mediation support deliberation in a citizens' assembly?
28:18 Summary
#artificialintelligence #aimodels #machinelearning #aimediator #reinforcementlearning #llm #llms #airesearch #aichallenges #aiprogress #robot #robotics #artificialsuperintelligence #artificialgeneralintelligence #tech #techtalk #techtalks #aitalks #aitalk #programming #googledeepmind #googleai #aiethics #habermasmachine
Social Links:
Newsletter: buzzrobot.substack.com
X: https://x.com/sopharicks
Slack: join.slack.com/t/buzzrobot/shared_invite/zt-2s067rv7n-guPIMGe62rbp9ncxdnOUfQ
How can #AI help bridge differing opinions in discussions? Finding common ground in group discussions can be slow and challenging. In this talk Michael Henry Tessler & Michiel Bakker from Google DeepMind share how they trained an AI to mediate human deliberation by generating and refining group statements based on participants' opinions and critiques.
Their study of 5,734 participants showed that AI-generated statements were rated as more informative, clear, and less biased than those from human mediators. Participants often adjusted their views after the deliberation, moving towards shared perspectives that respected both dissenting voices and majority opinions.
These findings were replicated in a virtual citizens’ assembly involving a demographically representative sample of the UK population.
Timestamps:
0:00 Introduction
2:22 LLMs for finding common ground
3:27 Habermas machine: a simulated election
8:40 Example deliberation round
10:25 Does AI-mediated deliberation help people find common ground? Human mediator comparison, impact of iterative critiquing
15:54 Does AI-mediated deliberation leave groups less divided? Analyzing belief change, comparison to unmediated exchange of opinions
18:30 Exploratory analysis: belief change under Human Mediator vs. Habermas Machine
19:35 Does the AI mediator represent all viewpoints equally? Embedding geometry
25:10 Can AI mediation support deliberation in a citizens' assembly?
28:18 Summary
#artificialintelligence #aimodels #machinelearning #aimediator #reinforcementlearning #llm #llms #airesearch #aichallenges #aiprogress #robot #robotics #artificialsuperintelligence #artificialgeneralintelligence #tech #techtalk #techtalks #aitalks #aitalk #programming #googledeepmind #googleai #aiethics #habermasmachine
Social Links:
Newsletter: buzzrobot.substack.com
X: https://x.com/sopharicks
Slack: join.slack.com/t/buzzrobot/shared_invite/zt-2s067rv7n-guPIMGe62rbp9ncxdnOUfQ
