The Habermas Machine — Google DeepMind's AI Mediator for Citizens'-Assembly-Style Deliberation

Google DeepMind, working with the University of Oxford and the citizen-assembly convenor the Sortition Foundation, built the 'Habermas Machine': a large-language-model system trained to mediate small-group political deliberation by drafting group statements that synthesise participants' individual opinions and reasons.

In a study published in Science (October 2024), 5,734 UK participants were organised into small groups (five to six people) to discuss divisive issues including Brexit, immigration, the minimum wage, climate change and universal childcare, comparing AI-drafted group statements against those written by human mediators.

The AI-drafted group statements were preferred over statements written by human mediators 56% of the time, and were rated higher by independent judges for quality, clarity, informativeness and perceived fairness. Groups that deliberated with the AI mediator also converged toward more common positions — an effect that did not occur when the same groups exchanged views directly without mediation.

The system currently cannot fact-check claims, keep discussion on topic, or moderate the conversation itself, and ethicists who reviewed the study (e.g. Joongi Shin) raised concerns that participants were not explicitly told an AI system was generating the statements presented to them (consent forms mentioned 'algorithms' generally). DeepMind has stated it has no plans to release the system publicly, so government adoption beyond further academic and pilot studies remains untested.