HABERMAS’S MACHINE: AI can reduce social conflicts
Jürgen Habermas is a German philosopher and social theorist working within the traditions of critical theory and pragmatism. Habermas’s theoretical framework explores the potential of reason, emancipation, and rational-critical communication inherent in contemporary institutions, as well as in the human capacity for reflection and the pursuit of rational interests / wikipedia.org
People sometimes find it very difficult to reach a compromise: everyone has different tastes, views on the future and the past, cultural preferences, and religious and political identities… However, it turns out artificial intelligence can help us even in this. A large language model can effectively serve as a mediator, assisting groups in achieving consensus.
GENERATOR OF SOCIAL CONSENSUS
Hardly had two programmers from Google DeepMind received the Nobel Prize in Chemistry for «predicting protein structures» when the company surprised the world once again. According to Science magazine, its specialists conducted an unusual experiment involving online discussion groups, where participants with diverse, sometimes even opposing views were brought together to tackle a complex problem.
Their task was to reach a consensus, formalized as a summary statement. Two types of mediators were involved in formulating compromise solutions: human moderators and AI. To the researchers’ astonishment, participants consistently preferred summaries crafted by artificial rather than human intelligence. But what does this mean in practice?
We already know that AI can replace humans in fields such as accounting and finance, advertising and marketing, design, and copywriting. Now, however, it appears capable of servicing the domain of «social consensus» — a critical foundation for effective collective action. Specifically, large language models (LLMs) can take over functions currently performed by citizen assemblies — first partially and possibly entirely in the future.
HABERMAS’S MACHINE: FROM THEORY TO PRACTICE
We know that diverse groups of people voice their opinions on public policy issues during civic assemblies. Politicians are expected to hear and consider these diverse perspectives. However, scaling such democratic initiatives is challenging, so these discussions are typically limited to small groups.
Christopher Summerfield, Research Director at the UK AI Safety Institute, once asked himself: could large language models (LLMs) assist in facilitating these discussions? To explore this, he employed a modified version of the pre-trained DeepMind LLM Chinchilla.
This model received a symbolic name — the «Habermas Machine». This choice wasn’t accidental: at 95, Jürgen Habermas is arguably the most influential living political and social philosopher. He developed theories of communicative action and discourse ethics, which outline ways to rationally resolve conflicts. Now, his philosophical ideas can be practically implemented — with the help of the Habermas Machine!
AI — THE BEST «MEDIATOR»
To test the model, researchers recruited a total of 439 British participants, dividing them into small groups. Each group was asked to discuss three questions related to UK government policy.
During the discussion, group members shared their individual opinions on each question, with participants allowed to critique and rank each other’s views. Afterward, the group’s consensus was formulated by two mediators: an artificial intelligence and a human moderator. Participants then assessed both versions of the final summaries, rating them on how well their views were represented.
The majority (56%) preferred summaries created by AI, considering them of higher quality, while 44% favored those prepared by human mediators. In addition to the participants, external reviewers were asked to evaluate the two summaries. They also gave the AI summaries higher marks for honesty, quality and clarity.
NEW OPPORTUNITIES, NEW RISKS
In the next stage of the study, another group was formed that closely matched the socio-demographic structure of the UK population. A virtual citizen assembly was conducted within this group, naturally leading to some controversial discussions. Nevertheless, following interaction with AI, the assembly was able to achieve consensus.
Researchers concluded that artificial intelligence could substantially contribute to enhancing deliberative democracy. However, they also highlighted the limitations associated with such technologies, noting that AI may neglect people’s intrinsic need for face-to-face communication, potentially narrowing certain types of communication spaces.
Christopher Summerfield emphasized that while AI could make a significant positive contribution to deliberative democracy, these technologies also present limitations — particularly in their vulnerability to including biases and overlooking the fundamental human need for genuine interpersonal dialogue.
Original Research:
