Reaching an agreement in debates, especially in democratic settings, could be difficult due to attendees' different ideological, political, and social views. However, recent research reveals that Google DeepMind has trained a system of LLMs to operate as a mediator, generating summaries outlining a group of areas of agreement on complex but essential social or political issues.
This breakthrough may revolutionize how we handle contentious issues, especially in group settings where finding common ground is essential.
The Habermas Machine: A Tool for Mediation
Named after the philosopher Jürgeb Habermas, Google DeepMind's AI tool, the "Habermas Machine"(HM), was designed to mediate group discussions by summarizing opposing viewpoints and highlighting areas of agreement. Unlike traditional AI models focusing on persuasion, HM has been developed to act as an unbiased mediator. It analyses group opinions and suggests all statements representing collective views.
More than 5000 people were used to test the tool's effectiveness. Groups were asked whether the National Health Service (NHS) should be privatized or if the voting age should be lowered to 16. The AI then summarises the group's opinions, which participants critique. When the participants were asked to rank the final AI-generated statements, more than half preferred these over those made by human mediators. This demonstrates how AI can mediate arguments more effectively than people, leading to better, more palatable results.
AI’s Role in Facilitating Agreement
AI's ability to combine various viewpoints and generate a collective summary highlights its function as a debate mediator. According to Christopher Summerfield, research director at the UK AI Safety Institute, AI can function similarly to citizen assemblies and deliberative posts aiming to provide a platform for divorce perspectives. With AI, these procedures can be expanded without compromising the standard of discussion or the inclusion of all voices.
The HM uses two large language models: one for generating statements reflecting the group's varied views and the other one to score the statements based on how appealing they are to participants. This approach has been shown to increase the possibility of collective action by assisting participants with different opinions in reaching an agreement. By encouraging agreement, AI can make debates more efficient and productive, especially in settings where finding common ground is essential for policymaking.
Potential and Limitations of AI Mediation
While AI has much potential to mediate debates, some limitations remain. The HM lacks certain features required for real-world mediation, like fact-checking and preserving topic significance, as noted by Google DeepMind research scientist Michael Henry Tessler. All the shortcomings highlight the requirement for careful implementation and additional research before the widespread adoption of AI mediation tools.
Additionally, concerns have been raised about AI's impact on the human aspect of deliberation. Many critics, including Sammy McKinney from the University of Cambridge, worry that AI may remove the essential human element from discussions. Even if AI is efficient, human engagement during arguments is vital as it promotes empathy and understanding.