Have you ever felt like you are talking to a brick wall during a discussion, unable to get the recipient to listen to your opinion? or found it difficult to voice your opinion in group debates, feeling overpowered by more confident, stubborn group members? While the idea of trying to find common ground is certainly appealing to me, dissenting views often cause conversations to boil over, descending into chaos, leaving us no further forward than before. But would you consider allowing for AI mediation to help reach a consensus during a debate?
Michael Henry Tessler and his fellow researchers did just that, training a large language model named the “Habermas Machine” to serve as a artificial intelligence (AI) mediator to assist small groups in the UK in identifying points of agreement while talking about contentious political topics like immigration, Brexit, and climate change. Their study explores how artificial intelligence (AI) might be able to get round differences of opinion by assisting people in finding common ground on difficult social and political issues using AI mediation.
The researchers posed the question of whether a large language model (LLM)-based AI system could effectively represent the underlying common viewpoints of group members by writing a “group statement” that the discussants would all agree with. They ran a number of tests with more than 5000 people from the UK to assess the effectiveness of AI-mediated deliberation. These tests examined how AI mediation affected the ability to identify common ground, how discussants’ opinions evolved during the process, how group statements balanced minority and majority viewpoints, and whether there were any biases in those statements.
Participants were asked to answer questions about everything from the use of monkeys in medical research to the use of religion in public education in each trial.
What were the results?
According to the study, group opinion statements produced by the Habermas Machine were consistently more favoured by group members compared to those created by human mediators. Additionally, external judges rated the AI-generated statements higher in terms of quality, clarity, informativeness, and perceived fairness. The use of AI-mediated deliberation also helped decrease differences within groups, as participants tended to align their views towards a shared position after discussing the issue. This outcome was not observed when individuals exchanged perspectives without any mediation.
The Habermas Machine itself, however, has sparked controversy due to some researchers pointing out that it does not effectively translate democratic discussions into policy outcomes. Dr. Melanie Garson, a conflict resolution specialist at UCL, expressed her concerns. She highlighted the potential issue where certain minority groups might not have enough representation to impact collective decisions, but could still face significant consequences from the outcome. She also pointed out that the Habermas Machine does not allow participants the opportunity to express their emotions, thus hindering the development of empathy towards those with differing opinions.
While the use of AI as a mediator in situations where we are simply going round in circles with no hope of reaching a consensus might offer a means of speeding up decision making, bringing opinions on different issues closer together, I am still a bit unsure about using it. Although some of the challenges mentioned at the start will always pose potential roadblocks, I still feel there is more to be gained from being able to express your opinions and feelings openly, aiming to develop more empathetic, human-led approaches to conflict resolution.
What do you guys think, do you have any other effective techniques for approaching conflict and disagreement? Why not share an example of an issue that has caused conflict and how you resolved it below? I would love to hear your thoughts.