The Habermas machine generates high-quality group opinion statements that are preferred over group statements written by humans, and criticism makes further improvements. Credit: Science (2024). DOI: 10.1126/science.adq2852
A team of AI researchers from Google’s DeepMind London group has discovered that certain large language models (LLMs) can serve as effective mediators between groups of people with divergent views on a given topic. The work is published in the journal Science.
In recent decades, political divisions have become commonplace in many countries; most have been characterized as liberal or conservative. The advent of the Internet served as an engine, allowing citizens on both sides to promote their opinions to a wide audience, generating anger and frustration. Unfortunately, no tools have surfaced to ease the tension of such a political climate. In this new effort, the DeepMind team suggests that AI tools such as LLMs could fill this gap.
To find out whether LLMs could serve as effective mediators, the researchers trained LLMs called Habermas Machines (HMs) to serve as huddle mediators. As part of their training, LLMs are taught to identify areas of overlap between the views of people in opposing groups, but not to try to change anyone’s views.
The research team used a crowdsourcing platform to test the mediation capacity of their LLM. Volunteers were invited to interact with a HM, who then attempted to elicit the volunteers’ views on certain political topics. The HM then produced a document summarizing the volunteers’ views, in which he was asked to give more weight to areas of overlap between the two groups.
The document was then given to all volunteers who were asked to provide criticism, after which the HM modified the document to take into account the suggestions. Finally, the volunteers were divided into groups of six people and took turns mediating critiques of the statements compared to the statements presented by the HM.
The researchers found that volunteers rated the HM’s statements as higher quality than the humans’ statements 56% of the time. After allowing volunteers to deliberate, the researchers found that the groups were less divided on their issues after reading the material provided by the HMs than after reading the document provided by the human mediator.
More information:
Michael Henry Tessler et al, AI can help humans find common ground in democratic deliberations, Science (2024). DOI: 10.1126/science.adq2852
© 2024 Science X Network
Quote: DeepMind researchers discover that LLMs can serve as effective mediators (October 18, 2024) retrieved October 18, 2024 from
This document is subject to copyright. Apart from fair use for private study or research purposes, no part may be reproduced without written permission. The content is provided for informational purposes only.
