AI as a bridge in the political debate | A compact guide to large language models pdf free | Llms examples | Popular language models | Turtles AI

A new study suggests that linguistic models can act as effective mediators in the political debate, opening the way to a more constructive communication between opposite groups.

Key points:

• Large linguistic models can mediate between opposite points of view.
• The research was conducted by the Google Deepmind team.
• Artificial mediators have shown that they produce superior quality communications than human beings.
• The groups were less divided after interaction with AI models.

A team of Deepmind researchers, part of Google in London, recently published a study on Science, revealing the potential of large linguistic models (LLM) as mediators between groups with contrasting opinions on political issues. Political polarization has increased over the years, also fueled by the massive use of the internet, which has amplified the items of both factions, generating conflicts and misunderstandings. In this context, research tries to fill a significant gap in the mediation of debates, suggesting that LLM can take on a key role in promoting a more constructive dialogue. Scientists trained a version of LLM called "Habermas Machines" (HM), designed specifically to identify areas of agreement between opposite positions, however avoiding modifying the personal opinions of the participants. To test the effectiveness of this approach, the researchers created a crowdsourcing environment in which volunteers were invited to discuss political issues with the support of an HM. This model summarized the opinions expressed, placing emphasis on the overlaps between the different positions. Once processed, the document was returned to the participants, who were able to provide feedback and criticism, allowing the HM to perfect the initial text. Subsequently, the volunteers were divided into groups of six and played the role of mediators, comparing the statements formulated by the HM with those proposed by human mediators. The results were surprising: the declarations generated by the HM were evaluated as superior in 56% of cases compared to those of human beings. In addition, the interactions facilitated by the HM led to a lowering of the tensions between the participants, showing that mediation through AI could reduce polarization and encourage a more open dialogue. These discoveries highlight an unpublished potential for the use of technology in improving human interactions, suggesting that AI could play a significant role in the management of contemporary political divisions.

In an era of growing conflict, the future of mediation could really depend on innovations such as those developed by Deepmind researchers.