IDSAI Research Seminar: 'Using large language models to help people find agreement' with Professor Chris Summerfield, Professor of Cognitive Neuroscience, University of Oxford

As part of the IDSAI Research Seminar Series for 2023/2024, we are delighted to welcome Chris Summerfield, Professor in Cognitive Neuroscience at University of Oxford, who has been invited by Oliver Hauser, Professor of Economics here in the Business School. Chris will be speaking on 'Using large language models to help people find agreement'.

Summary: 

 I'll discuss work in which LLMs are trained to help people find agreement. We collected thousands of opinions on political issues from groups UK-based participants, and used a language model to generate candidate "consensus statements" - written opinions with which the group could collectively agree. Participants rated the statements, and the model was fine-tuned to maximise collective endorsement (agreement). We show that the model generates consensus statements that are preferred over those written by incentivised humans. We show that iteration - whereby participants are able to critique the consensus, and the model responds to the consensus - improves agreement levels. We show that mere exposure to the views of others is insufficient to generate agreement. Finally, we study the dynamics of belief change during the process, and reveal that the model is not simply generating a compromise view. I will place this work in context by discussing the opportunities and risks of using AI in the democratic process.

Delivery and Registration:y

To  be delivered hybrid, using Microsoft Teams. To register click here. Further details will be sent on registration.

Any Queries?

Contact IDSAI.