Competing LLM Agents in a Non-Cooperative Game of Opinion Polarisation

ACL ARR 2025 July Submission267 Authors

Abstract: We introduce a non-cooperative game that incorporates elements of social psychology, including confirmation bias, resource constraints, and influence penalties, to analyse the factors that shape the formation and resistance of opinions. A key aspect of our simulation game is the introduction of penalties for Large Language Models (LLM) agents when generating messages to spread or counter misinformation, thus integrating resource optimisation into the decision-making process. Our study reveals that a higher confirmation bias leads to stronger opinion alignment but also amplify polarisation. Lower confirmation bias result in fragmented opinions with limited shifts. Investing in high-resource debunking strategy significantly boosts the population alignment with the debunking agent early in the game, but at the cost of rapid resource-depletion, limiting long-term influence.

Paper Type: Long

Research Area: Computational Social Science and Cultural Analytics

Research Area Keywords: misinformation detection and analysis, NLP tools for social analysis

Contribution Types: NLP engineering experiment, Data analysis

Languages Studied: English

Previous URL: https://openreview.net/forum?id=bicDtw12PO

Reassignment Request Area Chair: No, I want the same area chair from our previous submission (subject to their availability).

Reassignment Request Reviewers: No, I want the same set of reviewers from our previous submission (subject to their availability)

A1 Limitations Section: This paper has a limitations section.

A2 Potential Risks: Yes

A2 Elaboration: Ethical Statement

B Use Or Create Scientific Artifacts: Yes

B1 Cite Creators Of Artifacts: Yes

B1 Elaboration: Section 3

B2 Discuss The License For Artifacts: N/A

B3 Artifact Use Consistent With Intended Use: N/A

B4 Data Contains Personally Identifying Info Or Offensive Content: N/A

B5 Documentation Of Artifacts: N/A

B6 Statistics For Data: Yes

C Computational Experiments: Yes

C1 Model Size And Budget: Yes

C1 Elaboration: 3

C2 Experimental Setup And Hyperparameters: Yes

C2 Elaboration: 4

C3 Descriptive Statistics: Yes

C3 Elaboration: 4

C4 Parameters For Packages: Yes

C4 Elaboration: 3

D Human Subjects Including Annotators: No

D1 Instructions Given To Participants: N/A

D2 Recruitment And Payment: N/A

D3 Data Consent: N/A

D4 Ethics Review Board Approval: Yes

D4 Elaboration: Mentioning the ethics detail will void anonymity. The work is approved by UWA's ethics commitee

D5 Characteristics Of Annotators: N/A

E Ai Assistants In Research Or Writing: No

E1 Information About Use Of Ai Assistants: N/A

Author Submission Checklist: yes

Submission Number: 267