Persuasion Should be Double-Blind: A Multi-Domain Dialogue Dataset With Faithfulness Based on Causal Theory of Mind
arXiv:2502.21297v2 Announce Type: replace Abstract: Persuasive dialogue is central to human communication, yet existing datasets often rely on a single language model generating both roles, producing unrealistic interactions that violate the double-blind nature of persuasion. To overcome this, we propose ToMMA, a multi-agent framework guided by causal Theory of Mind that enforces role separation and prevents information leakage. Using ToMMA, we build CToMPersu, a large-scale multi-turn, multi-domain dataset capturing realistic persuasion dynamics. Automatic evaluations show that CToMPersu prod
The rapid advancement of large language models necessitates improved data generation methods to enhance their capabilities in complex human-like interactions such as persuasion, moving beyond single-model limitations.
Sophisticated AI agents capable of realistic and ethically sound persuasive dialogue have significant implications for commercial applications, information warfare, and human-computer interaction, making breakthroughs in this area strategically relevant.
The introduction of multi-agent frameworks like ToMMA and high-quality datasets like CToMPersu provides a new paradigm for training and evaluating AI in human-centric tasks, fostering more robust and realistic AI agent development.
- · AI Foundations
- · Generative AI Developers
- · Ethical AI Research
- · Monolithic LLM Architectures
More advanced AI models become capable of nuanced persuasive communication in various digital and real-world interfaces.
The improved realism of persuasive AI agents could lead to new forms of sophisticated digital marketing, customer service, and even psychological operations.
The ability to generate persuasive dialogue in a 'double-blind' manner might necessitate new regulations concerning AI transparency and responsible agent deployment.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.CL