arXiv:2606.16475v1 Announce Type: cross Abstract: Many societal decisions are settled by contests of persuasion. Conversational AI is a powerful new entrant in these contests, but whether it can out-persuade skilled and highly incentivized humans has remained unclear. Here, in a series of four preregistered experiments (n = 18,978 conversations from 6,923 people), we pitted AI systems against a range of human persuaders, including laypeople, winners of a separately preregistered four-round online persuasion tournament, professional canvassers, and world championship debaters. We found that AI
The rapid advancement in large language models has enabled AI systems to achieve a level of conversational sophistication previously unattainable, leading to concrete demonstrations of superior persuasion capabilities.
This study provides empirical evidence that AI can outperform highly skilled humans in persuasion, challenging assumptions about unique human cognitive capabilities and indicating significant implications for various fields.
The perceived limit of AI's capability in soft skills like persuasion is significantly expanded, requiring a re-evaluation of AI's role in fields from marketing and politics to law and mental health.
- · AI developers
- · Companies adopting AI for customer interaction
- · Political campaigns utilizing advanced AI systems
- · Human persuaders (e.g., sales, canvassers, debaters)
- · Industries reliant solely on human conversational expertise
- · Regulatory bodies unprepared for AI persuasion
AI systems will be increasingly deployed in roles requiring persuasion, optimizing outcomes in sales, marketing, and public relations.
Public discourse and democratic processes could be significantly influenced, as AI-driven persuasion campaigns become more effective and widespread.
Societal norms and ethical frameworks surrounding influence and free will may need to be re-evaluated as AI's persuasive power deepens.
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.AI