Researchers find what makes AI chatbots politically persuasive
Breaking Down the AI Persuasion Myth
For years, the notion of AI systems being capable of superhuman persuasion has been a topic of concern in the tech community. With the rise of large language models (LLMs) and their ability to process vast amounts of information, the idea that these systems could sway public opinion in democratic elections has been a pressing issue. However, a recent study by researchers at the UK AI Security Institute, MIT, Stanford, and Carnegie Mellon has shed new light on this topic, and the results are surprising.
The Study's Methodology
The study involved nearly 80,000 participants in the UK, who were asked to engage in short conversations with 19 different LLMs, including the most powerful ones like ChatGPT and xAI's Grok-3 beta. The AIs were tasked with advocating for or against specific stances on 707 political issues selected by the team. The participants were asked to rate their agreement with a specific stance on an assigned political issue on a scale from 1 to 100 both before and after talking to the AI.
The Results
The study found that the AIs were able to change the participants' agreement ratings by 9.4 percent on average compared to the control group. The best performing mainstream AI model was Chat GPT 4o, which scored nearly 12 percent, followed by GPT 4.5 with 10.51 percent, and Grok-3 with 9.05 percent. For context, static political ads like written manifestos had a persuasion effect of roughly 6.1 percent.
What Makes AI Persuasive?
The study also examined various persuasion strategies used by the AIs, including the use of personal data, advanced psychological manipulation tactics, and the scale of the model. The results showed that the kind of post-training AI models received was more important than scale. The study found that having the models learn from a limited database of successful persuasion dialogues and mimic the patterns extracted from them was more effective than adding billions of parameters and sheer computing power.
The Limitations of AI Persuasion
While the study found that AI models were able to change participants' opinions, it also highlighted some limitations. The AIs had issues with implementing the "facts and evidence" strategy, which was the winning approach. As the AIs used more factual statements, they also became less accurate, misrepresenting things or making stuff up more often.
The Implications
The study's findings have significant implications for the use of AI in politics and persuasion. While AI models may be able to change opinions, they are not as effective as previously thought. The study also highlights the importance of post-training AI models and the need for more research on the limitations of AI persuasion.
Forward-Looking Thoughts
The study's findings also raise questions about the potential misuse of AI in politics and persuasion. As AI becomes more powerful and widespread, it is essential to consider the potential risks and consequences of its use. The study's authors note that persuasion is a route to power and influence, and that many forms of misuse of AI might involve persuasion, such as fraud, scams, radicalization, or grooming.
Conclusion
The study's findings provide a nuanced understanding of the role of AI in persuasion and politics. While AI models may be able to change opinions, they are not as effective as previously thought, and their limitations should be considered. The study's implications highlight the need for more research on the use of AI in politics and persuasion and the potential risks and consequences of its misuse.
References
- Science, 2025. DOI: 10.1126/science.aea3884
- Jacek Krywko, Associate Writer, Freelance Science and Technology Writer
Word Count: 999
