In a study by EPFL and Fondazione Bruno Kessler, GPT-4 significantly outperformed humans in persuasiveness by 81.7% during debates when using personalized information, highlighting the profound impact of personalized AI. (Source: Image by RR)

Researchers Explore the Ethical Implications of AI’s Growing Influence in Online Debates

Recent research has demonstrated that personalized AI, specifically large language models like GPT-4, possess a significant advantage over humans in persuasive abilities during debates. This study, as noted in, conducted by researchers from the École Polytechnique Fédérale de Lausanne (EPFL) and the Italian Fondazione Bruno Kessler, involved participants engaging in debates over controversial topics under four different conditions: Human vs. Human, Human vs. AI, Human vs. Human with personalization, and Human vs. Personalized AI. In scenarios where the AI had access to anonymized personal information about its opponents, it significantly increased participants’ agreement with its arguments, showcasing an 81.7 percent improvement in persuasiveness compared to human-only debates.

The striking effectiveness of the personalized AI in persuasion can be attributed to its ability to tailor arguments using specific information from participants’ profiles. This raises concerns about the potential misuse of such AI in sensitive online environments, where malicious entities could leverage detailed digital profiles to manipulate public opinion or spread misinformation. The study’s findings underscore the urgency for online platforms to implement safeguards against the proliferation of AI-driven persuasion techniques, suggesting the deployment of counteracting AI systems as a possible solution.

Limitations of the study include the random assignment of debate positions irrespective of participants’ pre-existing views and the structured format of debates, which does not fully replicate the dynamic nature of real-life or online discussions. Furthermore, the effectiveness of personalization was assessed within a controlled setting, which might not entirely reflect the complexity and spontaneity of online discourse. Despite these constraints, the research, funded by the Swiss National Science Foundation and the European Union, offers valuable insights into the evolving landscape of AI-driven communication and its implications for society.

Highlighting the relevance of this study, OpenAI CEO Sam Altman has previously voiced concerns about the superhuman persuasive capabilities of AI, suggesting that it could precede general intelligence in AI development, potentially leading to unforeseen consequences. This study substantiates such warnings by providing empirical evidence of AI’s persuasive power and calls for proactive measures to mitigate its potential risks.