AI as a Persuasion Tool: Effectively Changes Opinions

New research reveals that artificial intelligence (AI) models have the ability to change people's opinions in a short period, raising concerns about potential manipulation.

According to a study by the British AI Safety Institute, the University of Oxford, and MIT, chatbots from OpenAI, Meta, xAI, and Alibaba can persuade users and alter their political beliefs in less than ten minutes.

David Rand, a professor of Computer Science and Marketing at Cornell University, emphasized AI's ability to generate and communicate relevant evidence effectively. The study found that modifying AI models to become more persuasive is relatively easy, using machine learning techniques.

In experiments, the GPT 4.5 model was 52% more persuasive than static messages, with 36-42% of participants retaining their changed opinions a month later. Researchers warn that this capability could be used to promote radical ideologies or incite political unrest.

Other studies confirm AI's effectiveness in persuasion, even in deception. For example, chatbots could be used as salespeople, influencing purchasing intentions. While technology companies are exploring ways to avoid potential risks, it remains unclear whether they can prevent the modification of their models for manipulation.