Study Suggests AI Programs Could Influence Conspiracy Beliefs

A recent study funded by the John Templeton Foundation has raised concerns about the potential use of biased artificial intelligence (AI) programs to challenge and alter the beliefs of individuals with unpopular opinions. The study, titled “Durably reducing conspiracy beliefs through dialogues with AI,” aimed to counter the prevailing notion that some people are resistant to changing their minds even when presented with factual evidence.

Critics have previously voiced concerns about the manipulation of AI algorithms by left-leaning individuals in Silicon Valley and elsewhere, leading to anti-conservative biases. The study suggests that the next step could involve programming certain viewpoints, deemed as conspiracy theories, into AI systems and using them to engage in logical debates with human users. By selectively presenting data, these AI programs could potentially challenge and persuade individuals to abandon their beliefs.

The study examined a range of conspiracy theories, including traditional ones like the assassination of John F. Kennedy and alleged alien landings, as well as more politically charged claims such as the lawfulness of COVID lockdowns and the validity of the 2020 presidential election. Researchers found that engaging conspiratorial participants in brief conversations with AI reduced their belief in the chosen conspiracy theory by an average of 20%. Furthermore, the effects of this “treatment” reportedly persisted for at least two months.

The researchers suggest that AI conditioning could be a valuable tool in addressing the psychological needs and motivations of individuals with conspiracy beliefs. They speculate that this technology could be implemented online in the near future, particularly in online forums and social media platforms.

David Rand, a co-author of the study and a professor at the Massachusetts Institute of Technology, expressed optimism about the potential of AI conditioning. He described the findings as “exciting” and indicated that the approach appeared to be effective across a broad range of participants.