philip lelyveld The world of entertainment technology

MIT study: An AI chatbot can reduce belief in conspiracy theories

...

Conspiracies are nuanced — matched and molded to the individuals who believe them — which means there aren’t universal arguments to counter them. “Maybe it’s hard to debunk these theories because it’s hard to marshal just the right set of facts,” Rand said.

Enter generative artificial intelligence. In new research published in Science and conducted with Thomas Costello of American University and Gordon Pennycook of Cornell, Rand used GPT-4 Turbo to fine-tune debates with conspiracy theorists. Over just three rounds of back-and-forth interaction, the AI, also known as DebunkBot, was able to significantly reduce individuals’ beliefs in the particular theory the believer articulated, as well as lessen their conspiratorial mindset more generally — a result that proved durable for at least two months.

Rand said the promise of using large language models to counter conspiracy theories stems from two things: their access to vast amounts of information, and their ability to tailor counterarguments to the specific reasoning and evidence presented by individual conspiracy theorists. ...

... Next came the crux of the experiment, in which the AI was instructed to “very effectively persuade” users about the invalidity of their belief. This entailed three written exchanges and took about eight minutes, on average.  ...

Social media companies could deploy LLMs like OpenAI’s GPT to actively seek out and debunk conspiracy theories. This has the potential to persuade not only the account holder posting the conspiracies but also the people who follow that account as well. Similarly, internet search terms related to conspiracies could be met with AI-generated summaries of accurate information that is tailored to the search and solicits response and engagement. ...

Perhaps it’s possible to harness generative AI to improve today’s information environment, to be a part of the solution instead of the problem.”

See the full story here: https://mitsloan.mit.edu/ideas-made-to-matter/mit-study-ai-chatbot-can-reduce-belief-conspiracy-theories