Did you know AI Fact-Checking Results in Mixed Outcomes, Sometimes Boosting Misinformation and Distrusting Truthful News

 

A recent study in Proceedings of the National Academy of Sciences explored how large language models (LLMs), like ChatGPT, affect people's perceptions of political news. The research, driven by the rise of misinformation, found that AI-based fact-checking didn't always help users identify truth from falsehood. In fact, sometimes it made people trust false headlines more and true headlines less.

LLMs like ChatGPT can quickly analyze vast amounts of content and spot errors. But when it comes to fact-checking, the study shows mixed results. While ChatGPT could identify false headlines with high accuracy (90%), it was uncertain about true headlines. In some cases, its uncertainty led people to doubt truthful information or believe lies.

The study involved 2,159 participants, who were shown 40 political headlines—half true, half false. They were divided into groups: some saw AI fact-checks, others saw human fact-checks, and some had no fact-checking at all. Those who saw human fact-checks performed better at distinguishing true from false news.

But the results weren't the same for AI. When ChatGPT was uncertain, people were less likely to trust true headlines, even if they were accurate, and more likely to believe false ones. Participants exposed to AI fact-checks were also more likely to share incorrect news, particularly if the AI didn’t provide a clear answer.

Interestingly, people who actively chose to view AI fact-checks were often already biased, more likely to share both true and false news, depending on their attitudes toward AI. The results show a potential problem: people may trust AI too much or not enough, depending on the situation.

The study's findings raise critical questions about how LLMs like ChatGPT are used in everyday life, especially on platforms where misinformation spreads quickly. It’s clear that these AI systems aren't foolproof and should be used with caution.

This research highlights the complexities of using AI for fact-checking. While these models are powerful, they need improvement. Future work will focus on refining their accuracy and understanding how people interact with them, especially in real-world settings like social media.

In the end, we need to be aware that technology, no matter how advanced, can’t replace human judgment. Understanding how to balance the use of AI in our digital world is crucial for making sure it serves society rather than misguiding it.