Researchers are testing large language models as tools to detect, explain, and even change people’s minds about false claims.
Misinformation is abundant online and can take the form of false images, fake voices, and bots made for propaganda purposes. Researchers like Jevin West (a misinformation researcher at the University of Washington) believe it is logical to use artificial intelligence (AI) as a way to combat the creation of misleading or outright false information, since the intention of the two technologies is the same.
From Machine Learning to Language Models
Earlier misinformation-detection tools relied on machine learning models trained on verified true-or-false claims, which could identify suspicious patterns with strong accuracy on narrow datasets. But these systems struggled to generalise across topics and platforms. Newer efforts have turned to large language models, which draw on vast amounts of internet text and can now search the web for current information before responding — though accuracy still varies significantly, with one early study finding a version of Grok agreed with human fact-checkers only about 55% of the time.
Handling Ambiguity and Context
One problem that researchers hope to answer regarding how LLMs can handle ambiguous or contradictory information. Many groups are developing LLM solutions that allow for the detection of ambiguous situations (for example, asking clarifying questions instead of making assumptions about statements). Some are creating tools, such as Dubawa in Nigeria. Some initiatives look at the relationships between language patterns associated with manipulation and the use of actual fact-checking; these types of projects typically find that LLMs produce answers that match human reviewers approximately 70% of the time.
Beyond Detection: Changing Minds
Using LLM technology to measure how a false narrative spreads is one way to use LLMs for journalism—using them to cover global stories rather than just to fact-check specific individual claims. For example, a recent Science study found that LLMs can reduce the confidence level of individuals in their belief of conspiracy narratives by as much as 20% through the use of LLMs (average of all participants); thus, using LLM technology to combat false narratives may be more effective than traditional methods.
A Tool, Not a Substitute
As noted earlier, AI and LLMs have come a long way in improving our ability to identify and respond to misinformation; however, they should be viewed as an aid and not as a replacement for humans who perform fact-checking. Thanh Thi Nguyen, an AI researcher, stated that LLMs require human supervision to operate correctly—similar to how a parent guides their child’s growth and development toward independence.
