A new study from the MIT Media Lab suggests that relying too heavily on artificial intelligence to verify news may make people worse at spotting misinformation on their own. The research found that while AI tools can help users detect false news in the moment, they may also weaken independent judgment when the tools are later removed.
The study comes at a time when more people are using large language models such as ChatGPT, Claude and Gemini to find, summarize and verify news. These tools can provide quick explanations and fact-checking support, but researchers warn that convenience may come with a hidden cost: users may gradually outsource their critical thinking to the system.
AI Helped Users at First
In the MIT Media Lab study, researchers tracked 67 participants over four weeks as they evaluated news headline-and-image pairs. When participants used an AI chatbot during a session, they were 21 percent more accurate at identifying fake news, confirming earlier findings that AI can reduce belief in false information when used as an active support tool.
However, the problem appeared when the chatbot was removed. By the fourth week, participants’ ability to detect misinformation without AI had fallen by 15 percentage points compared with their performance before the study began. Even more concerning, around a quarter of participants believed they were improving, despite their actual performance declining.
The AI Dependency Paradox
The researchers describe this pattern as part of the “AI dependency paradox.” In simple terms, AI can make people better while they are using it, but worse when they must perform the same task alone. This is similar to how GPS can weaken natural navigation skills, or how calculators can reduce mental arithmetic practice if people rely on them constantly.
The same problem has been observed in other fields. Studies have shown that professionals who use AI support may become less capable of performing certain tasks independently if they stop actively engaging with the underlying reasoning. In the context of news, this matters because misinformation detection depends not only on getting one answer right, but on developing habits of questioning, checking context and comparing sources.
Why AI Can Be Risky During Breaking News
The study also highlights the limits of AI systems during fast-moving or emotionally charged news events. Large language models are statistical systems that generate likely responses based on patterns in training data. They may appear confident, but they can still produce incomplete, outdated or misleading answers, especially when information is changing quickly.
This is particularly risky during breaking news, wars, political crises or major public incidents, when misinformation spreads quickly and verified facts may still be limited. If users treat AI responses as final authority, they may accept weak or incorrect claims without doing the additional work of checking original sources, image context or official updates.
From Active Thinking to Passive Acceptance
The researchers identified a group of participants they described as “Dependency Developers.” These users gradually shifted from actively evaluating news items to passively accepting the AI system’s guidance. Instead of using AI as one tool among many, they began allowing it to carry more of the judgment process.
One participant later noted that while the chatbot reminded them to check multiple sources, it did not teach them much about analyzing the context of the images themselves. That distinction is important. A tool that gives a correct answer may solve the immediate task, but it does not always help the user build the skills needed to solve similar tasks independently later.
AI Should Be a Coach, Not a Crutch
The MIT team argues that the way AI interacts with users is crucial. Systems that simply provide direct answers may encourage dependency, while systems that ask guided questions can help users learn how to think through the problem themselves. In other words, AI works better as a coach than as a crutch.
The researchers found that methods such as Socratic questioning and deeper follow-up prompts were more likely to support long-term learning. These approaches may slow users down at first, but they encourage active reasoning. Instead of telling someone whether a news item is real or fake, the system might ask what evidence supports the claim, whether the image matches the headline, or which sources should be checked.
Why AI Literacy Matters
The study’s findings point to a broader need for AI literacy, especially in schools and universities. As students and adults increasingly use AI to answer questions, write summaries and verify information, they need to understand both the strengths and weaknesses of these tools.
The key lesson is not that people should avoid AI entirely. Rather, they should avoid handing over their judgment completely. AI can help users compare claims, organize information and identify possible red flags, but the ability to question, analyze and form independent conclusions remains essential.
Final Outlook
The MIT Media Lab study shows that AI can be useful for detecting misinformation, but only if it is designed and used in a way that strengthens human judgment. If users rely on AI as a shortcut, they may become less capable of evaluating news when the tool is not available.
As AI becomes a bigger part of how people consume information, the challenge is to build systems that teach users to think better, not systems that think for them. In the long run, protecting society from misinformation will require both better AI tools and stronger human critical thinking.