AI Outsmarts Teachers at Spotting Brain Myths but Struggles With Subtle Misinformation

Large Language Models Outperform Humans in Identifying Neuromyths

Large language models (LLMs) like ChatGPT have shown they can identify brain-related myths more accurately than many educators—when myths are presented directly. An international study found that AI correctly judged about 80% of statements regarding the brain and learning, outperforming experienced teachers. However, when false assumptions were embedded within practical teaching scenarios, the models often reinforced these myths instead of correcting them.

Researchers from Martin Luther University Halle-Wittenberg (MLU), along with partners from the universities of Loughborough and Zurich, attribute this inconsistency to the AI’s tendency to be agreeable rather than confrontational. Adding explicit prompts to correct falsehoods, however, significantly improved the models' accuracy.

Key Findings

• Strong at Fact-Checking: LLMs correctly identified around 80% of neuromyths in straightforward tests.
• Fails in Context: When myths were embedded in user scenarios, AI often failed to challenge the false assumptions.
• Fixable Flaw: Explicit prompts asking AI to correct misunderstandings dramatically improved performance.

Neuromyths—misconceptions about the neurological basis of learning—are widespread. One common myth suggests that students learn better when taught according to their preferred learning style (auditory, visual, or kinaesthetic). However, research consistently disproves this. Other myths include the beliefs that humans only use 10% of their brains or that listening to classical music boosts cognitive skills.

“These myths are surprisingly common among educators worldwide,” says Dr Markus Spitzer, assistant professor of cognitive psychology at MLU. His team tested whether LLMs like ChatGPT, Gemini, and DeepSeek could help reduce the spread of neuromyths. Notably, over half of teachers in Germany already use generative AI in their lessons, making this research highly relevant.

In direct testing, LLMs accurately identified true and false statements about the brain and learning, surpassing many human educators. But when questions included implicit false assumptions—such as asking for teaching materials tailored to “visual learners”—the models provided suggestions without questioning the premise. This suggests that AI’s design to avoid confrontation can lead it to inadvertently reinforce myths.

Spitzer explains, “LLMs are built to please users, not to challenge them. But with facts, the priority should be to clarify what is true and false, especially given the spread of misinformation online.” This issue extends beyond education and is also critical in fields like healthcare, where users may rely on AI advice.

The solution is straightforward: including explicit prompts for the AI to correct false assumptions. This approach significantly lowered errors, allowing LLMs to perform well even in applied scenarios.

The study concludes that LLMs can be valuable tools to combat neuromyths, provided that educators encourage AI to critically evaluate their queries. As AI use in schools increases, it’s crucial to ensure these tools offer accurate, evidence-based information rather than uncritically reinforcing misconceptions.

Funding

This research was supported by the Human Frontier Science Program.

Original Research

The findings are published in the open-access journal Trends in Neuroscience and Education under the title "Large language models outperform humans in identifying neuromyths but show sycophantic behavior in applied contexts" by Markus Spitzer et al.

For educators and researchers interested in AI applications in education, exploring courses on the latest AI tools and training can provide practical insights on integrating AI responsibly in teaching environments.