Something Disturbing Happens When You “Learn” Something With ChatGPT
ChatGPT and other AI chatbots are increasingly replacing traditional search engines. Instead of spending time searching through various sources, these advanced large language models provide quick answers. However, there is a minor risk that these answers may be completely fabricated. Beyond this risk, something disturbing happens when people rely on AI chatbots to learn new information.
A recent study published in the journal PNAS Nexus reveals that while AI chatbots offer fast responses, they are not effective for deep learning. According to Shiri Melumad, a professor at the Wharton School of the University of Pennsylvania and co-lead author of the study, people who depend on large language models to summarize topics tend to develop shallower knowledge than those who learn through traditional Google searches. This finding highlights a significant downside to using AI chatbots for learning.
How AI Chatbots Affect Learning Depth
The study analyzed seven experiments involving more than 10,000 participants. In these experiments, participants were asked to learn about a topic. They were randomly assigned to use either an AI chatbot like ChatGPT or a standard search engine such as Google. After their research, participants were instructed to write advice for a friend based on what they had learned.
A clear pattern emerged from the results. Those who used AI chatbots wrote shorter advice with generic tips and fewer factual details. In contrast, participants who used Google searches produced more detailed and thoughtful advice. This difference persisted even when researchers controlled for variables such as the information presented and the tools used. Melumad explained that learning from synthesized responses generated by large language models led to shallower knowledge compared to actively gathering, interpreting, and synthesizing information from web links.
The Broader Implications of AI on Learning and Critical Thinking
Scientists are still exploring the long-term effects of AI on the brain, but early evidence raises concerns. One notable study by researchers from Carnegie Mellon and Microsoft found that people who trusted AI tools too much experienced a decline in critical thinking skills. Another study linked heavy reliance on ChatGPT among students to memory loss and declining academic performance.
Melumad emphasized a fundamental principle of skill development: people learn best when actively engaged with the material. Using Google search involves more “friction” because users must navigate multiple web links, read different sources, and synthesize information themselves. In contrast, AI chatbots perform this entire process for the user, turning learning into a more passive experience.
Despite these risks, AI technology is rapidly entering education. It has become a popular tool for cheating on assignments. Companies like OpenAI, Microsoft, and Anthropic are investing millions to train teachers on how to use AI products effectively. Universities are also partnering with these firms to develop customized chatbots for their students. For example, Duke University has collaborated with OpenAI to create “DukeGPT,” a chatbot designed specifically for its students.
As AI continues to dominate, concerns grow about its impact on learning. The evidence suggests that something disturbing happens when people “learn” something with ChatGPT: their understanding remains shallow, and their critical thinking skills may suffer. This calls for careful consideration of how AI tools are integrated into education and learning processes.
For more stories on this topic, visit our category page.
Source: original article.