Illustration: Tag Hartman-Simkins / Futurism. Source: Getty Images
ChatGPT and other AI chatbots are replacing search engines. Instead of going through the pains of looking up sources, these powerful, large-scale language models simply make up an answer for you, but there’s a small chance it’s a complete fabrication.
A new study published in the journal Science finds another danger. This method may be quick to get answers, but it is not very suitable for actual learning. PNAS Nexus.
“When people rely on large language models to summarize information about a topic, they tend to have less knowledge about that topic than if they learned it through a standard Google search,” said study co-lead author Shiri Mehrmad, a professor at the Wharton School at the University of Pennsylvania. conversation About her work.
The findings are based on an analysis of seven studies with more than 10,000 participants. The gist of their experiment is: Participants were asked to learn about a topic and randomly assigned to either use only an AI chatbot, such as ChatGPT, or a standard search engine, such as Google, to conduct a survey. Finally, participants were asked to write advice to a friend about what they had learned.
A clear pattern emerged. Participants who conducted the research using AI produced short advice with less general tips and factual information, while participants who used Google search produced more detailed and thoughtful tips. This pattern remained even after controlling for factors such as the information users saw during the study and the tools they used by presenting each group with the same facts.
“Even holding facts and platform constant, our findings confirm that learning from synthesized LLM responses results in less knowledge than when collecting, interpreting, and synthesizing the information yourself via standard web links,” Melumad wrote.
Scientists are only just beginning to understand the long-term effects of AI use on the brain, but the body of evidence so far suggests worrying risks. One large, high-profile study conducted by researchers at Carnegie Mellon University and Microsoft found that people who trusted the accuracy of AI tools had atrophied critical thinking skills. Another study found that students who relied heavily on ChatGPT were associated with memory loss and tank grades.
“One of the most fundamental principles of skill development is that people learn best when they are actively engaged in the content they are trying to learn,” Melmad explained. “When we learn about a topic through Google search, we face even more ‘friction’: we have to navigate various web links, read information sources, and interpret and synthesize them ourselves.”
“However, with LLM, this entire process is performed for you, turning learning from a more active process to a passive one,” Melad added.
While scientists continue to investigate the risks of AI, the technology is rapidly making inroads into education and has become a popular tool for cheating on assignments. While major companies like OpenAI, Microsoft, and Anthropic are spending millions of dollars to train teachers on how to use their AI products, universities are also partnering with similar companies to create their own customized chatbots to push to students, such as Duke University’s creatively named collaboration with OpenAI, DukeGPT.
Learn more about AI: As AI takes over, student math and reading scores hit record lows
