AI and social media are rotting your brain cells, warns research

A new research led by Shiri Melumad, professor at the Wharton School of the University of Pennsylvania, suggests that larger language models (LLMs) such as ChatGPT may reduce the depth of learning as compared with traditional web search. The study indicate that while LLMs offer quick, synthesised answers but they come at a cost, as due to the ease the users exert less effort and also develop shallower knowledge and produce less original ideas.

The study published in PNAS Nexus, tested around 10,000 participants access several experiments. The results showed that the users who used ChatGPT reported learning fewer new things, felt less ownership over their knowledge and wrote advice that was shorter, more generic and less persuasive as compared to the users who used Google search as a medium for learning.

ChatGPT vs Google Search: Why efforts matter in learning

The findings of the study are build on the concept of ‘desirable difficulty’. The idea is that effortful learning often leads to deeper understanding. Traditional web search needs the users to navigate links, interpret multiple sources and also synthesise information themselves. This research argue and friction force the users to process the searched information more deeply.

On the other hand, the LLMs like ChatGPT perform this synthesis automatically which lead to reduction in the cognitive work involved. As a result of this, the users may feel they have learned less and generate advice that is less informative and unique.

Evidence across experiments

ul1
• li1Experiment 1: Participants asked to advise a friend on planting a vegetable garden learned either via Google or ChatGPT. Those using Google spent more time searching, reported deeper learning, and wrote longer, fact-rich advice.
• li1Experiment 2: Even when the information was identical across conditions, the format mattered. Google users engaged more deeply, while ChatGPT users produced sparser, more similar advice.
• li1Replications: Effects held across topics of personal relevance (healthy lifestyle, financial scams) and even when LLM summaries included real-time web links.

Implications of the use of AI

The study revealed that while LLMs have proven to be effective in tasks like writing, coding and exam preparation, but overreliance of the model can hinder your knowledge development and the content created may also lack originality and deeper understanding of the topic. AI tools save time but may discourage the kind of active engagement that builds lasting knowledge structures.

Melumad and colleagues argue that while AI can be a powerful aid, it risks turning users into passive recipients of information rather than active learners. The message is clear that AI and social media may be reshaping not just how we access information, but how deeply we understand it.

GN Awards 2025: Vote for your favorite Gadgets