Artificial intelligence (AI) has continued to impact various areas of life, both positively and negatively, and the education sector is one that is being affected, particularly due to the rise of generative AI (Gen AI) tools, such as OpenAI’s ChatGPT.
A recent study from the Massachusetts Institute of Technology’s (MIT) Media Lab, titled “Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task,” has shown some worrying results in the area of education.
The research included 54 participants, aged 18 to 39 years, all of whom were students from various educational institutions in the Boston area. The participants were then divided into three groups, all of which had one specific task — writing essays.
Each group was then assigned to do things differently. One group can only use ChatGPT (aka LLM Group), another uses Google’s search engine without the AI functionality (aka Search Engine Group), and the last would have no tools at all (aka “Brain-only” Group).
The students then underwent three writing sessions, during which they could select their preferred SAT essay writing prompts.
Using an electroencephalogram or EEG, a test that measures the electrical activity of the brain across 32 regions, the researchers found that the ChatGPT user group had the lowest brain engagement and “consistently underperformed at neural, linguistic, and behavioral levels.”
- Undetectable AI? OpenAI says AI writing detectors don’t work
- ChatGPT messaged first: Should we be worried?
Nataliya Kosmyna, one of the main authors of the research, even commented that the LLM group had “pretty homogeneous” outputs, which used a lot of very similar words.
Kosmyna also noted that high neural connectivity occurs for brain-only participants, but there’s a “scaling down” of brain activity when participants transition from using only their brain to using the LLM.
Over the next several months, the LLM group “got lazier” with each of the following essays, even resulting in simply copying and pasting.
After the initial three essays, the researchers invited the participants again for a fourth session, but only a third of the original participants showed up. This meant a smaller sample size was observed from an already-small original sample size of 54.
For the fourth essay, the students were reassigned to different groups. For instance, those in the Brain-only group were given ChatGPT, while those who initially used ChatGPT would no longer be able to use it. They were then tasked with writing an essay on the topics they had previously written about.
The switching of tools for the groups yielded a surprising result, as the brain-only group for the first three sessions performed even better with the help of ChatGPT.
The researcher added that these findings can be crucial for future studies, particularly in highlighting the importance of timing the introduction of an LLM tool to children. The paper suggests that using LLM for younger users can harm their learning, particularly for small children whose brains are still developing.