Giving writers the freedom to use AI as much as they want resulted in some negative outcomes, according to a new study from researchers at the Massachusetts Institute of Technology, Wellesley College, and the Massachusetts College of Art and Design, as reported in Education Week.
Participants who constructed essays with the assistance of ChatGPT exhibited less brain activity during the task than participants asked to write on their own, the study found. AI-users were much less likely to recall what they had written and felt less ownership toward their writing. AI-supported participants lacked both individuality and creativity, according to evaluators who reviewed the essays.
The outcome was different when participants first wrote essays on their own and then used AI to write on the same topics. These writers showed a boost in brain activity.
Timing could be very important for when you integrate these tools, says Nataliya Kosmyna, a research scientist at the MIT Media Lab and the lead author of the paper.
If writers first spend time thinking about their topic, organizing their thoughts, and then use generative AI, it’s possible that they can get more benefits from using the tool, she says. “Maybe now you can ask questions, go back and forth. You have your opinions on the topic, you can prompt in different directions.”
The findings could offer insight about the potential for generative AI to either circumvent the learning process or deepen students’ thinking.
Researchers divided 54 participants into different groups to write three essays each, across three separate sessions, responding to prompts drawn from the SAT college-entrance exam.
Participants in the first group could use Open AI’s GPT-4o in any fashion to assist in the writing process. In the second, participants could search the internet and use any website to inform their writing—except ChatGPT or any other large-language AI model. Participants in the third group couldn’t use any research tools – they were called the “brain-only” group.
Participants’ brain connectivity was monitored during writing tasks through electroencephalogram, or EEG, which measures electrical activity in the brain. Participants were also asked questions after their writing and had independent raters—two English teachers and one AI—score their essays.
Writers who didn’t rely on any outside support had the strongest, widest-ranging neural activity, according to EEG results. The least activity was seen in the group that used ChatGPT. The group that searched the web fell in between.
AI use can also affect a sense of ownership about the final writing product. Almost all the brain-only group believed their essays were their own. But the ChatGPT group wasn’t so sure. This group “presented a fragmented and conflicted sense of authorship,” researchers wrote. “Some participants claimed full ownership, others explicitly denied it, and many assigned partial credit to themselves.”
The two English teacher judges wrote that AI-crafted writing showed “a close to perfect use of language and structure while simultaneously failing to give personal insights or clear statements.” The AI judge, by comparison, couldn’t tell the difference between essays written with and without AI’s help.
Researchers asked participants back for a fourth session. Those assigned to the brain-only group had access to ChatGPT, and those who initially could have used ChatGPT were asked to write unaided. Eighteen of the original participants completed this session. They picked one of the three essay topics they had already written on so they were familiar with the subject matter.
The findings:
- When the ChatGPT group switched to writing without support, they didn’t reach the same level of brain activity as the group that never had access to AI.
- When the brain-only group switched to using ChatGPT, they saw an increase in brain connectivity.
“These results suggest that strategic timing of AI tool introduction following initial self-driven effort may enhance engagement and neural integration,” the researchers wrote.
Students need time to develop writing skills and form ideas without AI, educators say
Brett Vogelsinger, a high school English teacher in Bucks County, Pa., says the findings offer a red flag: “Outsourcing too early is the biggest risk here.”
This could mean incorporating AI only for revision, says Kristina Peterson, an English teacher at Exeter High School in Exeter, N.H., and the co-author of AI in the Writing Workshop with her colleague, English teacher Dennis Magliozzi.
If students are using AI as a writing partner, Magliozzi says, “you have to show up to the table having written first.”
It’s not clear how much writing expertise students need before they can make good judgments about the quality of an AI’s output, says one of the study’s authors. The study sample was small: 54 participants completed the first three sessions, and only 18 came back for the fourth.
What is clear is that it’s important to carefully consider how teachers say AI could be best used—and to heed their warnings.
Support and listen to teachers, and don’t rush to push all the tools available without understanding, says the researcher.
Education Week
