Your Brain on ChatGPT: New Neuroscience Reveals The Hidden Cost of AI Writing

The "Zombie" Writer Phenomenon

This is the fear that has loomed over English departments and classrooms ever since ChatGPT went public: Are we raising a generation of students who can prompt but not think?

Teachers always knew that when a student outsourced their essay to a computer program, they were doing something to avoid the work but also to avoid something more substantial. The essay was finished. The grammar was flawless. However, the student's intellectual faculty was not exposed to any ideas they supposedly took ownership of.

Until very recently, this has mainly been anecdotal. The educators had to rely on gut reactions to recognize in AI-produced text the "hollow" tone that they sensed was absent. However, a very interesting new study has moved this question from the faculty room to the neuroscience laboratory.

This data reveals that once writers are very reliant on AI, the activity level in their brains drastically collapses. However, this research also discovered something very interesting - a certain usage pattern for AI that increases brain power.

Within the Experiment: The Spark of Thought

To investigate how our minds work while collaborating with silicon, a research team coordinated by Nataliya Kosmyna at MIT Media Lab conducted a significant experiment on 54 participants. The group, comprising undergrads and faculty, and staff at the university, was asked to write compositions using college entry essay questions from the SAT.

The authors divided participants into three groups with distinct rules for interactive engagement:

• AI Group: Free usage of GPT-4o to aid their writing.
• The Search Group: Allowed to use search engines for research purposes, but not allowed to use Generative AI.
• The "Brain-Only" Group: No technology whatsoever is to be used. It's just the author and the empty page.

As participants sat with pen to paper to write, much more than their time was measured. Literally, they were observed as they wrote, as their brains' connectivity patterns were tracked by electroencephalogram technology to form a pattern of actual activity.

The Results: The Neural Cost of Convenience

The EEG results presented a disquieting image for those in favor of using AI freely in education.

The "Brain-Only" group had the strongest and most widespread activity in the brain: Their brains were lighting up for tasks of creative idea generation, integrating scattered bits of information, and self-monitoring. The difficulty with organizing thoughts, formulating arguments, and editing sentences created a very high level of internal demand, keeping their brains busy and "fit."

In contrast, the AI group demonstrated the least activity within their brains. They had outsourced "the hard work of organizing and drafting" to ChatGPT. As a consequence, "their brains functioned in a kind of 'power-saving mode.'"

However, this was only the tip of the iceberg in terms of the implications. The studies demonstrated significant memory and ownership deficits:

• Memory Loss: Later, when they were asked to reproduce lines from their own writings for their essay quotations, only 11% could not remember their writings, while 83% of AI writers failed to do so.
• Loss of Ownership: The psychological ties to the work were lost. The individuals within the AI group felt that "there was a 'fragmented and conflicted sense of authorship' whereby many individuals did not feel that the essay was, in fact, very much their own."

Independent evaluators verified these internal flaws: Although grammar in AI-assisted essays was perfect, evaluators did find they lacked "personal insights" and "individuality."

The Plot Twist: How to Wake Up the Brain

Had they ended there, the implications would be rather frightening: AI will rot your brain. However, they did add a fourth session to this experiment that turned everything upside down.

This is the final stage where they flipped the scenario by allowing "Brain-Only" writers to utilize ChatGPT, while those using AI had to perform manual writing.

What happened instead was that when writers, who had already created content on a topic manually, meaning those belonging to the "Brain-Only" group, proceeded to work on the next level using AI, they started increasing their brain activity.

"What it could potentially tell us is that timing could be very important when you integrate these kinds of tools," said lead author Nataliya Kosmyna.

This means that it is not the tool that constitutes a problem, but rather its deployment at a given time. For example, where a student has brainstormed, outlined, and written out ideas prior to using ChatGPT, this AI becomes more of a collaborator to enhance cognition rather than substitute for it.

"Maybe now you are able to ask questions, go back and forth," Kosmyna observed. "You have your opinions on the topic, you can prompt in different directions."

The Pedagogical Crisis: "Outsourcing Too Early

For educators, this work offers the scientific basis for "scaffolding," as it is known in classrooms.

"The big risk here is outsourcing too early," says Brett Vogelsinger, an English teacher at a high school and AI in education blogger. "You need to have some writing time that's you, and the paper, and your thoughts."

What is at stake in introducing AI too quickly into the learning process is that children are not able to develop that "muscle memory" that comes with written expression. Steve Graham, professor at Arizona State University, explains that children need to understand complex syntax to write a sentence effectively, which children are not able to do orally. ChatGPT is always building the house, but children are not laying bricks.

By not possessing this essential ability themselves, they are unable to assess what their AI does. They can't tell whether or not their essay is good because they never learned what high quality feels like to create.

The Future of Student Writing: Strategic Timing

This preliminary version, awaiting review, outlines what AI in education could look like in the future. "A total prohibition on AI is not required, but to give students 'free rein' to utilize AI is detrimental."

"The sweet spot" for optimal brain function and education appears to involve "A Sandwich Method":

• Human First: "The actual work on the idea is done by using the brains of the students to write out ideas by hand. This entails high activity of the brain."
• Collaboration with AI: "The actual work on the idea is accomplished using AI to critique, expand on, or refine existing ideas." This entails increased activity between the brains.

Now, as we bring this very powerful technology into our schools and our workplaces, let's not forget to include efficiency over cognition as our aim. As shown in the data from the EEG scan research, to create stronger brains, our brains must work against something in order to resist. And as we remove all resistance to writers and to writing, perhaps we are actually smoothing out our own brains as well.

Takeaway for Students and Writers:

• "Do not start with the prompt. Start with a blank sheet."
• "Your brains must develop the idea's 'neural pathways' before you develop it using AI."
• The 'Recall' Test: can you orally cite your own essay or remember its principal arguments from memory?" Then you likely overused AI assistance.
• Use AI as "editor," not "ghostwriter." Hence, AI must only add to ideas, but not start ideas."

Now, as shown in this research above, "this research illustrates that using AI later in the 'writing' act increases activity in your brains."