The rapid integration of AI tools like ChatGPT into everyday life has been framed largely as a leap forward—an upgrade to how we work, learn, and communicate. But beneath the surface of this technological fluency, a quieter shift may be taking place. What happens to the brain, and perhaps to the self, when the effort of thinking is repeatedly outsourced? A new study from researchers at MIT’s Media Lab offers early but sobering insight into that question. Using EEG brain scans, linguistic analysis, and behavioral observation, the researchers tracked how different methods of essay writing—using ChatGPT, a search engine, or no tools at all—affected participants over time. The results were not just statistically significant. They were deeply human.

Participants who relied solely on their own minds showed stronger neural connectivity, greater memory retention, and a deeper sense of authorship. Those who leaned on ChatGPT, by contrast, exhibited diminished brain activity, less originality, and an increasing disconnect from their own writing. Over time, their thinking became more automated, their memory thinner, their engagement shallower. While the study is still awaiting peer review, its findings echo larger questions many are already sensing: What is the cost of convenience when it comes to cognition? And what might we be losing—not just intellectually, but spiritually—in the process?

How AI Changes Brainwaves

The first brain scan study of ChatGPT users has raised serious concerns about the neurological consequences of relying on large language models (LLMs) for cognitive tasks like writing. Conducted by researchers at MIT’s Media Lab, the study used electroencephalography (EEG) to track brain activity in participants as they wrote essays using either ChatGPT, a search engine, or no digital assistance at all. The differences were both stark and consistent.

Participants who used no tools—the “brain-only” group—demonstrated the strongest and most distributed neural activity across the cortex. EEG scans showed robust connectivity in alpha, theta, and delta frequency bands, patterns typically associated with creative ideation, memory retention, and semantic processing. This group also reported a greater sense of ownership over their work and higher levels of satisfaction, suggesting deeper cognitive and emotional investment in the task.

In contrast, the group using ChatGPT showed the weakest neural connectivity. Executive control and attentional engagement were notably low, especially by the third session, where many participants had fully delegated the task to the AI. Rather than using the tool to augment thinking, they used it to bypass it. EEG readings confirmed a downscaling in cognitive load: memory networks were scarcely engaged, and there was minimal activation in regions linked to complex reasoning or idea generation.

This pattern worsened over time. The study followed users across multiple sessions, and ChatGPT users showed a steady decline in neural engagement. When asked to rewrite an earlier essay without using ChatGPT, these participants struggled to recall their own content and showed reduced alpha and beta brainwave activity—markers of shallow cognitive imprinting.

The research also revealed that switching tools mid-study produced measurable changes. Those who moved from using ChatGPT to relying on their own minds exhibited under-engaged brain patterns, indicating that previous reliance on AI had dampened their cognitive readiness. Conversely, participants transitioning from brain-only writing to ChatGPT use still maintained strong engagement, suggesting that prior cognitive effort may buffer against immediate neural drop-off.

While the sample size was modest and the study has not yet undergone peer review, its methods and conclusions align with broader concerns in neuroscience and education. Cognitive activity appears to scale inversely with tool dependence: the more the machine does the thinking, the less the brain is called upon to participate.

The Real Cost of Thinking Less

The MIT researchers coined a powerful term to summarize the trade-off they observed: cognitive debt. This concept describes how the immediate gains in efficiency and convenience offered by AI tools accumulate into a long-term deficit in our core cognitive abilities. Each time we outsource a demanding mental task, we bypass the cognitive effort necessary to build and maintain our neural pathways. The task gets done, but we’re left neurologically and intellectually poorer for it. This debt, the study shows, isn’t just a metaphor; it manifests in a series of clear, measurable, and deeply concerning behavioral symptoms.

The first and most striking symptom is a near-total collapse of memory and comprehension. In a post-writing interview conducted just minutes after the task, participants were asked to quote a sentence from the essay they had just produced. The results were stark. In the very first session, virtually none of the participants in the ChatGPT group could accurately quote a single sentence from their own work. Another analysis put the failure rate at 83% for the AI group, compared to just 11% for the brain-only and search engine groups. This profound memory deficit persisted across all sessions, even after participants gained more practice with the tool. This isn’t a simple failure of memorization. Being able to recall what you’ve written is evidence of deep engagement—a sign that you’ve wrestled with the ideas, structured the logic, and chosen the words carefully enough for them to become part of your own cognitive framework. The inability to do so points directly to a breakdown in the learning process. The neurological data provides a clear explanation: the weaker alpha and theta brainwave activity in the LLM group suggests a “bypassing of deep memory processes.” Because the brain didn’t generate the content, it didn’t properly encode it. The work was completed, but no knowledge was internalized. This is cognitive debt in its purest form: a finished product that leaves no cognitive asset behind.

The second symptom is a fragmented sense of authorship and a diminished sense of self. When asked about ownership of their essays, brain-only writers almost unanimously claimed full credit and expressed satisfaction with their work. In stark contrast, LLM users reported a “fragmented and conflicted sense of authorship.” Some denied ownership entirely, while others assigned only partial credit to themselves, expressing a clear disconnection from the final text. The researchers link this psychological distancing to the diminished connectivity in brain networks responsible for self-monitoring and evaluation. When we think and write for ourselves, we engage in a constant internal dialogue of drafting, critiquing, and revising. This feedback loop is fundamental to self-awareness. By outsourcing this process, we appear to quiet that internal narrator, leading to a feeling that the work—and the thoughts within it—aren’t truly our own.

The third symptom is what one report called the “flattening effect”: the production of homogenized and “soulless” content. Linguistic analysis revealed that essays written with AI assistance showed significantly less variation in vocabulary, sentence structure, and conceptual approach. When mapped mathematically, the AI-assisted essays clustered tightly together, occupying a small, generic conceptual space, whereas the brain-only essays were spread across a much more diverse landscape. This was confirmed by two human English teachers who graded the essays. They described the AI-assisted writing as having a “close to perfect use of language and structure while simultaneously failing to give personal insights or clear statements.” They called the work “soulless.” This finding reveals a critical truth: original expression is born from the struggle to find the right words. When we rely on an LLM, we’re not just getting help with grammar; we’re adopting its probabilistic, often bland, way of assembling ideas. The resulting “homogenized communication” is a direct reflection of a homogenized thought process. The terrifying implication is that by habitually using these tools, we risk not only writing like machines but, through the persistent mechanisms of neuroplasticity, beginning to think like them, sacrificing our unique intellectual and personal voice in the process.

Using AI Wisely For a Sharper Mind

Perhaps the most insightful part of the MIT study was its final phase. After three sessions in their assigned groups, the researchers introduced a crucial twist for the fourth and final session. They switched the conditions for a subset of the participants: habitual ChatGPT users were asked to write without any tools (the “LLM-to-Brain” group), while those who had previously relied only on their minds were given access to ChatGPT (“Brain-to-LLM”). This elegant design allowed the scientists to move beyond a simple snapshot of brain activity and investigate two crucial questions: Do the effects of AI use linger even after the tool is gone? And can AI be used in a way that enhances, rather than diminishes, cognitive function? The answers to both are profoundly illuminating.

For the LLM-to-Brain group, the results demonstrated a persistent cognitive deficit. When these habitual AI users were forced to write on their own, their brain connectivity did not simply rebound to the level of the brain-only group. Instead, their neural engagement remained significantly weaker, showing continued “under-engagement of alpha and beta networks.” Even without the tool, their brains were not working as hard as those who had consistently practiced unassisted thinking. They failed to “catch up” to the cognitive engagement levels of their peers. This neurological finding was mirrored in their writing. They showed a clear behavioral residue, exhibiting a bias toward using the specific vocabulary and phrasing favored by the AI in their previous sessions. The AI’s linguistic patterns had been absorbed, altering their natural writing habits. This is the cognitive debt coming due—a measurable, lasting impact on both brain function and creative output, suggesting that repeated reliance on AI may inhibit the development of the strong, independent neural pathways that emerge from focused, unassisted practice.

However, the results from the Brain-to-LLM group tell a different, more hopeful story—a potential path to redemption. When participants who had first spent three sessions building their cognitive “muscles” were finally given access to ChatGPT, their brain activity did not quiet down. On the contrary, their brain-wide connectivity surged, spiking across all frequency bands to a level that exceeded not only their own previous brain-only sessions but also every session recorded for the habitual LLM users. This finding is the single most important and actionable insight from the entire study. It suggests that the impact of AI on the brain is not predetermined; it is conditional. The order of operations is critical. When a mind has first established a strong cognitive foundation through self-driven effort, AI ceases to be a crutch that causes atrophy. Instead, it can become a powerful tool that augments and even enhances neural engagement. As the researchers themselves concluded, these findings support an educational model that “delays AI integration until learners have engaged in sufficient self-driven cognitive effort.”

It is vital to approach these findings with nuance and a balanced perspective. The study, while groundbreaking, is a pre-print, meaning it has not yet completed the full peer-review process, and its sample size of 54 participants is relatively small. Lead author Nataliya Kosmyna has been careful to clarify that the results do not show “brain rot” or permanent, irreversible damage. The study captures a reduction in cognitive engagement in a specific context, not necessarily a permanent loss of cognitive capacity. As some expert critics have pointed out, the LLM group didn’t need to think as hard to complete the task, so they didn’t. The brain, like a muscle, is a “use it or lose it” organ, and the EEG scans may simply be showing a muscle at rest. This distinction between engagement and capacity is the central scientific and philosophical tension of the study. While a single instance of non-engagement does not equal atrophy, the study’s four-month duration and the lingering deficits observed in the LLM-to-Brain group suggest a dangerous trajectory. Chronic under-engagement is precisely the path that leads to diminished capacity over time. The MIT study provides the first clear, neurological warning signal on this slippery slope. It suggests that while we may not be losing our ability to think, we are losing the habit of it—and the long-term consequences of that habit loss are what make these initial findings so terrifying.

The Power of Thinking For Ourselves

The MIT study offers more than a scientific warning; it provides neurological evidence for a truth long held by spiritual traditions: effort isn’t a burden to be avoided, but a necessary path to growth, self-awareness, and a richer inner life. By translating concepts like attention and engagement into the language of brainwaves and neural connectivity, the research elevates the conversation about AI from a technological debate to a spiritual imperative. Psychologists call the intrinsic motivation to engage in challenging mental tasks “Need for Cognition” (NFC). Individuals with high NFC enjoy effortful thinking, leading to deep learning and greater well-being. Conversely, those with low NFC tend to act as “cognitive misers.” The “cognitive debt” described by the study is the price of becoming a cognitive miser—a debt paid with diminished memory, fragmented ownership, and a quieted mind. This study stands as a stark warning against engineering a society that systematically discourages the very effort required to build a robust mind and a stable sense of self.

From a wellness perspective, this effort is the engine of neuroplasticity—the brain’s remarkable ability to reorganize itself by forming new neural connections. Just as muscles need resistance to grow stronger, the brain needs cognitive challenges to forge and reinforce its pathways. When we wrestle with a difficult problem or write an essay from scratch, we are actively sculpting our own neural architecture. The quieted brain scans of the ChatGPT users are a picture of neuroplasticity in reverse—a brain that isn’t being challenged, and therefore, isn’t growing. At the heart of this process is our most precious resource: our attention. As neuroscientists and spiritual teachers alike have long asserted, how we direct our attention literally changes the brain; the brain becomes what we regularly focus on. The “dimming of the mind” seen in the EEG scans is the direct consequence of withdrawing our focused attention from the creative process. When we outsource our thinking, we are not just saving time; we are divesting our conscious awareness, risking a state of profound disconnection where we become passive observers of our own output.

The future of the human mind isn’t a battle against technology, but a choice about how we engage with it. AI is a powerful tool that can augment human capabilities. The study’s hopeful finding—that a prepared mind can use AI to achieve even greater levels of neural engagement—points the way forward, suggesting AI designed to engage us in thinking rather than doing the thinking for us. Ultimately, the “terrifying” results of this study aren’t a prophecy of doom but an empowering call to action. They reveal that in an age of increasingly intelligent machines, the most vital human skill is the deliberate, conscious choice to engage in effortful, focused, and original thought. This is a spiritual practice—a radical act of preserving and expanding the unique, felt quality of human consciousness. Choosing to think for ourselves, to embrace the struggle and the effort, is how we ensure that the future of the mind remains vibrantly, irreplaceably human.

Source:

1. Kosmyna, N., Hauptmann, E., Yuan, Y. T., Situ, J., Liao, X., Beresnitzky, A. V., Braunstein, I., & Maes, P. (2025, June 10). Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task. arXiv.org. https://arxiv.org/abs/2506.08872

Loading...