This is your brain. This is your brain on ChatGPT.

Should ChatGPT carry a warning label? The science seems to suggest yes. But can we use it safely?

Of course using ChatGPT changes the physical structure of your brain! But not any more so than doomscrolling or web surfing or binge watching or practically anything else that you do to fill your mortal hours. This is not so much an excuse for ChatGPT, rather than a reminder that the squishy ball of dough we carry upstairs will happily reshape itself in response to whatever we’re doing with it, whether we want it to or not. That’s neuroplasticity, baby, and it’s the only reason our silly species was able to invent reading and writing in the first place! Maybe this explains why your (my) consciousness seems to have the same vibe as an email or text thread or whatever media form has seized the majority of your waking life. If you had been born fifty years earlier, it would have been cinema! Sorry!

Anyway, before we jump into the delicious doomsplaining research related to the cognitive effects of ChatGPT, let’s do a quick refresher on neurons, synapses, and the electrochemical Burning Man that turns our passing thoughts into physical structures in the brain. To understand how ChatGPT-like tools can shape our brain dough, we need to first understand how that brain dough works. (I believe the more appropriate metaphor is circuit, to which I say, oh well!)

Your brain, according to neuroscience

Your mental activity is the result of countless neurons firing in coordinated patterns across specialized networks in your brain. Every thought, memory, decision, or sentence you write arises from electrochemical signals traveling across synapses, forming dynamic neuron dough circuits shaped by your experiences, habits, and environment. According to the last count, the human brain contains about 86 billion of them with around 100 trillion connections. They look like prehistoric sea life from the Cambrian period about 500 millions years ago, and for all I know they are, because didn’t we come from the sea? Below is a picture of a neuron with a smiley face, because putting smiley faces on complicated things enhances my cognitive stamina when my neural circuitry is pretty much crawling out of my head to check my phone. For what?! Like what could possibly be of any interest on my phone at this stage of life at 10:08 p.m. I am a poor, foolish heap of neuron dough, that’s for sure.

A smiling neuron that wants to be understood.

Anyway, when you’re born, your baby skull already contains nearly all 86 million neurons, but the connections between them, called synapses, are still forming. Initially, they are lonely, dark islands with no wifi, no phone lines. But in the first few years of life, the brain forms millions of new synapses each second as sights, sounds, and sensations trigger neurons to fire, which send signals that help form and strengthen connections through repeated use. This overproduction is followed by synaptic pruning, where useful connections are strengthened and unused ones are eliminated. The result is a more efficient brain, wired for the things you do most. Like Instagram scrolling! Or saying mean things to yourself!

After early childhood, neuroplasticity continues in a more targeted way. The brain stays flexible, reshaping pathways based on what you do. Reading and writing, for instance, are acquired through this ongoing process of forging and reinforcing neural circuits. Regularly reading challenging texts can strengthen circuits for deep focus and reasoning. Writing by hand can reinforce memory and language networks more effectively than typing. Over time, the brain shifts resources toward frequently used circuits and prunes the rest. In short, your daily cognitive diet literally sculpts your mind, or what researchers call neural architecture.

High speed microscope captures fleeting brain signals (UC Berkeley)

As you can probably imagine, there are quite a few neural architecture possibilities when you have 86 billion different building blocks. Everyone’s neural architecture is different, shaped by a unique combination of genetics, biology, and the accumulated effects of experiences, habits, and learning over a lifetime. London cab drivers, for example, have huge hippocampi as a result of memorizing the city’s 25,000 labyrinthine streets. However, there are also shared patterns between individuals, especially when engaging in culturally universal activities like reading, writing, or language processing. These shared patterns form the basis of what researchers call functional networks—common circuits that activate across brains during specific cognitive tasks even as the precise wiring and strength of those circuits remain individualized.

Functional networks are key for understanding how tools like ChatGPT affect our neural circuitry, because they provide a map of expected cognitive engagement for specific tasks like deep reading and essay writing. Deep reading, essay writing, speech, for example, all have individual functional network patterns across individuals. If we expect these networks to activate in predictable ways, but see altered or reduced engagement when humans rely on AI assistants, we can begin to identify how such tools may be rerouting or disrupting cognitive behavior and reshaping how thought is formed.

So what does the research say?

I’m sorry, it’s so boring. I mean, it’s important and blabbity blah, but “I’d rather be reading this amazing 19th-century book on the philosphy of beards” if you know what I mean. Here’s the deal. You have a whole bunch of stressed out students and stressed out knowledge workers living on a burning planet being told that it is very important that they write essays or emails or reports or whatever for a grade or a paycheck. There is such extreme cognitive dissonance between the action they are being told to take and the existential crisis at hand that their neural circuitry is pressurized into a silky lace that elegantly suppresses their critical capacities and true desires. I am procrastinating in delivering the scientific research, I’m sorry, I can’t help it. I really do have the goods. It’s just that you already know exactly what ChatGPT does to your brain. Can you guess. Please. The scientists say struggle is good for your circuits. Prompt yourself: “You are a human with reasonable intelligence and are guessing what the cognitive effects of ChatGPT are.” Do not scroll past the image until you’ve generated your output.

Yes, you did it! ChatGPT enables our laziness in ways that can atrophy our cognitive capacities for memory, critical thought, judgement, and deep thinking. They are crossing the rainbow bridge to join our recently-departed attention spans and ability to navigate our own neighborhoods without GoogleMaps. We wish them well! We show them the door! We tell them we’re happier without them! We will never, ever miss them!

Yes, this is all backed by science. First, it’s important to note that there is much we can infer about ChatGPT from decades of related studies in media and cognitive science. Long before LLMs, researchers were already documenting how technologies like GPS, web search, social media, and even airplane autopilot systems facilitate cognitive offloading, or the process of outsourcing cognitive tasks to external tools. These tools can lead to greater efficiency, but sometimes at the cost of memory, spatial reasoning, or critical thinking. Studies on skimming versus deep reading, for instance, show that fast, hyperlinked environments promote shallow processing and reduce long-term comprehension. So ChatGPT isn’t some brand new devil, it’s an intensification of existing trends:

• Over reliance on AI can cause “cognitive muscle atrophy”: A survey of 319 knowledge workers found that while generative AI tools make critical thinking feel less effortful, they paradoxically lead to less actual critical thinking. People who trust AI more think critically less, while those confident in their own abilities engage more deeply despite the extra effort. (Microsoft Research, 2025)

• Students using ChatGPT show increased procrastination and memory problems: Research with 494 university students revealed that heavy ChatGPT users were more likely to procrastinate, experience memory difficulties, and achieve lower academic performance—especially when facing high workloads and time pressure. (Abbas et al., 2024)

• Cognitive offloading creates a dangerous dependency: While external memory aids boost immediate performance, losing access to saved information results in worse performance than never using external aids at all. This research suggests our brains adapt to rely on these tools in ways that may weaken internal processing. (Nature Reviews Psychology, 2025)

• "Google effect" is real and getting stronger: We tend to forget what we believe is easily accessible. A meta-analysis confirmed that people increasingly outsource memory to digital devices rather than remembering information themselves, with the effect being particularly pronounced when using smartphones versus computers. (Frontiers in Public Health, 2024)

• Younger users show more AI dependence and weaker thinking: Among 666 participants (clearly, a Satanic study), frequent AI tool users demonstrated significantly lower critical thinking abilities. The effect was strongest in younger participants, who showed both higher dependence on AI tools and lower critical thinking scores overall. (Societies Journal, 2025)

• Student writing quality deteriorates with "metacognitive laziness": In an analysis of 56,000 post-ChatGPT essays, researchers found student writing was significantly simpler with less varied vocabulary. Research shows students offload crucial learning tasks like evaluation, planning, and reflection to AI, leading to longer but less complex texts, increased use of AI-specific phrases, and more generic "robotic" tone lacking personal voice and creativity. (University of Helsinki, 2025)

In the last bullet, we are offered the very wonderful term “metacognitive laziness.” I’d like to underscore it, as aside from personally being an Olympian in metacognitive laziness, I think it explains the sort of overarching cause of all the complaints listed in the research above. The AI does XYZ for you, and so, you become weaker in those areas, or worse, fail to develop those capacities in the first place. Laziness is an interesting term, however, because the word is so loaded. If you’re lazy, you’re obviously undisciplined or morally depraved. Lazy people deserve the nothing that’s coming to them!

Or so the thinking often goes. But there is in fact another side of the story. I was especially pleased to find through my own unmemorable web surfing, that metacognitive laziness, however embarrassing, is in fact a triumph of human evolution. We are designed to be lazy. It helps us survive the long cold hungry winter. So when these tools present their range of opportunities to be lazy, we have to go against our evolutionary intelligence to resist them. In theory, we might use them unlazily to all sorts of untold advantage. But metacognitive opportunities for laziness are like potato chips. If they are near me, I am going to eat them all.

It’s one thing to binge on chips and know better. It’s another thing to evolve your neural architecture in a perfect body temp aquarium of metacognitive laziness. When I posted the question about whether ChatGPT should carry warning labels, an educator friend commented, “Absolutely. And I'd go further -- treat them like atom splitters and keep them in labs.” Another friend called it a white collar form of “workplace hazard.” “My employer,” she wrote, “should not be entitled to making me use tools that diminish the abilities I rely on to make a living for the sake of their own profit and productivity.” A Wall Street tech journal confessed that “Artificial intelligence was eating my brain!”

But seriously, people, calm down. When you read that Google, is rolling out its AI chatbot for the under 13 crowd, do not grieve the future generations! As Einstein said, “Part of us is still stuck in the pinhead that was the universe before it did its big booming bang. And the other part is tired.” You can use ChatGPT. Safely. You just need to reincorporate the struggle, find a way to keep those neurons firing! Here’s how I did it while writing this very post. And now it’s time for me to say goodbye, as my head is about to explode in a neural supernova. Try it. Good luck!