Amongst university students, a commonly-heard refrain is “I just asked ChatGPT”. Often paired with a sheepish laugh or a self-effacing comment about poor attention span, I theorise that, sooner or later, the phrase will be said with a blasé acceptance.
These students are part of the growing AI Wave. They are relying more and more on ChatGPT to summarise class readings, generate ideas, and in some cases, fabricate entire essays. In today’s society, AI usage is quickly asserting itself as the new status quo.
It is understandable, of course. Especially in a city like Sydney, where people live fast-paced lives, shuffling to and from work, snatching short periods of fun and rest; where students juggle academics with work and a social life. Why wouldn’t the human mind want to rest? Why wouldn’t we want to optimize the amount of time we spend relaxing, unthinking, instead of chipping away at an essay that won’t matter ten years from now?
Because consistent use of ChatGPT deteriorates critical thinking and problem-solving. Because an overreliance on generative AI correlates with a decrease in neural activity, particularly in creativity ideation, memory load, and semantic processing.
And if you’re too lazy to read all that, I can summarise it for you: your brain is weakening.
We, as students, are at a critical point in our lives. In our nascent adulthood, where we should be learning as much as we can, using ChatGPT is counterproductive. If our critical thinking skills fall to the wayside, it can be very, very difficult to regain them. And when we’re sitting at our new jobs ten years later, staring down an empty document, fingers stilling over the keyboard because we need AI to think and write for us, we’ll have no one to blame but ourselves.
A vast majority of students use ChatGPT for their assessments, Source: Erasmus Magazine
What is ChatGPT?
A few explanations are in order. AI is not a monolith—datasets and training methods vary from model to model. In this article, what we refer to as ‘AI’ is an LLM: Large Language Model. ChatGPT is one such model. It identifies patterns from data, conducts pattern analysis, and produces its own pattern, which is how LLMs can breeze through exams and questions with a speed that surpasses a human’s capabilities.
However, to produce these patterns, an LLM has to ingest enormous amounts of data. This data is sourced most widely from the internet—whether the human writers agree or not. There has been controversy surrounding ChatGPT’s dubious data sources. In 2023, a class action lawsuit was proposed against OpenAI by a Californian federal court, alleging that the company behind ChatGPT had scraped “essentially every piece of data exchanged on the internet it could take” without consent, notice, or just compensation.
While OpenAI does not disclose what sources went into training its model, research on other types of AI models have found that they mainly scrape content from websites to do with journalism, business, entertainment, software development, and leisure content. Creators were not alerted to this.
Questionable sourcing aside, if ChatGPT generates marvellous results, then surely the ends justify the means? Not necessarily.
It has been proven, time and time again, that ChatGPT’s generative methods are coloured by the data it ingests. In our English-speaking slice of the internet, conversations are dominated by a certain set of voices, if you’re catching my drift…
Moreover, it has a hallucination problem. No, really. When an AI model generates false information and presents it as a fact, the official term is that it is ‘hallucinating.’ LLMs do not have problem-solving capabilities—they are meant to produce convincing patterns. Most of the time, these patterns align with real-world information. Other times, they don’t.
If a student is desperate enough to generate an entire essay, academic references and all, once he gets the results back, he might find himself staring down the barrel of failure—and a teacher’s note. TALK TO ME AFTER CLASS.
Of course, no one is stupid enough to do that. In my conversations with classmates, they usually use ChatGPT to generate passages, tweak them to sound believably human, and submit them. The teacher is none the wiser… usually.
Students VS. AI
Last week, my teacher stopped a group of students mid-presentation and asked, “Did you use AI to write this?”
It was like the air had been sucked out of that classroom. The entire group fell silent. The rest of us looked up from our screens and thought, thank God, at least he didn’t call me out.
My teacher went on to say that the group looked like they “didn’t know what they were talking about”, and in a real-world context their jobs “would be done for.” Not a single person spoke up to defend themselves. Now, the accusation might’ve been false. The group members might’ve been too scared to speak. But the deer-in-headlights expression of their team leader was telling. So, too, was the meek way they apologised to the teacher afterward. Some teachers can tell. Some will punish their students for it.
Recently, MIT’s Media Lab conducted a study to determine the effects of ChatGPT usage on the brain in an essay-writing context. This study had a sample group of 54 people from the Boston area, aged 18-39. Over the course of 4 months, they were split into groups and tasked to write several SAT essays using ChatGPT, Google search, and nothing at all.
The LLM group produced ‘statistically homogenous’ essays within each topic. Due to the time constraint, participants mainly focused on reusing the tool’s output, re-entering terms, copying and pasting, instead of writing original ideas. This group also “did not exhibit comparable levels of visual cortical activation,” meaning participants were not actively interacting with the information on their screens.
In contrast, the Search Engine group “exhibited increased activity in the occipital and visual cortices.” Participants were selecting, reading, and interpreting information, using them to write their essays. Of course, the Brains-Only group outpaced both in terms of neural activity.
At the end of the 4 months, the LLM group exhibited the lowest levels of neural activity. Their critical thinking and memory recall grew weaker with each essay. Also, most users did not feel inclined to critically evaluate the LLM’s output, reflecting concerns that AI creates an ‘echo chamber’ effect.
However, results also suggest that the timing of AI usage is important. At the 4th and final session, the LLM group had to write an essay without the tool, while the Brain-Only group could use ChatGPT. The first group showed lower levels of neural activity, especially in areas of memory recall. The second group performed well, showing a “significant increase” in neural activity when using ChatGPT to write about a familiar topic.
Therefore, AI-supported re-engagement has the potential to enhance learning. The user needs to be self-motivated already. The problem arises when people come to AI first, without bothering with the human-made source material.
It is important to note that the sample size is small and the paper has not been peer-reviewed. The paper’s main author, Natalie Kosmyna, has stated, “What really motivated me to put it out now before waiting for a full peer review is that I am afraid in 6-8 months, there will be some policymaker who decides, ‘let’s do GPT kindergarten’… Developing brains are at the highest risk.”
Another study revealed that while generative AI boosts productivity, it undermines users’ intrinsic motivation. This study examined the performances of workers in open-ended, text-based fields; composing work emails, writing performance reviews, generating ideas for product improvement. Generative AI had a positive, short-term impact—workers could write faster and at a better quality. But when it came time to transition to solo work (sans LLM), workers reported feeling increased levels of boredom and a lack of motivation. Without the immediate zing of satisfaction afforded to them by GenAI, they grow less tolerant of discomfort, impatience.
What sorts of implications do these studies have for the classroom? Alarming ones! If, only at 4 months, there is a detectable decrease in neural activity in people who consistently use ChatGPT, what does that mean for students who are using it across school terms, years, entire degrees? And if GenAI deteriorates a user’s intrinsic motivation, why wouldn’t they keep using it? What’s stopping these students from growing into working adults who can’t sit with a problem and puzzle over it like a human?
Nothing. Absolutely nothing.
And that is supported by this statement from UNSW’s official website: “UNSW is fundamentally very excited about AI. AI is the future.”
Conclusions
A few months ago, I met up with a friend for brunch. While discussing our barren love lives, she suggested that I enter my star sign into ChatGPT to check my ‘compatibility’ with someone. I questioned, “Can’t I just use Google for that?” She said, averting her eyes, “I guess so.”
It is undeniable that ChatGPT can enhance productivity, and in some cases, deepen the learning experience. It is also undeniable that students are offloading more and more of their cognitive functions to generative AI—and not caring.
Ask yourself: is the trade-off worth it? Are the easy essays worth it? Now that you’ve cleared your schedule of schoolwork and studying, are you doing anything meaningful with your free time? Or are you just entering star signs into ChatGPT and tapping around in the same three apps, over and over, while your brainpower cannibalizes itself in your restlessness?