Cognitive Surrender: How AI Is Changing the Way We Think
I remember preparing for WAEC a few years ago. I attended online tutorials because of the COVID outbreak, and we had this group where our instructors sent questions to solve. Sometimes they were assignments; other times, they were competitive questions to determine the smartest students.
Some questions were straightforward, but others made you struggle for hours with no clue how to start.
Of course, I used Google Image Search a few times and hoped someone somewhere had solved the exact problem or something similar. I got lucky a few times, but other times, Google brought up everything except what I was actually looking for.
I’d go through multiple textbooks, look at similar questions, try formulas, and eventually I’d figure it out. I remember that feeling. The little moment of satisfaction when I finally got that answer. I didn’t just get the answer. I understood how I got there and how to solve similar questions I encountered during that predicament.
Fast-forward a few years.
Today, I can take a picture of that same question, upload it to ChatGPT, and have a detailed solution in seconds. Pretty incredible, right?
It is. And I’m genuinely grateful that technology has made learning this accessible.
But I started noticing something about myself. There were questions I knew I could solve. Maybe they’d take more time, but I was capable of figuring them out. Yet whenever I encountered one, my first instinct was increasingly to ask AI.
The thought that something else could solve it faster gradually made me lose the willingness to struggle through problems.
It turns out there’s a name for something very similar.
Cognitive Surrender.
In January 2026, the Massachusetts Institute of Technology (MIT) established an ad hoc committee to examine how artificial intelligence was changing teaching, learning, and research. The committee was tasked with understanding how students and lecturers were using AI, identifying its opportunities and risks, and recommending policies for its use.
By August, the committee had released its 40-page report on AI and education.
The report discussed cognitive surrender, a situation where students begin relying on AI at the first sign of difficulty instead of attempting to work through problems themselves. The report drew on research by Steven Shaw and Gideon Nave, who studied how people could become willing to accept AI-generated answers without properly questioning them.
The researchers found that access to AI improved people’s accuracy when its answers were correct. But when AI gave incorrect answers, people’s accuracy dropped, even as their confidence increased.
You could be confidently wrong simply because an AI system sounded convincing.
MIT’s committee also observed that students were increasingly tempted to hand over difficult tasks to AI, especially when deadlines were approaching. And it’s understandable. When you’re dealing with assignments, exams, projects, and several other responsibilities, efficiency becomes attractive.
Why spend hours completing something when a chatbot can finish it in minutes?
But the committee warned that this convenience comes at a cost: weaker independent thinking and reduced confidence. Getting the right answer can create the illusion that you’ve learned something.
But have you?
Your Brain on ChatGPT
Interestingly, this wasn’t the only MIT-related research raising questions about AI and learning.
In June 2025, researchers associated with the MIT Media Lab released a study titled Your Brain on ChatGPT. They recruited 54 participants for the first three sessions of an experiment involving essay writing.
One group used ChatGPT, another used search engines, and the third had to write without either tool. While participants worked, researchers measured their brain activity using EEG technology.
The results revealed differences. The group writing independently showed the strongest and most widespread brain-connectivity patterns during the tasks. Search engine users showed moderate connectivity, while ChatGPT users showed the weakest.
Participants who relied on ChatGPT struggled more to accurately recall passages from their own essays and reported feeling less ownership of their work.
Imagine submitting an entire essay and struggling to remember what you wrote.
The researchers raised concerns about what they described as cognitive debt, where reducing mental effort through AI could have consequences for learning.
Of course, the study was relatively small and remains a preprint. It doesn’t prove that ChatGPT permanently damages our brains.
But alongside MIT’s newer report, it provides a pattern.
The rise of the ChatGPT student
Not everyone who uses ChatGPT to study is doing something wrong, but I worry we’re heading towards a generation of students who can produce impressive academic work without necessarily developing the skills behind it.
Get The ChalkBoard’s stories in your inbox
Join Medium for free to get updates from this writer.
Think about a mathematics student who submits ten perfectly solved problems using AI. That’s ten correct answers on paper. But change the values, rearrange the question, or introduce a condition they’ve never encountered before.
Can they still solve it?
If not, what exactly did those ten correct answers represent?
There’s actual research that makes this even more interesting.
In a 2025 study published in PNAS, researchers conducted an experiment involving nearly 1,000 high school mathematics students.
Students who had access to a general-purpose AI assistant performed better while practising. However, when the AI was taken away, their exam performance was 17% worse than that of students who had practised without AI.
But students who used a specially designed AI tutor that encouraged them to work through problems rather than simply providing solutions largely avoided that decline.
So the issue wasn’t simply that students used AI.
It was how they used it.
Learning was never just about the answer
One thing the MIT committee emphasised was the importance of productive struggle. The mental effort involved in figuring things out, making mistakes, revisiting concepts, and gradually building understanding.
We’re increasingly measuring our progress by how much work we complete rather than how much we understand. It looks productive, except the student may not develop the ability to do anything independently.
Before AI, we had study groups; we asked our friends or visited teachers after lectures for clarity. Today, it’s often easier to open a chatbot than to have that conversation.
MIT’s report raised concerns about this too, warning that AI could reduce the human interactions and collaborative problem-solving experiences that make education valuable.
We risk losing not just the habit of thinking independently, but also the habit of thinking together.
So, should students stop using AI?
Definitely not.
I still use AI, and I don’t intend to stop. In fact, I think it has made learning more accessible than ever.
But we can change how we approach AI.
If you’re a student, a few things are worth trying to prevent cognitive surrender.
Attempt a problem first. Even if you don’t get far, identify what you know and where you’re struggling before asking AI.
Ask AI to guide you. Instead of requesting a complete solution immediately, ask for hints, an explanation, or feedback on your approach.
Test what you’ve learned. After understanding a solution, close the chatbot and try a similar question independently.
Don’t abandon human learning. Discuss difficult ideas with classmates, ask lecturers questions, and try to explain a concept you understand to someone else.
But the responsibility isn’t entirely on students. Schools and universities also need to rethink how they teach and assess learning.
MIT’s committee recommended clearer AI-use policies for individual courses, more emphasis on practical and collaborative learning, and assessments that better reveal what students genuinely understand.
The point wasn’t to ban AI from education. It was to ensure that AI supports the process of learning rather than quietly replacing it.
The goal of education was never simply to produce students who could submit the correct answers. It was to develop people capable of understanding problems, questioning assumptions, reasoning, and applying what they’ve learned.
Because eventually, you’ll encounter a problem that doesn’t look exactly like the one you’ve seen before.
And when that moment comes, what matters won’t be how quickly ChatGPT could solve yesterday’s assignment.
It’ll be whether you learned enough to solve today’s.
AI should make us more capable thinkers, not give us more reasons to stop thinking.
Originally published at https://chalkboard.usechalkos.com on October 10, 2026.
