Cognition
When the Mind Stumbles, It Grows
Is AI smoothing our thinking at the cost of the friction that fuels insight?
Updated July 30, 2025 Reviewed by Abigail Fagan
Key points
- LLMs smooth thought, but friction is where insight and originality are born.
- AI may be erasing the tension that makes us think deeply.
- LLMs can reduce cognitive friction, but used wisely, they can also support deeper thinking.
There’s a strange new smoothness creeping into our thinking—yours and mine.
Large language models (LLMs) have quietly woven themselves into our cognitive routines. They offer instant answers, well-crafted prose, and the ability to finish our thoughts before we’ve fully had them. OK, so far so good. But lately, I’ve been wondering about what happens when thinking itself becomes too perfect or too complete? And is our cognitive navigation becoming more like Waze, which lets us take the quickest and easiest path?
That's worrying me, because human thought was never meant to be smooth.
Our thoughts hesitate and often contradict. They get stuck, recalibrate, and sometimes even fragment before something new takes shape. That friction isn’t a flaw. It’s how we grow. The resistance we try to avoid is often the very force that drives deeper insight, better ideas, and, at times, transformation.
Minds Are Built on Instability
The mind doesn’t operate like a clean, linear processor. It’s a living system that is dynamic and often rather unstable. Competing beliefs, sensory inputs, emotional responses, and shifting expectations constantly jostle for primacy. And here's the key point: real insight doesn’t arise despite that instability. It emerges from it.
Even our highest-performing mental states, like flow, creativity, and peak experience, are born through tension, not ease.
Take flow, for instance. Psychologist Mihály Csíkszentmihályi described it as the immersive state where action feels seamless and time distorts. But flow isn’t what happens when everything is smooth; it’s what happens when challenge and skill lock into fragile balance. Too little challenge and the mind drifts. Too much, and it fragments. Flow lives on the edge of chaos, and it’s an active negotiation with instability—not an escape from it.
Peaks Begin in the Climb
Or consider Maslow’s idea of the peak experience—those rare moments of insight, awe, or even self-transcendence. These aren’t states of casual bliss. They often come after deep effort, doubt, or conflict. Maslow tied them to self-actualization, a process built on cognitive friction and the reconciling parts of the self, and pushing through limitations. The peak doesn’t arrive unless there’s something to climb.
This is the architecture of cognition, its instability followed by adaptation or even dissonance, followed by synthesis. Where the mind stumbles, it learns to reconfigure itself. And that reconfiguration is often where the growth happens. And it's this very human realization that stands in dramatic counterpoint to the anti-intelligence of AI.
When Machines Remove the Friction
LLMs are designed to erase that fertile instability. Their purpose is to predict and complete, and not to dwell in contradiction or to pause for reconsideration. They’re optimized for fluency, not for struggle.
And as we come to rely on them for everything from writing, thinking, and even imagining, we may start to internalize that fluency as the new model for cognition or perhaps even as a "colonizer" that may impart new and concerning cognitive hegemony, And while we might get faster answers, it may also be a narrowing of thought itself.
The dangerous path that we might find ourselves on is one of completion over contemplation. That is a slippery (frictionless) slope.
A Quiet Signal From the Brain
Recent research at MIT reinforces this concern. In a study using EEG to monitor brain activity during writing tasks, participants who used ChatGPT showed significantly weaker neural connectivity than those who wrote unaided. Not only was their cognitive engagement lower during use, but the effect persisted even when they returned to independent work. Their essays were less diverse, their memory of the material poorer, and their sense of ownership diminished. The researchers call this “cognitive debt.” But I still wonder if this is a new form of efficiency and not simply the early signs of disengagement?
What Happens When We Stop Struggling?
I think we all need a level of “desirable difficulty” that provides a template for productivity. It’s the same principle at work when students learn more by wrestling with a problem than by being shown the solution. Simply put, struggle matters, and without it, we don’t have that template for lasting insight.
LLMs, by design, can smooth that struggle out. They offer the finished thought before the false start. But if we consistently bypass the hard part that includes the tension, the confusion, the reworking, we’re not thinking more efficiently. We’re thinking less fully.
This isn’t a call to reject the tools but a challenge to stay alert to the hidden trade-offs. Because when our tools stop challenging us, we risk losing complexity. But when used correctly, they offer a powerful and transformative opportunity to take that bumpy road that today may be one less traveled.
