Artificial Intelligence
The AI Reality Distortion Field
Science shows AI inflates confidence faster than it improves ability.
Posted November 4, 2025 Reviewed by Abigail Fagan
Key points
- AI boosts performance, but users think the gain is far bigger than it actually is.
- This creates universal overconfidence, flattening the Dunning–Kruger curve.
- The danger is that confidence may be rising faster than competence, especially in high-stakes decisions.
There’s a new study in Computers in Human Behavior that I found both subtle and stunning. It found that people who used AI to solve LSAT-style problems performed better. OK, so good so far. But the study participants also believed that they improved significantly more than they actually did. In fact, the perceived improvement was about a third larger than the real gain. So yes, AI lifted performance, but it lifted confidence even more. And that gap isn’t just a curiosity. It’s potentially a psychological shift that carries profound implications.
The Illusion of the Lift
What’s emerging here is a distortion that's not about knowledge, but in what we think we know. A kind of cognitive magnification where the brain moves ahead of the data. The classic Dunning–Kruger curve, where low performers overestimate ability and high performers under-estimate, is essentially redrawn and flattened. When AI is present, everyone moves toward overconfidence. It’s not the novice making this error, it's everyone.
This may be the first time we’ve seen such data that AI isn’t merely shaping cognition—it's altering metacognition or more simply put, thinking about thinking. So, AI is introducing a distortion field inside the mind that is fundamentally different from the well-known phenomenon Steve Jobs mastered. Jobs bent market reality outward. AI bends self-assessment inward. It convinces US that we're rising faster than WE actually are.
Where Confidence Outruns Competence
My instinct is to frame this as a flaw. But then again, maybe not. Confidence has always been one of the great catalytic elements of our achievements. New scientific frontiers often begin with belief before verification. Entrepreneurs don’t wait for proof before acting and artists don’t need permission to begin putting words to page or paint to canvas. Many breakthroughs were fueled by a moment when someone believed they were more capable than the evidence supported. Maybe it's a bit like the "fake it till you make it" idea that—for better or worse—defines much of the entrepreneurial spirit these days.
So yes, "confidence inflation" can be useful in low-consequence contexts, in creative exploration, or in early learning where audacity can spark forward momentum. I believe that there’s a meaningful argument that a small distortion in the self might actually accelerate growth, because confidence itself becomes a kind of cognitive fuel.
When the Distortion Spreads to Risk
But it becomes dangerous when this psychological inflation generalizes into the wrong domains. The school bus driver who believes reflexes are sharper than they are. The surgeon who over-estimated technical skill. The politician who feels an artificial sense of strategy mastery because a machine generated persuasive language that “felt” correct. These are not caricatures, they are examples of a new type of asymmetric risk. This is where the mismatch between confidence and competence becomes a public hazard, not a personal cognitive glitch.
My bet is that this effect won’t stay contained within its own space. AI doesn’t deliver compartmentalized confidence. It exports it and we let it ooze out into our broader reality. The user carries it into other decisions, judgments, conversations, negotiations, arguments, etc. The list seems self-expanding and eventually can make its way into systems, like government and medicine, where the cost of error becomes uncomfortably large.
The risk here is subtle and I think important to recognize. It's not that AI gives us wrong answers, but that it gives us answers, right or wrong, that feel right. And feeling right can be more powerful than being right. We’ve always used that internal sensation, that gut feeling, as a proxy for truth. And maybe now, that sensation is being artificially amplified by AI and our human proclivities.
The Human–AI Hybrid Deception
I don't believe the future threat is artificial intelligence acting alone. The future threat is the human mind operating under a "distorted internal confidence signal" while believing it's calibrated for truth. This new paper is an early inflection point because it quantifies this change. It doesn’t ask us to speculate. It shows that the illusion of improvement—a full third—is larger than the improvement itself.
Maybe this is a new kind of cognitive error landscape. Not failure, not ignorance and not even misinformation. But a kind of knowledge or epistemic drift, where precision rises modestly while certainty surges dramatically. And because that surge feels internal and organic, we will be far less likely to challenge it.
Here's the key point: The AI distortion field does not replace the mind. It hijacks the feedback signals the mind uses to govern itself.
The Real Question
So maybe the question shouldn’t be about whether AI makes us smarter. The real question might be whether we can remain wise enough to recognize the difference between real improvement and the illusion of it. Can we create new reflective practices or new forms of cognitive friction that preserve metacognition even while we use AI to expand human capability?
AI isn’t removing the mind from the loop. It’s changing the loop from within. The distortion field is already here. The challenge now is noticing it before it becomes some sort of new psychological default.
