AI Boosts Performance but Blurs Self-Knowledge
A practical path to preserve personal agency In 2026.
Posted January 7, 2026 | Reviewed by Gary Drevitch
Artificial intelligence is increasingly praised for making us sharper, faster, and more productive. From drafting text to solving complex reasoning problems, tools like ChatGPT promise cognitive augmentation at scale. Yet a growing body of psychological research suggests that something subtler, and potentially more consequential, is happening alongside these gains. AI may be improving what we do while quietly distorting how well we understand our own competence.
A recent study published in Computers in Human Behavior offers one of the clearest empirical demonstrations of this tension to date. The researchers examined what happens when people use generative AI to complete logical reasoning tasks, and how accurately they can judge their own performance in the process. The results are both reassuring and unsettling.
Better Scores, Weaker Self-Assessment
The research team conducted two experiments using logical reasoning problems drawn from the Law School Admission Test (LSAT), a domain well suited for studying both performance and metacognition .
In the first study, 246 participants in the United States solved 20 reasoning problems using a custom interface. Each question appeared alongside a ChatGPT window, and participants were required to consult the AI at least once per item, either to request a solution or an explanation. After completing the task, they estimated how many questions they had answered correctly and rated their confidence for each response.
Objectively, the AI helped. Participants scored about three points higher on average than a historical comparison group whose members completed the same test without AI support.
Subjectively, however, accuracy collapsed. Participants believed they had answered roughly 17 out of 20 questions correctly. Their actual average score was closer to 13. The technology improved outcomes while inflating self-perception.
When the Dunning–Kruger Effect Disappears
The most intriguing theoretical contribution of the study lies in how AI altered the classic Dunning–Kruger effect. Under normal circumstances, people who perform poorly tend to overestimate themselves the most, while higher performers are relatively well calibrated.
With AI assistance, that pattern vanished.
Overconfidence became evenly distributed. Low and high performers alike overestimated their results by similar margins. AI acted as a leveling force, not only for performance, but for misjudgment.
To test whether this was a statistical fluke, the researchers conducted a second study with 452 participants. This time, one group completed the task with AI assistance, the other without. Participants were also told they would receive a financial bonus if their estimated score matched their actual score.
The findings replicated cleanly. The unaided group showed the traditional Dunning–Kruger pattern. The AI-assisted group again displayed uniform overestimation, even when accuracy was financially incentivized.
AI, it appears, fundamentally reshapes how people perceive their own competence.
Why AI Feels Like Understanding
Several psychological mechanisms may explain this effect. One is the “ illusion of explanatory depth,” the tendency to believe we understand something well until we are forced to explain it independently. Fluent AI explanations can amplify this illusion, making borrowed reasoning feel like personal insight.
Another factor is effort. People often use ease as a heuristic for success. When a task feels smooth, we assume we performed well. AI reduces cognitive friction, and the brain interprets that ease as evidence of mastery.
The result is a subtle decoupling: Performance improves, but the internal feedback loop that supports learning and self-correction weakens.
This is not an argument against AI. The data show that people perform better with assistance. The deeper concern is human agency—the ability to understand, own, and be accountable for one’s thinking and decisions.
When the boundary between personal performance and delegated performance becomes blurred, responsibility erodes quietly. People may learn less from success, overlook gaps in understanding, and carry inflated confidence into future decisions.
This matters in education , professional judgment, leadership , and any domain in which knowing the limits of one’s competence is as important as competence itself.
Takeaways Through the A-Frame
The A-Frame—Awareness, Appreciation, Acceptance, and Accountability—offers a psychologically grounded way to respond to this shift. It helps illuminate the causes and consequences of AI-assisted miscalibration across aspirations, emotions, thinking, and behavior.
Awareness: Recognizing the gap between assistance and authorship.
Practice: After using AI, explicitly identify which parts of the reasoning you could reproduce unaided.
Appreciation: Valuing AI as a partner, not a proxy.
Practice: Use AI to reveal gaps in your reasoning rather than to conceal them.
Acceptance: Acknowledging limits without outsourcing agency.
Practice: Normalize naming when AI was used, and where understanding remains partial.
Accountability: Owning judgments, decisions, and learning.
Practice: Treat AI outputs as drafts. Make final judgments only after articulating the logic yourself.
AI can make us better performers. Whether it also makes us better thinkers depends on whether we preserve the psychological conditions that sustain agency. AI is a means to an end. It is our task to maintain self-knowledge in its presence; that requires NI: natural intelligence.
Share this post Facebook Bluesky Linkedin Email
There was a problem adding your email address. Please try again.
By submitting your information you agree to the Psychology Today Terms & Conditions and Privacy Policy
Cornelia C. Walther, Ph.D., is an Associate Professor at Sunway University and a Wharton/University of Pennsylvania Fellow who researches hybrid intelligence and ProSocial Al.
Get the help you need from a therapist near you–a FREE service from Psychology Today.
This article is part of the Bringwise Psychology Journal — daily insights on human behavior, mental health, and personal growth.