I’ve previously introduced the phrase borrowed competence. The idea that AI lends us the appearance of capability we have not built and do not own, and that the loan eventually comes due.
If borrowed competence were obvious to the people borrowing, they would notice and self-correct. The reason it matters is that the borrower cannot tell. Recently, researchers at Aalto University in Finland published a study that tested this directly. Two experiments, with 246 and 452 participants, on logical reasoning problems from the Law School Admissions Test (LSAT). Half of each group used ChatGPT. Half worked alone.
The people using ChatGPT did better on the reasoning. That part was expected. The unexpected finding was about how accurately people could judge their own performance afterwards.
Most of you will know the Dunning-Kruger effect. People who are bad at something are confidently bad at it. People who are good at something underestimate themselves. It has held up across studies for twenty-five years. In the Aalto study, with AI use, the pattern disappeared. Not weakened. Disappeared. Everyone overestimated their performance, regardless of actual ability.
And here is the part that genuinely worried me. The people who reported the highest AI literacy were the most overconfident of all.
Technical sophistication did not protect against miscalibration. It made it worse. The researchers called this the reverse Dunning-Kruger effect. The published paper is aptly titled, “AI makes you smarter but none the wiser“. This is the cognitive mechanism that makes borrowed competence invisible. The borrower cannot tell because the cues that ordinarily calibrate confidence are gone. Most participants asked AI one question, accepted the answer, and moved on. They never developed the feedback that would tell them whether the answer was any good. For anyone thinking about how their workplace uses AI, this should change the question. The training that matters is not how to prompt better. It is how to know when you do not know.
A much harder skill to teach, and I think the one most likely to separate organisations that thrive from those that quietly accumulate borrowed competence until the loans come due. As a trainer, I find this finding both vindicating and alarming. Vindicating because it confirms why the slow, awkward work of gaining competence matters.
The struggle is the calibration. Alarming because the people most likely to skip the struggle are the people most likely to think they have not skipped anything. Next week, the Boston Consulting Group study that puts numbers on what borrowed competence costs in real workplace tasks.
This piece used Claude (Anthropic) for literature review, structural drafting, and editorial pushback. The thesis is mine, the sources are checked, the words have been earned.