AI and Dunning-Kruger Effect

Scientists at Finland’s Aalto University have found that using artificial intelligence (AI) almost removes and reverses（逆转）the Dunning‑Kruger effect, a phenomenon in which people who aren’t very good at a given task are overconfident, while people with high ability tend to underestimate their skills. Their research showed that when using common chatbots to solve problems, everyone (regardless of their skill level) tended to put too much faith in the quality of the answers, with the most experienced AI users doing so the most.

In the study, scientists gave 500 subjects logical reasoning tasks, with half allowed to use ChatGPT. Both groups were later quizzed on both their AI literacy and how well they thought they performed, and promised extra rewards if they assessed their own performance accurately.

The reasons behind the findings are varied. Because AI users were usually satisfied with their answer after only one question or prompt（提示）, accepting the answer without further checking or confirmation, they can be said to have engaged in what Robin Welsch, an Aalto University computer scientist, calls “cognitive（认知的）offloading”— exploring the question with reduced reflection, and approaching it in a more “shallow” way.

Such less engagement in our own reasoning means we bypass the usual feedback process of critical thinking, which reduces our ability to assess our performance accurately. Even more revealing was the fact that we all overestimate our abilities when using AI, regardless of our intelligence, with the gap between high-and low‑skill users reducing. The study owed this to the fact that large language models(LLMs) help everyone perform better to some degree. Meanwhile, the Aalto team warned of several potential impacts as AI becomes more widespread.

Firstly, metacognitive accuracy overall might suffer. Without reflecting on results, error checking or deeper reasoning, we risk weakening our ability to source information reliably, the scientists said in the study. What’s more, the flattening of the Dunning‑Kruger effect will mean we’ll all continue to overestimate our abilities while using AI, with the more AI‑literate among us doing so even more— leading to an increased climate of miscalculated decision‑making and a drop in their skills.