Ten minutes. It is the time it takes for AI to have a negative effect on our ability to reason and solve problems, or at least that is what they have concluded in a new study in which they have measured how the use of AI assistants not only improves immediate performance, but also reduces persistence and worsens performance when we do not have access to AI.
The study. Researchers from Carnegie Mellon, MIT, UCLA, and Oxford have published a randomized, controlled experiment that measures the impact of using AI on the ability to solve problems independently. In total, more than 1,200 people participated in three different experiments. The researchers’ conclusion goes in the direction of what we have seen in other previous studies: using AI enhances our productivity, but It has a cognitive cost.
The experiment. A first experiment was carried out with 354 participants in which they had to solve twelve simple fractions. Some of the participants had a side panel with an AI assistant (GPT 5) that they could use to solve the operations. The curious thing came when their access to the chatbot was removed and they had to answer three more questions without the help of the AI. The result was that people who had used AI made more mistakes in their answers than the control group.
Confirming results. The researchers did a second experiment in which they duplicated the participants (667) and did a pretest to measure the level. In addition, they added a “placebo” side panel (without AI) to the control group participants, so that there were no interface differences. The results again showed that people who used AI failed more than the control group.
There was a third experiment in which reading comprehension problems were asked with 201 participants and the same thing happened again: when AI was removed, that group performed the worst.
The key nuance. There is an important detail of the study and that is that they measured how the participants used AI. 61% used it to give them answers directly, while others used it to give them clues or clarifications. The results of this second group were more similar to those of the control group. On the other hand, those who asked for AI solutions as they were failed much more when it was withdrawn. This suggests what we have said above: the negative effect of AI on our cognition. It depends largely on how we use it. Copying answers without questioning is not the same as using them as support in the cognitive process.
The new silly box. The fear that technology makes us stupid is not something that has arisen with AI, it happened with the calculator, it has happened with television, with video games and it is happening with cell phones. Although there are studies that point in that direction, there is no clear evidence that technology damages our cognition. However, it is also true that until now we had not had access to technology to which we could delegate all our thinking.
Cover image | Xataka
GIPHY App Key not set. Please check settings