Researchers examine whether generative AI may weaken some thinking skills

Generative AI chatbots can now write emails and computer code, translate, plan trips and suggest gifts from simple natural-language prompts. Their spread through schools, universities, workplaces and personal life has led researchers to ask whether relying on them could reduce some forms of human mental effort.

Recent studies cited in the report point to possible risks for memory, decision-making and critical thinking. One American-British study of 1,222 people, which is still under peer review, found that AI tools helped participants perform arithmetic and reading-comprehension exercises in the short term. But the study reported that, over time, participants did worse and were less willing to keep trying when the tools were not available.

Grace Liu, a Carnegie Mellon University doctoral student and the study's main author, said AI can remove learning opportunities because it can be used across many intellectual, reasoning and cognitive tasks. The report contrasted this with tools such as electronic calculators, which can help solve equations while still leaving much of the method and reasoning to the user.

A 2025 MIT study also found that students who used generative AI to write essays showed less critical thinking capability. Other research has described the issue as “cognitive offloading,” or, in stronger terms, “cognitive surrender.”

Johann Chevalere, a social and cognitive psychology researcher at France's CNRS institution, said people often try to save mental energy by using shortcuts. He said generative AI could strengthen that tendency, and that the brain may not maintain connections linked to activities that are never used.

In response to criticism, generative AI developers have started adding functions designed to encourage reflection. Some systems aimed mainly at students give hints or ask questions instead of immediately providing answers. Examples cited include OpenAI's ChatGPT study mode and Google's Gemini guided learning.

Microsoft said it had added warnings about possible mistakes to its Copilot models and reminders for users to check the information they receive. The company said the risk of excessive cognitive offloading is real, especially when AI automates tasks that are also important for skill development, and said users need training to use the tools correctly.

Researchers cited in the report said large-scale, long-term studies are still lacking, so the full effect of generative AI on human brains remains uncertain. Chevalere said people will need to use AI carefully and adapt to the technology as they have to earlier technological changes.