Generative AI has evolved from an academic curiosity to a routine part of daily study. Our 2026 Student Generative Artificial Intelligence Survey confirmed that 95% of UK undergraduates now use AI in some capacity, and 94% use it for graded work. This transformation prompts an important question: What occurs to the thinking required for a task when students allow an AI system to perform, solve or explain that task? Here we present our synthesis of recent experimental and survey data on AI dependence in education, based on EEG results from MIT, a randomized trial at Harvard, a field study with almost 1000 high school students in Turkey and with survey data from China and Peru. Our results supported a striking distinction. Use of open-ended AI systems resulted in reduced scores for critical thinking, and with poorer performance on examinations when access to AI was eliminated and decreased cognitive activity while composing text. Al tutors that refrain from providing direct answers and require students to make an effort first result in learning improvements that surpass regular classroom instruction. The document states that the construction of an Al engagement, rather than the existence of Al, determines whether a complex tool develops a skill or supplants it, and ends with possible strategies to differentiate the two.
Velugeti et al. (Wed,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: