Researchers and educators examine whether the convenience of generative AI is improving learning or allowing students to bypass the mental effort that makes knowledge stick.
The rapid spread of generative artificial intelligence (AI) into classrooms, universities and home study routines is prompting a new debate about what students actually gain when an AI system can produce an answer in seconds.
The technology has already demonstrated its ability to improve productivity. Students may use AI to organise information, explain difficult concepts, generate ideas and receive immediate assistance with assignments. Yet emerging research suggests that better-looking homework does not necessarily translate into stronger learning, raising questions about what happens when some of the most demanding parts of thinking are routinely handed over to machines.
A recent report by ABC News highlighted a study involving more than 26,000 students in Grades 7 to 12 in China. Guided by this report, students using generative AI achieved an 18% improvement in homework scores and completed their work about 30% faster. Conversely, their monthly examination scores declined by 20% within 6 months.
The contrast is significant because homework and examinations measure different aspects of academic performance. Completing an assignment efficiently may demonstrate that a student can obtain and assemble information. An examination, particularly one completed without technological assistance, places greater emphasis on memory, understanding, reasoning and the ability to work through a problem independently.
The findings, therefore, add weight to concerns about what researchers describe as ‘cognitive offloading’ — the process of transferring part of a mental task to an external tool.
Research published in ‘Computers in Human Behaviour: Artificial Humans’ examined AI-based cognitive offloading among 667 university students in mainland China. The researchers found that students’ perceptions of AI’s value were associated with their willingness to transfer both lower-order and higher-order cognitive activities to AI. Importantly, the study also found that AI evaluation capability affected these relationships, suggesting that the consequences of AI use depend partly on whether students can critically assess what the technology produces.
That distinction may become increasingly important as educational institutions attempt to establish sensible rules for AI-assisted learning. The issue is not simply whether students use AI, but ‘how much of the intellectual process remains theirs.’
New research from the Massachusetts Institute of Technology (MIT) also provides a more complicated picture. A study published by the MIT Media Lab compared unrestricted conversational AI with a Socratic system designed to guide students through hints and an adaptive tutoring system that responded to indicators of cognitive engagement. The experiment involved 50 participants learning about nuclear safety protocols.
Interestingly, the unrestricted chatbot produced higher immediate learning gains than the constrained approaches. However, the researchers cautioned that the result did not necessarily demonstrate more profound learning. Many participants using unrestricted AI adopted an answer-retrieval strategy, while the adaptive system generated significantly greater EEG-measured engagement.
The findings illustrate why the debate cannot easily be reduced to the argument that AI is either beneficial or harmful. An AI assistant may help a learner overcome an obstacle, but the educational value of that assistance can depend on whether the student is encouraged to understand, question and reconstruct the answer.
Longer-term research is now becoming particularly important. A study published in September in the ‘Journal of Computer Assisted Learning’ examined generative AI and learning dynamics across an entire semester, considering knowledge gain, motivation, cognitive load, critical thinking and reflective AI use. The researchers noted that much earlier research had concentrated on individual outcomes rather than examining how these factors interact over time.
This growing body of evidence is arriving as educators reconsider traditional assignments. If a student may obtain a polished essay, computer program or research summary almost instantly, simply grading the finished product may provide less information about what that student has actually learned.
Some institutions are therefore exploring greater use of supervised classroom assignments, oral examinations, handwritten or closed-device assessments and activities requiring students to explain their reasoning. Such approaches do not necessarily reject AI, although they attempt to separate legitimate technological assistance from the fundamental educational objective of developing independent intellectual abilities.
The question is particularly relevant for younger students, whose learning habits besides academic foundations are still developing. Neuroscientists cited by ABC News were of the view that repeatedly outsourcing difficult mental work may affect the development of skills associated with critical thinking. At the same time, experts also stress that AI may support critical thinking when it is used deliberately rather than simply as an answer-generating shortcut.
For schools and universities, the challenge may therefore be less about banning AI than about designing learning environments in which students still have to struggle productively with difficult questions.
Generative AI can explain a complicated subject, challenge an argument, provide alternative examples or identify weaknesses in a student’s reasoning. Used in that manner, it becomes a learning partner rather than a substitute for learning.
The emerging research suggests that the crucial issue is not how quickly AI may complete a task. It is whether the human using it has acquired the knowledge and reasoning ability to understand the result, challenge it and eventually perform the underlying task independently.
As AI becomes an ordinary part of education, that distinction may shape the next generation of teaching and assessment — and determine whether technological convenience strengthens human learning or quietly replaces some of the mental effort through which learning occurs.
