In July, the University’s Generative Artificial Intelligence in Teaching and Learning Committee released a report finding that over half of undergraduate respondents reported using generative AI tools on a regular basis.
As part of the report’s feedback process, students raised questions about the negative cognitive effects of AI usage and the technology’s impacts on long-term learning.
The Herald spoke to researchers across different disciplines about their work in AI and education.
According to Megan Wei GS, a computer science doctoral candidate, when students are using large language models to complete assignments, they are “basically offloading the task to the LLM” instead of “learning the actual material in class.”
Cognitive offloading, as defined in a paper Wei co-authored this summer, refers to “delegating cognitive effort to another entity, such as an AI tool, and thus reducing a user’s effort and involvement in the process and output.”
Melanie Subbiah, a co-author of the paper and a postdoctoral associate at the LUNAR Lab, believes that the impact of AI on cognitive abilities depends on the user’s effort and process.
Subbiah noted that if users engage with AI properly, usage of the technology doesn’t necessarily have to entail complete cognitive offloading.
Since the quality of an AI model’s writing is “often 50-50,” users find themselves shifting their attention to judging the quality of the AI’s output, Subbiah said. When users are fully committed to a task, “there actually still is a lot of effort going on,” Subbiah said. The extent of cognitive offloading “depends (on) how much you’re shortcutting in a given instance,” she added.
Zoe Weissman ’28 said she has seen many of her peers taking these kinds of shortcuts in class.
“People have completely offshored their thinking skills to a machine because of convenience,” Weissman said. “I remember seeing someone in my anthropology class use ChatGPT to come up with talking points to raise his hand to talk about.”
According to Annaliese Paulson, a postdoctoral research associate at the Annenberg Institute, the effort students put into the learning process is directly tied to their cognitive development.
“Learning happens through experiences, and we have to have cognitively effortful experiences that reshape the brain in some ways,” she said. “If I engage in the process of learning a multiplication table, I’m having that effortful experience that reshapes how my brain operates.”
Paulson, who researches the role of college coursework on students, explained that there are two main perspectives in research regarding the effect of generative AI on learning.
One perspective argues that generative AI is “augmenting” learning and thereby cognitive abilities by giving students more access to feedback on their work, specifically for students who previously did not have access to resources such as tutoring, Paulson said.
This pattern was observed in the GAITL report, where a large percentage of students who responded to the survey reported using generative AI to learn new concepts.
But the second perspective argues that generative AI is “automating the process” of learning and limiting students’ cognitive growth. “We learn the best when we’re really actively constructing ideas and grappling with cognitive dissonance,” Paulson said.
Cat Gao ’27 said that while she understands how AI could be used to learn new concepts, she has also noticed that her classmates have been defaulting to using AI.
“It’s nice that there’s so much information at your fingertips, but I think it’s really impeded a lot of person-to-person interaction in classes, which is not good for (an individual’s) cognitive ability,” Gao said.
Assistant Professor of Entrepreneurship and Education Saloni Gupta is exploring an approach to evaluate how AI is impacting students’ cognitive abilities.
In her study, 4,000 students were given an open-ended question about how they would approach a given task. Then, students were asked to complete the task with access to AI. Gupta found that students who completed the initial strategy were “far more engaged with AI.”
“When you take a step back and think of a big picture of ‘what am I, what is my goal,’ then your engagement with AI will be specific and you will get much better learning out of it,” Gupta said.
Paulson said that the process of learning matters much more than a task’s results in college. “No freshman writing instructor is really the most concerned with the paper you produce so much as the process by which you produce it.”
“If we automate that process, we might be really harming students,” she added.
While Paulson is only in the preliminary stages of her research, she hypothesizes that the more skilled an individual is in a particular field, the more likely it is that generative AI will help them learn.
To help offset some of cognitive offloading’s negative effects, Subbiah suggested more honest AI disclosure policies.
“Cognitive offloading happens to the most extreme degree when it is like ‘I’m gonna get away with this,’” she said. “If we’re all more honest about what AI is doing versus not doing, I think it naturally shifts the focus onto what you are doing as a student beyond what AI is doing or even how you are using AI.”
Because there is still ambiguity surrounding the effect of AI on cognitive abilities, Paulson believes that before introducing generative AI in educational institutions, more research should be conducted.
“The lack of clear evidence on (AI’s) effects on cognitive ability is something I think should give us a little bit more pause before we really push” AI tools, Paulson said.
