Recent experiments have shown evidence that relying on artificial intelligence to learn new material tends to reduce the amount of information students remember after a few weeks. The findings suggest that while these tools may speed up initial learning, they may actually weaken the deep mental processing required for long-term retention of knowledge. The study was published in the journal Open social sciences and humanities.
Generative artificial intelligence refers to computer programs that can create text, images, or other media in response to a user’s prompts. These systems can answer complex questions, integrate vast amounts of information, and write essays in seconds. Because of this convenience, millions of college students use these programs to help with their coursework.
Andre Barcaoui, professor of business administration at the Federal University of Rio de Janeiro, conducted the study to find out how this technology affects memory. He wanted to test whether getting quick answers from a chatbot would be detrimental to long-term learning.
When students delegate mental tasks to external tools, psychologists call it cognitive offloading. Previously, this meant using a calculator for calculations or a notebook to memorize shopping lists. Generative AI platforms take this a step further by performing the actual thinking, analysis, and problem-solving.
“What motivated me the most was my undergraduate students,” Barkaoui told SciPost. “In recent years, I have felt that there has been a loss of awareness and interest among students. In teaching and teaching with AI, I have seen the full potential of this tool and consider myself passionate about the beneficial and responsible use of AI.”
“However, I began to feel that there was a risk, as teenagers’ interest in reading declines over the years, and with it their creativity, critical thinking, interpretation of texts, and even writing skills and expression. I also spoke with professors from other countries, and I don’t think this is a phenomenon specific to Brazil. So I decided to investigate this issue and see if other comparable studies pointing in the same direction corroborated it.”
Barkaoui based his research on the concept of desirable hardship. This psychological principle suggests that when students face some productive struggle, they actually learn better. Activities like actively trying to remember facts or trying to understand difficult concepts cause the brain to build strong memory pathways.
Automated chatbots have the potential to remove these necessary hurdles by providing sophisticated answers instantly. Without mental friction, students may have weaker memory retention. Memory consolidation is a biological process by which fragile new memories become stable and safely stored in the brain over time.
To explore this dynamic, Barcaui recruited 120 business administration students. Participants included 68 men and 52 women between the ages of 18 and 24. None of the students had any formal training in computer science or machine learning.
Students were randomly divided into two equal groups of 60 students each. Both groups received the same assignment and had to explore basic artificial intelligence concepts. Topics include ethics, social impact, and technological foundations.
The assignment required students to spend up to two weeks researching an assigned topic and preparing a 10-minute presentation. The first group was instructed to use a specific AI chatbot known as ChatGPT as a learning aid. They freely interacted with the program to explain concepts, generate examples, and structure presentations.
The second group used traditional learning methods. These students were banned from using automated chatbots. Instead, they relied on course notes, academic databases, and standard Internet search engines.
After the initial research period, participants gave a presentation to a small group of their peers. After that, the students had a normal academic life for one and a half months. This delay allowed researchers to test natural memory decline over a realistic academic period.
Forty-five days after the first study phase, researchers surprised participants with a memory test. Eighty-five students returned to take this 20-question multiple-choice exam. This test was specifically designed to measure how well students understand concepts, rather than simply their ability to memorize definitions.
Students who used traditional learning methods performed better. On average, traditional learners answered 68.5% of the questions correctly. In contrast, students who learned using the chatbot answered only 57.5% of the questions correctly.
The negative impact of chatbots was most noticeable when students were learning highly technical topics. The software also reduced memory for less specialized topics, such as ethics and social studies, but the difference between the two groups was not large. This pattern suggests that productive struggle is especially important when mastering complex or structurally difficult content.
Barcaui also tracked the amount of time students spent preparing their presentations. Automated chatbots allow students to complete tasks much faster. The technology-assisted group spent an average of 3.2 hours on the task, while the traditional group spent 5.8 hours.
Even when the data was adjusted to account for this difference in study time, traditional learners still scored higher on the final test. This detail shows that learning disabilities are not simply the result of less time spent with the material. In fact, the quality of study time seemed to be declining.
Traditional learners may have spent extra time rereading the material, testing themselves, and manually connecting different ideas. These traditional study habits give your brain natural practice in retrieving information. Chatbot users are probably spending their time writing prompts and reading the responses generated, and while they feel productive, they are not engaging their brain’s search network.
The researchers also found that the results did not change depending on students’ previous experience with chatbots. Frequent users of the technology performed just as poorly on retention tests as users who were relatively new to the technology. Evidence suggests that software can make students feel like they understand the material better than they actually do, creating an illusion of competence.
Mr. Barcaui emphasized three main points.
“Productivity is not a substitute for ability: There is a big difference between delivering a work and understanding the creative process. Indiscriminate use of AI can create an ‘illusion of competence’ where individuals obtain results without developing the synapses necessary to independently reproduce that reasoning. ”
“Atrophy of important “muscles”: Just as continuing to use a calculator can diminish your mental arithmetic skills, relying on AI to create and interpret text can atrophy your overall and critical thinking abilities. Without the mental “friction” of reading and writing, we lose the ability to articulate complex ideas and question information. ”
“AI as co-pilot, not autopilot: The main lesson is that AI should be used to augment human capabilities (extend reach), not replace human capabilities (eliminate effort). Human value will increasingly shift from the ability to execute to the ability to ask the right questions and critically curate the data generated. ”
As with all research, this study has some limitations. Nearly 30% of the original 120 participants did not return for a surprise test 45 days later. Although dropout rates were similar in both groups, this loss of data may slightly skew the final numbers.
The sample was limited to business students from one university in Brazil. Students from different disciplines, such as engineering and humanities, may interact with software differently.
Future research should examine how student outcomes change with different guidelines for technology use. Scientists can test scenarios where students attempt to solve problems on their own before asking the chatbot for feedback. Researchers can also use digital tracking software to accurately measure how students allocate their time while studying.
The study, “ChatGPT as a cognitive crutch: Evidence from a randomized controlled trial on knowledge retention,” was authored by André Barcaui.
