AI and Your Brain's Behavior: Understanding the Impact on Learning
AI and How We Learn: A Behavioral Look
In the world of behavioral psychology, we often study how new tools and environments change human behavior. The rise of Artificial Intelligence (AI) tools like ChatGPT in schools and universities is a major new influence. Since its introduction in 2022, many educators have expressed clear anxiety about the behavioral shifts they observe in students. They suspect that relying too much on these AI tools for academic work might lead to what they call 'cognitive atrophy.' This is a significant behavioral concern because when AI provides ready answers, it can short-circuit the entire learning process, weakening the application and development of essential thinking and reasoning skills.
Two recent studies, one from the Massachusetts Institute of Technology (MIT) Media Lab and another from the University of Pennsylvania (UPenn), aimed to understand these behavioral and cognitive effects of AI on learning outcomes.
The Behavioral Impact of Over-Reliance on AI
From a behavioral perspective, when we are given an easy way out, we often take it. If AI consistently provides immediate solutions, it can discourage the natural human behavior of problem-solving and critical thinking. This can lead to a reliance behavior where students seek quick answers rather than engaging in the deeper cognitive processes required for true understanding. This over-reliance can reduce the mental effort involved in learning, potentially making our "thinking muscles" weaker over time.
The MIT Study: Memory and Brain Activity Behaviors
Researchers at MIT designed an experiment to observe these behavioral patterns with 54 participants, divided into three equal groups of eighteen students. Each group engaged in different essay-writing behaviors. One group was tasked with writing essays using only ChatGPT and was not allowed to use anything else. A second group used the search engine Google and was not permitted to use any LLMs. The third group had no digital assistance—neither LLMs nor search engines—and relied entirely on their own cognitive abilities for essay writing.
Each of the three groups completed three sessions under the same conditions, with electroencephalography (EEG) used to monitor and record their brain activity during the writing sessions. In a subsequent fourth session, the group tasks were reversed. Those who had used ChatGPT for the first three sessions were asked to write using only their brains, while the group that had previously written without any digital tools was directed to write their essays using ChatGPT. The results were evaluated by both human and AI judges.
⚠️ The Dangerous Face of AI
AI is no longer just helping—it's starting to behave like us. Learn how some AI models are now lying, threatening, and strategically deceiving their creators. This chilling behavior challenges the very safety of human-AI interaction.
🔎 Read the Full Article →
Key Behavioral Observations:
The MIT researchers found that while essays written with AI were more grammatically polished and properly structured, they often lacked the originality and creativity that was more evident in the essays written by the group with no digital assistance. This suggests a difference in creative expression behavior. A striking behavioral outcome was that essay writers who used LLMs could barely recall anything they had written when interviewed just minutes after task completion. This phenomenon has been termed 'cognitive alienation.'
The EEG data provided an explanation for this memory blankness: their brains were simply not effectively encoding information because they were not processing the content for learning. ChatGPT users showed weaker activity in parts of their brains connected to attention and critical thinking. These are crucial cognitive behaviors for learning. In contrast, those who relied on their own minds demonstrated ownership over their work. They developed the mental scaffolding required to write an essay, which involves undertaking a greater cognitive load, rather than resorting to the copy-paste behavior observed in most ChatGPT users. The test group limited to Google use showed a moderate degree of brain activity, less than those relying solely on their brains but more than ChatGPT users. The MIT finding suggests that even traditional search engines like Google require users to engage in substantial cognitive work, such as forming queries, evaluating sources, synthesizing information, and formulating ideas, indicating more active learning behaviors.
It's worth noting the limitations of the MIT study: the sample size was small, the task focused solely on essay writing, and the study lasted only four months. Additionally, it has not yet been peer-reviewed. Nevertheless, the study's findings point to a significant learning gap between AI users and those who rely on their memories and past learning.
The UPenn Study: Copying Behavior vs. Learning Behavior
The University of Pennsylvania conducted a randomized controlled trial (RCT) in a Turkish high school, involving nearly 1,000 students studying mathematics. Students were randomly assigned to one of three behavioral learning conditions over four class sessions. One group relied only on textbooks, serving as the control group. A second group used "GPT-base," a version that mimicked a standard ChatGPT interface. The third group used "GPT-Tutor," a version with learning-focused prompts and teacher-designed safeguards that guided students to answers without completely completing them. The research aimed to compare performance and learning retention across these groups, assessing both immediate task performance and knowledge retention when AI was removed.
Behavioral Outcomes:
The results were significant. In practical problems, students using GPT-base performed 48% better than those using textbooks. GPT-Tutor users scored an impressive 127% higher than those who only used textbooks. This shows an immediate performance boost behavior with AI. However, when students were tested without AI assistance, the outcome reversed. The GPT-based group performed 17% worse than the textbook group (control). The GPT-Tutor group, on the other hand, performed just as well as the control group.
The reason for this result reversal was a behavioral one: students using GPT-base frequently asked the AI for complete answers. They copied these answers without truly attempting to solve the problems themselves. Consequently, during the test, when AI was not present, they could not solve the problems on their own. There was also an "illusion of learning" behavior. The ChatGPT-based group thought they were improving. In post-test surveys, many expressed confidence in their performance. However, their sub-par performance data showed they were mistaken.
Cognitive Debt: How Our Brains Fall Behind
What unites these two studies, from a behavioral psychology viewpoint, is the fundamental idea that when people allow AI to perform the "heavy lifting" involved in learning, they incur what MIT researchers call 'cognitive debt'. This refers to the accumulation of thinking deficits over time that occurs when students offload their thinking and critical reasoning skills onto AI.
This is analogous to students who have experienced human math tutors completing all their homework, leading to perfect scores on their assignments throughout the school term. However, when taking tests in school, they struggle to answer questions because they have not genuinely engaged with and/or internalized the subject matter. Their homework 'behavior' was outsourced, leading to a deficit in their own learning behavior when independent cognitive effort was required.
The Behavioral Benefits of Effortful Learning
From a behavioral perspective, learning something new inherently involves discomfort and even failures. These experiences, however, are crucial for helping to erase cognitive debt. Metacognition, which fundamentally involves self-reflection on how one learns, only develops when students immerse themselves in the learning process. This includes behaviors such as taking notes during lectures and summarizing them and asking self-directed questions like, "Do I understand this?" AI tools can disrupt this vital behavioral process by providing immediate answers, sometimes inaccurately and without the necessary reflection. As a famous quote attributed to Euclid states, when asked by King Ptolemy I if there was an easier way to learn geometry, he replied, "There is no royal road to geometry." This emphasizes that there's no shortcut to the true behavioral acquisition of knowledge.
Shaping AI Use for Positive Learning Behaviors
Obviously, we cannot prohibit AI, and even if we could, it would be akin to "throwing the baby out with the bathwater." The real challenge lies in finding the 'sweet spot.' We want to utilize AI's potential to personalize learning for each student while ensuring it does not do the thinking for them. A clue on how to achieve this is provided by the UPenn study's experience with the GPT-Tutor interface. This demonstrated that when AI is designed to guide rather than simply provide answers, it encourages more active and beneficial learning behaviors.
Ultimately, the behavioral goal is to integrate AI as a tool that enhances, rather than replaces, our fundamental human cognitive and learning behaviors. It's about empowering the learner, not enabling cognitive shortcuts.
0 Comments