Scaffolding and fading generative AI support shapes dependency and anxiety among undergraduate law students
This study challenges the assumption that restricting generative AI reduces dependency by demonstrating that in a Sri Lankan law cohort, gradually fading AI support increased academic anxiety—particularly among students with pre-existing dependency—without significantly lowering reliance, suggesting that structured integration rather than outright restriction is necessary to support equity and student well-being.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
In the modern classroom, a new kind of helper has arrived: artificial intelligence that can write essays, solve problems, and explain complex ideas in seconds. For universities, this tool presents a difficult choice. Many institutions have simply banned it, fearing that if students can get answers too easily, they will stop learning how to think for themselves. This approach assumes that taking the tool away will force students to become independent. However, learning science suggests that the way a tool is used matters more than whether it is present at all. Just as a child learning to walk needs a hand to hold at first, but must eventually let go to build strength, students using AI need a specific kind of guidance that starts with support and slowly fades away. If the support is removed too abruptly, or if the tool is never used in a structured way, students might not learn to stand on their own; they might simply become anxious and feel helpless without the machine.
This question of how to guide students through the use of artificial intelligence was the focus of a five-month study conducted with first-year law students in Sri Lanka. The researchers wanted to see if simply restricting access to AI would make students more independent, or if a carefully designed sequence of using the tool would work better. They worked with a group of 126 students in a Bachelor of Laws program, a field where learning to reason and analyze arguments is the core skill. The study did not just ask students what they thought; it watched how they actually behaved over time as the rules for using AI changed. The researchers divided the semester into three distinct phases to test different ways of interacting with the technology.
In the first month, the students were allowed to use AI however they wanted, with no special rules. This established a baseline of how they naturally used the tool. By the third month, the rules changed. Students were required to write their assignments completely on their own first. Only after they finished their work could they submit it to the AI, which was programmed to act as a strict reviewer. The AI would not give them answers or write new text for them; instead, it would challenge their reasoning, point out gaps in their logic, and ask them to think deeper. This phase was designed to let the students do the hard work of thinking while using the AI to check their understanding. In the final month, the support was pulled back even further. The AI was allowed only to flag where a student's reasoning was incomplete, without offering any explanations or suggestions on how to fix it. This final stage was a test of "fading," where the safety net was removed to see if the students could still function.
The results of this experiment challenged the common belief that banning AI is the best way to build independence. When the researchers measured the students' confidence in their own abilities, they found that the students felt most capable during the middle phase, when they had to do the work first and then use the AI to review it. This suggests that the act of struggling with a problem before getting help creates a sense of mastery. However, when the AI support was reduced in the final phase, the students' anxiety levels rose significantly. The study found that simply taking the tool away did not make them more confident; instead, it made them feel more stressed and more dependent on the tool than before.
Perhaps the most important discovery was about who suffered the most when the support was removed. The researchers found that students who had relied heavily on AI at the very beginning of the course were the ones who became the most anxious when the tool was faded out. For these students, the AI had become a crutch, and removing it did not force them to learn to walk; it just left them feeling unable to move. The study showed that a blanket ban on AI might actually hurt the very students who need the most help to develop their own skills. Instead of creating independence, a sudden restriction can deepen the gap between those who are ready to work alone and those who are not.
The researchers concluded that the design of the interaction is far more important than the mere presence of the tool. They found that using AI as a reviewer of independent work helped build genuine confidence, while removing support too quickly caused anxiety and a sense of helplessness. The study suggests that universities should not just decide whether to allow or ban AI, but should design a path where students learn to use the tool to check their own thinking before the tool is gradually taken away. This approach ensures that students build their own reasoning skills first, rather than becoming dependent on a machine that might one day be switched off. The findings offer a clear path forward for education: support students as they learn to think, and then slowly step back, rather than simply taking the tool away and hoping they figure it out on their own.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.