About the study
It is clear from existing research that students are using AI to various degrees inside and outside the classroom. It is also clear that different professors are taking different approaches, often within the same program. Some require it, some encourage it, some allow it for various specific items, and some ban it completely. What is not clear is how students feel about their AI use. Specifically, are students using AI more than they would prefer, less than they would prefer, or in line with their preference?
This research measures that gap, what students attribute it to, and how it relates to why they are using AI in the first place. It also asks students one direct question about learning: across their CS coursework last semester, did using AI help their learning or hurt it? That answer sits beside the gap, so the study can say whether students who used AI more than they would have chosen think it cost them something, and whether students who used it less think the same. Students who did not use AI at all are asked the matching question about not using it. The survey also asks whether the amount a student used felt like their own decision, whether they often did not know if using it was the right thing to do, and whether they wished someone would just tell them how much to use it. Because these tools are everywhere, students are making high-stakes moral and academic decisions about them every day, for every kind of coursework, with little guidance, and these questions let students say what that has been like.
The survey begins by asking which CS courses students took last semester and what each of those courses said about AI, whether it required it, encouraged it, permitted it, discouraged it, or prohibited it, and because a student can answer that they are not sure, the study can also report how often students did not know what a course’s rules were. The questions that follow are about one of those courses. A student who took one CS course answers about it, and a student who took more than one picks the course where AI came up the most, whether in the rules, in their own use, or in how they felt about using it, with the course they spent the most time on as the fallback. Every later question names that course, so that students answer about a specific course they actually took instead of averaging across several classes or thinking about their personal use, and the study sets each student’s gap beside the rules of the very course they answered about. The results can say which courses the responses came from, although they do not report findings for individual courses. Finally, one separate question asks how often students used AI to translate English course material into another language, and it is reported on its own instead of being folded into the gap, because translation is a common use that is different from the others.
These are three separate things, and the study keeps them separate on purpose. The gap is about behavior, meaning how much a student uses AI compared to how much they would choose to. The attributions are about causes, or what the student says is driving the difference. The motivation is about quality, and it asks whether a student’s use feels like something they have chosen and endorse, something they are complying with, or something they feel they have to do to avoid feeling bad about themselves. A student can be using AI more than they would prefer for reasons that are entirely their own and entirely endorsed, and another student can be using it exactly as much as they prefer while feeling constantly pressured. Measuring the gap alone would miss that difference, which is why the study measures all three and looks at how they relate.
When students use AI more than they would prefer, that often looks like pressure. It can come from instructors and course requirements, from peers, from beliefs about what employers will expect, or from dependence, the sense that they cannot do the work without the tool. Dependence comes in two forms. The first is a student who avoids getting help from people and leans on AI instead, and the second is a student who feels that they could no longer do the work on their own. People worry most about the second form, because it sounds like a skill that has worn away. However, the survey can only record what students believe about themselves, and it cannot tell us whether a skill was lost, was never built in the first place, or is still there and simply feels out of reach.
Guilt and shame run through this. Students feel guilty about using AI, and some of them also feel guilty about not using it, and those two point in opposite directions. However, shame is different from guilt, because guilt is about the thing you did and shame is about who that makes you. Students describe worrying that using AI means they are not a real programmer. Recent work with computing students found that these feelings mostly do not stop students from using AI, but they do change whether students hide it.
When students use AI less than they would prefer, that often looks like constraint. That constraint can come from access and cost, from whether the tool actually fits the task, from worry about being accused of cheating, or from whether the student knows how to get the help they want from the tool.
But the direction is not fixed by the cause. A professor who bans AI is a source of pressure that pushes use down instead of up. A student who cannot afford the paid tier faces a constraint, but a student who finds the free version good enough does not. The study does not assume which direction any given cause pushes, and instead looks at which causes students point to when they use AI more than they would choose and which they point to when they use it less.
Through all of this, convenience runs as a frictionless pull for students who might not use the tool otherwise. Convenience is the easy on-ramp to using AI, and it is tempting to assume that dependence is simply where that road ends. Unfortunately, one survey about one semester cannot show one turning into the other, so the study measures both and reports how they show up together.
After the survey, small groups of students chosen for their survey patterns talk through what happened, in sessions led by graduate students who are not their instructors. They are asked about a time one course felt different from another, who they were comfortable telling about their AI use, how they see other students’ use compared with their own, and how they decided how much to use it, whether there was a moment when they decided or it just happened. They are also asked about a time using AI changed what they ended up understanding, for better or for worse, and whether they think they would have learned more, less, or about the same if there had been no rules about AI. The study does not measure learning. It records what students believe their use did to it, and the two are not the same thing, which is part of why the question is worth asking.