Why the System Does Not Correct Itself
The loss of developmental practice is not reproduced only by an individual choosing a shortcut. The focus here is why classroom norms, deadlines, evaluation systems, and experience-level perspectives can make that shortcut difficult to refuse and difficult to recognize as a problem. Understanding those conditions matters because a remedy aimed only at learner discipline or mentor goodwill cannot repair a pathway whose incentives are collective.
The pattern has two connected parts. In education, AI use becomes normal enough that working without it can carry a competitive cost. In the workplace, seniors and juniors can occupy positions from which the same change looks fundamentally different. Together, these conditions allow the old pathway to weaken without a shared decision to remove it.
When resistance becomes costly
The junior participants in the software-engineering study did not describe GenAI use in university as a simple matter of preference. A student could decide to work independently, but if peers used AI to produce nearly perfect answers, the decision changed the student’s position in the grading system. Working slowly through mistakes no longer looked like a private learning choice. It looked like accepting a disadvantage while everyone else used a tool that the institution was not effectively regulating.
One participant described the pressure directly: when everyone else was using GPT and receiving high scores, independent work no longer felt like a free option. Another said that not using it was “just stupid.” A third described the feeling of falling behind. These accounts represent different emotions—loss of autonomy, strategic resignation, and anxiety—but the underlying condition is similar. The payoff from using AI is partly determined by what other students do.
That is the defining feature of a collective constraint. If one student studies without assistance while others use AI to finish comparable assignments, the independent student bears the time cost and may receive no corresponding recognition for the additional learning. Even a learner who understands that struggle could be useful has a reason to abandon it. The choice is not made in a vacuum; it is made inside a curve, a deadline, and a peer environment.
Time pressure intensifies the effect. A student may intend to work through a problem unaided, then turn to an AI tool when the deadline approaches. The immediate result is legible: the assignment is submitted. The avoided cost—a lost opportunity to discover what the student does not yet understand—is invisible. Because the evaluation often rewards the submitted product, the system can reinforce the shortcut even when the learner knows that the shortcut is educationally weaker.
Institutional silence adds another layer. The participants reported that instructors often appeared to know students were using ChatGPT but did not prohibit it. This does not show that every faculty member approved of the practice or that a ban would be effective. It shows a gap between the learning objective and the conditions under which the objective is assessed. If a course is meant to develop a capability that AI can perform on the student’s behalf, leaving the choice entirely to individuals may make that capability impossible to observe.
The result is not simply more efficient studying. The study describes a divergence between visible performance and underlying development. Two courses can produce similar grades while giving the student different kinds of growth. In one, the learner develops domain knowledge through effort and revision. In another, the learner mainly becomes more fluent at operating GPT.