Co-written by Paul Surlis, Ava Gilmartin and Michael Hogan.
Imagine that a group of graduate students were given a complex task management problem Their mission is to develop a sustainable water management plan for a busy tourist city that balances environmental protection, economic growth, and increasing demand for scarce resources.
Working in small groups, they have ninety minutes to understand the problem, discuss possible solutions and agree on a plan. In a few minutes, several groups will start consulting with ChatGPT. At the end of the session, each group will present a polished proposal. Although the groups have never interacted with each other, many of their solutions are strikingly similar: in structure, in the assumptions they make, and in the possibilities they never learn. Nothing is copied. However, something important is missing.
This is the scene behind a recent study Romero (2025). As generative AI becomes an increasingly familiar part of higher education, the most important question may no longer be whether students should use AI at all. A more useful question is when should AI enter the learning process. Introducing AI too early can destroy it creativityThe debates and cognitive movements that universities are trying to cultivate?
Testing the impact of AI access time
To test whether time really matters, Romero (2025) conducted a simple but revealing experiment. Thirty-six graduate students were randomly assigned to one of three conditions while working in small groups on a water management problem. One group had access to ChatGPT from the start. The second group spent the first fifteen minutes working without AI, brainstorming ideas, questioning assumptions, discussing different points of view, and co-creating an initial solution, with AI then introduced as a tool for improvement. The third group completed the task completely without AI. Although this was a relatively small quasi-experimental study involving a single task, its design isolated one variable that is often overlooked in discussions of AI in education: time.
The results were amazing. Groups with immediate access to AI have consistently converged on similar solutions, demonstrating that early reliance on AI narrows the diversity of ideas and encourages familiar lines of thought rather than independent research. Conversely, students who grappled with the problem first generated broader ideas before using AI to test, expand, and refine their thinking. Interestingly, the groups that had never used artificial intelligence produced the most diverse solutions, although their work was also variable in quality. Romero (2025) derives from this an important design principle: artificial intelligence appears most valuable when entered as a means to an end rather than a starting point.
In Romero’s (2025) study, only thirty-six students completed a single collaborative task, meaning that the findings should be viewed as preliminary evidence rather than definitive evidence. Nevertheless, the sample is suggestive. By delaying AI, educators can protect a period of productive cognitive struggle before the technology begins to shape students’ thinking. This naturally raises a deeper question: Why should fifteen minutes without AI make such a difference?
AI leads to metacognitive laziness
Science and others. (2025) Offer a plausible explanation by exploring what happens in students’ thinking when AI becomes part of the learning process.
In a randomized experiment with 117 university students, participants completed a two-stage English reading and writing task before revising their work with one of four forms of support: ChatGPT, a human expert, a structured writing checklist, or no additional support. The researchers then used the observational data to map how students organized their learning during the revision process—what they observed, what they evaluated, when they stopped to reorient themselves.
The ChatGPT condition focused students’ self-directed learning on AI. The interactions were extensive. Essays improved. But compared to the human expert and checklist conditions, the ChatGPT group showed relatively fewer metacognitive processes—evaluation and orientation—that involve stepping back, assessing where you are, and deciding what to do next. Expert support created transitions between orientation and assessment that ChatGPT did not. And most importantly, when knowledge acquisition and transfer were measured not only by the completion of tasks, the advantage of the ChatGPT group disappeared.
This Fan et al. metacognitive laziness: not laziness in the everyday sense, but subtle shift self-regulatory thinking that reinforces learning. The connection with Romero is straightforward: the pursuit of artificial intelligence immediately runs the risk of missing a period of struggle and the real cognitive effort that precedes good ideas. Romero shows how valuable this can be in terms of creative diversity. Science and others. show how much it costs in terms of metacognitive development. Together, the studies point to the same educational principle: giving students time to think before introducing AI helps them retain both.
Creativity researchers have long recognized that good ideas rarely emerge when presented with a problem. Instead, they emerge through something closer to Wallace’s (1926) classic four-step sequence of preparation, incubation, comprehension, and verification. The incubation stage, where people continue to think without immediate answers, is where unexpected connections often begin to form. In Romero’s study, it can be understood as a delayed entry condition pedagogical try to protect the hatching area or it will never open.
From pedagogical time to personal habit
If delaying AI can help preserve the thinking that underpins creativity and learning, the next question is how educators should design it. Romero’s research cannot tell us whether fifteen minutes is enough or whether the same principle applies to different subjects, learning activities, or assessment formats. These questions remain open. Timing is just one design decision. Educators must also consider which tasks can truly benefit from AI, how assessment rewards thinking rather than smooth results, and how learning activities can foster metacognitive processes that AI might otherwise replace.
Ultimately, the most profound challenge is designing a learning environment in which students become the authors of their own thinking. Getting the timing right is one of the important steps. As AI becomes a permanent feature of education, keeping students empowered can be a much bigger challenge. However, agency cannot be indefinitely dependent on pedagogical structures that only delay access to AI. When students encounter AI outside of carefully designed classroom activities, they will need to regulate their use of these tools.
This means developing practical habits of good AI use: brainstorming before making suggestions, critically evaluating AI suggestions, and using AI to augment rather than replace one’s own thinking. In this sense, agency is not simple self regulationbut self-regulation is based on understanding how to sustain learning in the long term. As AI becomes a permanent feature of education, the quality of learning may increasingly depend on students being able to think before offering. But these habits do not appear by themselves. Teachers play a central role in educating them through thoughtful pedagogical design, helping students to gradually assert their agency even when external constraints no longer exist.
