When Is It Safe to Stop?

Is it cheating to let AI generate design ideas for you? Where is the threshold of ethics here? How far do we allow it, and when is it “safe” to stop?

These questions are no longer for conferences. They are being asked quietly, every week, in classrooms across Bangladesh, often by students who are afraid to ask them aloud.

Last semester, a student pinned up a beautiful set of drawings. The renders were clean, the diagrams were graphically sound, the concept had a confident name. I asked a simple question: why does the building face west? The room went quiet. The student was neither lazy nor dishonest. The student had simply never been asked to stand behind a decision that someone, or something, else had made.

I used to wonder whether that silence was about the tool. I have started to think it may be more about distance. Technology might not create the gap between a student and a classroom so much as move into a gap that was already there.

Walk through a university corridor in Dhaka, and you may well see the most powerful creative tool ever released, open on a phone. Look closer, and it is at times being asked to brighten a photograph, or to write a caption in English the student does not fully understand. This may be less a failure of the student than of how the tool was introduced. A tool that could test a structural idea or compare ten site strategies is often used like a filter, perhaps because no one has shown what else it could be.

A deeper difficulty may be speed. AI seems to have arrived faster than any syllabus could respond, and many classrooms may still lack a shared understanding of disclosure, authorship, or acceptable help. In that space, each student may end up inventing an ethics alone, and each teacher improvising in the corridor. Undefined territory is rarely neutral: it may reward those least troubled by the question, while more conscientious students hide their use or quietly fall behind.

We often hear that students can no longer pay attention, and that anyone can focus for only ten minutes. The evidence behind that slogan seems thinner than we might expect. A 2016 review in Advances in Physiology Education traced the popular 10 to 15 minute rule and reported that the research behind it appears not to support it. The largest variation it found in attention was between teachers rather than between teaching formats.

When researchers actually probed students during live university lectures, minds wandered about a third of the time, but more than half of those episodes were intentional, and they did not grow as the lecture went on. Phone use behaved differently: it became more common as a lecture progressed and was linked to weaker learning. Even so, the effect is modest. A 2025 meta-analysis of 45 studies found a small negative association between smartphone use and academic performance, and a two-year study of 470 Danish university students suggests earlier estimates may have been inflated.

So students may not be simply unable to focus. Sometimes they might be deciding, in small ways, that this hour is not for them, and a phone may be what a student reaches for when a class has stopped reaching for them. This is my interpretation rather than a finding, though it may resonate with what many teachers see.

I do not think the teacher in this picture is a villain. Many of us, in my experience, were hired for what we know and trained far less in how to teach, then handed large classes, fixed rows of seats, and a syllabus to finish. In a study of 40 final-year English students at the University of Dhaka, the barriers to classroom interaction included seating arrangements, overcrowded rooms and teachers’ preference for lecturing over communicative activities.

Language may widen the distance further. A 2014 study of first-year students at five Bangladeshi universities found that English-medium instruction led some to see themselves as deficient and left them feeling excluded from classroom activity. When the language is foreign, the examples are foreign, and the teacher feels far away, a machine that translates and explains at midnight may feel less like a temptation than like the only helper in the room.

The evidence for what might help seems clearer than the evidence for the problem. A meta-analysis of 225 studies in science, technology, engineering and mathematics found average failure rates of 21.8 percent under active learning, against 33.8 percent under traditional lecturing. A randomized comparison in physics added a warning: students in the active classroom learned more but felt that they learned less, because effort feels like struggle. Teachers who add activity might find it useful to share this with their students.

For a two-hour class, research does not offer a magic interval, but some approaches seem worth trying: changing the activity from time to time, inviting students to recall and explain what they have just heard, and offering short breaks. In Bangladesh, a translanguaging intervention at a private university reported deeper engagement when Bangla text and topics from students’ own lives were brought into the lesson.

Four areas may deserve more attention than they currently receive. The first is space. A “smart” classroom that keeps rows of chairs facing one speaker may have digitised the old arrangement without changing it, and as an architect I wonder what a room invites a student to do. The second is language: where are the Bangla-language learning tools, and AI that serves a Bangla-medium student as well as an English-medium one? The third is the teacher: much of the debate centres on students, while teachers are often expected to help define the rules with little training or time. The fourth is watching. Attendance trackers and learning analytics raise a question we seldom ask: who is allowed to see what a student does with a tool, and for what purpose?

My answer, offered tentatively, has two halves. For a student, a useful signal to pause might be the moment one can no longer explain what is on one’s own page. Could I defend it? Did I say that I used the tool? Did I grow from it? For an institution, it might help to stop treating technology as the answer and ask what it is standing in for. One sign that a tool may have replaced the relationship is when the student no longer asks the teacher, and the teacher no longer asks the student.

Some steps might help, and none of them seems complicated. We could mark the process and not only the final image, and invite students to explain their work aloud. We might write plain guidelines on AI use together with students, early, before habits harden into assumptions. Institutions could support teachers with training in how to teach. Classes might open with a question from students’ own world, and rooms could be arranged so that people can see each other.

The next time a student pins up a flawless drawing, I will ask about the courtyard again. I hope the silence is shorter. Better still, I hope the answer begins with, “Because I walked the site.”