Universities Need Comprehensive AI Policies Now

The Professor Who Watched Her Students Cheat With AI and Decided It Was Her Fault

A few months into the 2023 spring semester, a professor at a Hong Kong university noticed something strange in her inbox. Students she had never seen before were emailing her with perfectly written queries. The prose was polished. The grammar was flawless. The tone was professional. And the content was wrong.

The emails had been generated by ChatGPT. The students were outsourcing basic communication to a machine. The professor, Cecilia Ka Yuk Chan, did what any responsible educator would do. She started a study.

What she found, after surveying 457 students and 180 teachers across multiple disciplines at Hong Kong universities, was not what she expected. The problem was not that students were using AI. The problem was that universities had no idea what to do about it. And neither did the students. Neither did the teachers. The institutions had been caught flat footed by a technology that had been improving in plain sight for years.

Chan's proposed solution, published in the International Journal of Educational Technology in Higher Education and now cited over 1,100 times, is called the AI Ecological Education Policy Framework (Chan, 2023). It is the most comprehensive attempt yet to answer a question that every university on earth is now facing: What do you do when a machine can write a better essay than most of your students?

The short answer, according to Chan's data, is that universities need three things at once. A pedagogical plan. A governance structure. And operational support. Most institutions have none of these. Some have one. Almost none have all three.

This is not a future problem. This is a now problem. And the window for getting it right is closing fast.

What 637 People Actually Told Cecilia Chan About AI

Chan's study is unusual because it does not just ask students what they think about AI. It asks teachers and administrators the same questions, then compares the answers. The gap between the two groups is where the real story lives.

The students were across 10 disciplines including arts, business, education, engineering, law, medicine, and science. The teachers came from the same fields. Both groups answered quantitative surveys and participated in qualitative interviews and focus groups. Chan wanted to know three things: how people were using AI, how they felt about it, and what they thought universities should do.

The answers were not clean. They were contradictory. And that contradiction is exactly what makes Chan's framework necessary.

The Students Were Already Using AI. The Teachers Were Not.

Most students in the study reported using text generative AI tools like ChatGPT for their coursework (Chan, 2023). They used it to brainstorm ideas, to improve their writing, to translate concepts they did not understand, and sometimes to generate entire assignments. The teachers, by contrast, were largely aware that AI existed but had not incorporated it into their teaching. They had not been trained on it. They did not have guidelines for it. They were making it up as they went along.

This creates a dangerous asymmetry. The students are operating in a world where AI is normal. The teachers are operating in a world where AI is a disruption. Neither group has a shared understanding of what constitutes acceptable use. And without that shared understanding, every interaction becomes a negotiation.

Both Groups Agreed on One Thing: They Needed Rules

When Chan asked participants what universities should do, the most common answer was not "ban AI" or "embrace AI." It was "give us clear policies." Students wanted to know what they were allowed to do without getting in trouble. Teachers wanted to know what they were supposed to police. Administrators wanted to know how to enforce anything at all (Chan, 2023).

This sounds obvious. It is not. Most universities have not produced any AI policy. Some have produced vague statements about "academic integrity" that do not mention AI at all. A few have banned AI outright, which is about as effective as banning calculators in the 1970s. The students in Chan's study were not asking for permission to cheat. They were asking for clarity. They wanted to know where the line was so they could stay on the right side of it.

The Disagreement That Matters Most

Here is where Chan's data gets uncomfortable. Students and teachers disagreed sharply on one question: Should AI be allowed in assessment?

Most students said yes. Most teachers said no (Chan, 2023). This is not a minor disagreement. It is a fundamental conflict about what education is for. Students see AI as a tool that makes their work better. Teachers see AI as a tool that makes their assessment meaningless. Both are right. And neither has a framework for resolving the tension.

Chan's framework is designed to address exactly this kind of conflict. It does not pick a side. It creates a structure where both perspectives can be accommodated, but only if the institution is willing to do the hard work of building that structure.

The Three Dimensional Framework That Actually Makes Sense

Chan's AI Ecological Education Policy Framework has three dimensions: Pedagogical, Governance, and Operational (Chan, 2023). Each dimension addresses a different set of problems. Each dimension requires a different set of actions. And each dimension is useless without the other two.

The Pedagogical Dimension: What Are We Actually Teaching?

This is the dimension that most universities think they have covered. They have not.

The Pedagogical dimension asks a simple question: How does AI change what students need to learn and how we teach it? Chan's framework proposes that universities should not just teach students how to use AI. They should teach students when to use AI, when not to use AI, and how to evaluate what AI produces (Chan, 2023).

This is harder than it sounds. It requires redesigning assignments so that they test skills AI cannot replicate. It requires teaching critical thinking as a specific, assessable skill rather than a vague aspiration. It requires faculty to become fluent enough in AI to know when a student is using it well versus using it to bypass learning.

Chan's data shows that most teachers are not prepared for this. They have not been trained. They have not been given time. They have not been given resources. The Pedagogical dimension is not a suggestion. It is a demand that universities invest in their faculty the way they invest in their technology.

The Governance Dimension: Who Makes the Rules and How Do We Enforce Them?

This is the dimension that makes everyone uncomfortable. Because it requires universities to make choices they have been avoiding.

The Governance dimension addresses privacy, security, accountability, and academic integrity (Chan, 2023). It asks: What data does the AI collect about students? Who owns that data? What happens when a student uses AI to cheat? What happens when a teacher accuses a student of cheating but cannot prove it?

Chan's research reveals that universities are paralyzed on these questions. They do not have policies for data privacy related to AI. They do not have clear definitions of what constitutes AI assisted cheating versus AI supported learning. They do not have appeals processes for students who are accused of AI violations.

The Governance dimension forces universities to make these decisions. It does not tell them what to decide. It tells them they must decide something. The worst option, Chan's data suggests, is to leave these questions unanswered. Because unanswered questions get answered by whoever shouts loudest. And that is not how good policy gets made.

The Operational Dimension: Can the Infrastructure Handle It?

This is the dimension that gets ignored until something breaks.

The Operational dimension addresses infrastructure, training, and support (Chan, 2023). It asks: Do students have reliable internet access? Do they have devices that can run AI tools? Do faculty have training on how to use AI in their teaching? Does the university have IT support for AI related issues?

Chan's study found that these operational questions are often treated as afterthoughts. Universities announce AI policies without ensuring that students and faculty can actually implement them. They mandate AI literacy without providing training. They ban AI tools without understanding how those tools are already embedded in students' workflows.

The Operational dimension is not glamorous. It is about bandwidth, hardware, and help desks. But without it, the other two dimensions are just words on a page.

What the Research Does Not Prove

Chan's framework is based on a large sample across multiple disciplines. But it has limits that matter.

First, the study was conducted in Hong Kong. The cultural context of higher education in Hong Kong is not identical to the cultural context in North America, Europe, or Africa. Students in Hong Kong may have different attitudes toward authority, different expectations about academic integrity, and different access to technology. Chan's framework is a starting point, not a universal solution.

Second, the study was conducted in 2023. AI technology is moving faster than academic publishing. The tools available to students when Chan collected her data are already obsolete. The ethical questions have only gotten more complex.

Third, Chan's framework is a proposal, not a tested intervention. It has not been implemented and evaluated at scale. It is a carefully reasoned argument based on solid data, but it is not a proven cure. Universities that adopt it should treat it as a hypothesis to be tested, not a recipe to be followed.

These limitations do not weaken Chan's argument. They strengthen it. Because they remind us that this is an ongoing process, not a one time fix. The framework is not the end of the conversation. It is the beginning.

The Hardest Question No One Is Asking

Chan's data reveals something that most discussions of AI in education avoid. The real problem is not that students will use AI to cheat. The real problem is that AI makes visible something that has always been true: our assessments are not measuring what we think they are measuring.

When a student can generate a passable essay in 30 seconds using ChatGPT, the essay is no longer a valid test of writing ability. But it was never a pure test of writing ability. It was always a test of time management, access to resources, stress tolerance, and privilege. AI just makes the flaws obvious.

Chan's framework does not solve this problem. It creates the conditions for solving it. By forcing universities to articulate what they are actually trying to teach and how they will know if students have learned it, the framework opens the door to a deeper conversation about what education is for.

That conversation is overdue. AI is not going to wait for us to catch up.

What This Actually Means

• ▸Universities must create AI policies now, even if they are imperfect. The cost of having no policy is higher than the cost of having a bad policy that gets revised. Students and faculty need guidance, not perfection.

• ▸The policy must address three dimensions simultaneously: what to teach (Pedagogical), how to enforce rules (Governance), and what infrastructure to provide (Operational). A policy that covers only one dimension will fail.

• ▸Faculty training is not optional. Universities cannot expect teachers to navigate AI without resources, time, and support. The Pedagogical dimension requires investment in human capital, not just technology.

• ▸Assessment redesign is the most urgent task. If an assignment can be completed by AI, it is not testing what you think it is testing. Universities need to identify which skills AI cannot replicate and design assessments around those skills.

• ▸Students want clear rules. Chan's data shows that students are not asking for permission to cheat. They are asking for boundaries. Give them boundaries, and most will stay within them. Leave them uncertain, and they will make their own rules.