When generative artificial intelligence began showing up in schools a few years ago, much of the conversation focused on a familiar problem: cheating. Could teachers tell whether a student or ChatGPT had written an essay? What counted as plagiarism? Those were understandable, yet simple, questions.
Michigan schools are now confronting something harder.
This year, the Michigan Department of Education released new guidance encouraging districts to develop policies and practices for using AI responsibly. It asks schools to think about AI literacy, human oversight, privacy and how the technology can be used appropriately in teaching and learning.
That shift reflects what is already happening among students. A 2026 Michigan Virtual study found that 58.4% of surveyed students had used AI for schoolwork in the previous 30 days, while nearly a quarter said they had received no guidance about acceptable AI use. Students and teachers also agreed more about what constituted inappropriate use than about appropriate use. The question facing schools is no longer simply how to keep students from using AI improperly. It is what using it properly should look like.
Schools have been here before. Sort of.
I watched an earlier technological transformation unfold during my years as a high school teacher. Beginning around 2010, schools embraced personal technology with considerable enthusiasm. Laptops, tablets and eventually Chromebooks promised greater access to information, more engaging instruction and new ways for students to learn.
The technology has brought some real benefits, and research has found positive academic effects from one-to-one computing. What I remember most clearly, though, is something we didn't fully anticipate: the effect those devices could have on student attention.
Before every student had a screen in front of them, a distracted student might stare out the window, whisper to a friend or doodle in a notebook. Personal devices created a different level of temptation. Students could work on homework for another class, browse the internet or find any number of ways to occupy themselves while appearing to be engaged in the lesson in front of them. In a strange way, we had provided the distraction ourselves.
Some of that was probably unavoidable; those devices reflected the world students were entering. Looking back, though, I'm not sure we spent enough time considering how putting a personal device in front of every student might affect their ability to focus.
AI makes that earlier transition look almost straightforward. A laptop gave students another way to do their intellectual work. Generative AI can do some of that intellectual work for them.
That is much more consequential.
AI changes something more fundamental
Suppose I assign a student an essay. If AI corrects a few grammatical errors, has it interfered with learning? Probably not, assuming grammar isn't what I'm assessing. What if it suggests ways to strengthen an argument? Helps organize evidence? Proposes the thesis? Writes a paragraph? Produces the first draft?
Somewhere along that continuum, a tool that assists thinking can become a substitute for it. The challenge for schools is to identify the difference.
In a 2025 randomized study of 120 college students, those allowed unrestricted use of ChatGPT while studying scored an average of 57.5 percent on a surprise test 45 days later, compared with 68.5 percent for students who studied without AI. Researchers suggested that reducing effort during learning can sometimes make it harder for information to stick.
A small MIT study produced a similar warning. People writing essays with ChatGPT showed weaker neural connectivity than those writing without outside tools and had more difficulty recalling what they had written. The study was a preprint involving a small sample, so its findings are preliminary, but they raise questions about what may be lost when AI reduces intellectual effort.
Learning requires some degree of that effort. We learn to write partly by struggling to decide what we think and how to express it. Sometimes the difficulty isn't an obstacle to learning; it is the learning.
That doesn't mean every difficulty is valuable or that AI necessarily undermines education. A 2025 randomized study of college physics students, for example, found significant learning gains when AI was deliberately designed as a tutor rather than an answer machine.
What should students still have to do themselves?
Michigan's new guidance cannot answer that question for every classroom, nor does it try. The state is encouraging districts to develop their own approaches while keeping AI purposeful and maintaining human oversight. Some Michigan districts are already trying to answer those questions.
St. Joseph Public Schools has been developing a three-year AI integration project. One example offered by district leaders is revealing: students might first brainstorm an idea themselves and then use AI to help develop a thesis or bounce ideas around. That is very different from asking AI to write the essay.
Ann Arbor Public Schools has involved students directly in developing AI expectations, including a draft AI Bill of Rights. Neither example tells us where every school should draw the line. They show that Michigan schools are beginning to wrestle with where that line belongs.
There probably isn't one rule that can tell a fifth-grade teacher, a high school English teacher and an Advanced Placement physics teacher exactly when AI supports learning and when it replaces it. Each will have to consider their specific learners, but all must ask a broader question:
At what point does assistance become substitution?
The lesson from the earlier technology transition isn't that schools should resist this one. It is that adopting a technology is not the same thing as deciding its educational purpose.
