Joseph Hennessey, MSEd, is an educator who has served in classroom and building-level leadership roles, most recently in central Maine. He was the 2018 Piscataquis County Teacher of the Year, the 2019 Maine Teacher of the Year, the 2020 National University Teacher of the Year for Maine and is currently an American Service Fellow at the Harvard Kennedy School of Government.

Toward the end of the last school year, I sat with a middle school teacher and her student in the classroom to address misuse of AI on an important assignment. Were Google and Gemini (Google’s AI assistant) really different forms of sources? How much “assistance” from AI qualified as too much, at this developmental level, in this content area, at this juncture in history?

Our district’s acceptable use policy had been provisionally updated, but we had not yet had time to share the changes with students. The teacher did not want to penalize the student unfairly; the student did not want to receive an “incomplete” for the assignment; and I did not want to establish a precedent that would be difficult to maintain.

In the end, we used this meeting as an educational opportunity for the student, and the teacher and I revised plans to deliver even more digital literacy to all middle school students in the coming year — this was a positive result in the face of complex circumstances.

***

Public education has been grappling with the AI issue for several years. As Kate Meyer and Nicole Davis reminded me in their recent book “EduProtocol Field Guide: AI Literacy Edition,” however, we are only limited in our use of AI by the constraints of our imaginations.

There are many reasons one could be optimistic about AI’s application in public schools, but students’ and schools’ relationships to this emerging technology are still very unclear, as evidenced in my anecdote above.

What’s undeniable, though, is that artificial intelligence affects the environment much more than it should. Schools should thus treat AI as what it is: a very expensive, very precious commodity that is not to be misused, squandered or discarded. Casual use of AI in education at this stage is careless use of AI that we can ill afford.

In 2024, the Berkeley Lab projected that AI “hyperscale” data centers would consume 98% of all water used by the sector by 2028. New Yorker contributor Steven Witt noted in his 2025 article that each new AI data center built in the United States is expected to use one gigawatt of electricity per year, which is more than the entire annual energy use of Philadelphia.

According to Noman Bashir of MIT and his colleagues, the global electricity consumption of data centers is expected to reach 800 terawatt hours in 2026, which is more than three times the annual energy use of California.

And while this is a small portion of the overall environmental demands of the United States and globe (according to CBS News, more water is needed to produce one hamburger than to conduct hundreds of AI queries), it is very important to consider. AI and its natural resource usage is part of a critical context concerning energy behaviors. And if there are popular movements to limit red meat consumption (i.e., Meatless Mondays addressing the hamburger problem), a similar standard should be applied to AI in general and in education in particular.

Schools already teach about natural resource conservation and climate change. AI must be a distinct part of that discourse precisely because we do not know how large it will be. When the student described above had used AI so frequently and inefficiently to complete a single assignment, they likely had not considered their environmental impact as much as their desire to produce excellent work.

Our colleagues and communities expect that schools will embrace AI in a manner that is net productive and equips our students with the knowledge and skills to confront a rapidly evolving society. These are large tasks along an uncertain path. In response, we should tread conscientiously on that path, not fearfully.

In accordance with the federal and various worthy state-level AI frameworks on offer, three questions guide my use:

1. Am I producing something concrete that will make sense to others and mean something to others?
2. Am I producing something durable that will stand the test of time rather than become immediately irrelevant?
3. Am I producing something reusable that can be taken offline and into the real world with necessarily needing more AI?

If the AI task satisfies those criteria for me, then the product will satisfy the different human forward frameworks. For students, this means putting in the real work of learning, whether AI is used in the beginning, middle or end stages of a task. No machine or algorithm should be doing the thinking for you.

For teachers, this means generating the core materials (lesson plans, supplemental handouts, differentiated readings), and then using your own professional judgment to create replicable learning experiences for students. Teachers are still the experts; AI is a tool.

For administrators, this means crafting queries carefully the first time and working with colleagues to polish the message. Use efficient prompting to get it right the first time and do the important refining work offline.

And for all of us in public education, this means not demanding pristine, symmetrical perfection from student work. We should be far more interested in work that looks like a young person made it with an authentic combination of time, effort and risk-taking. Let’s let children express themselves through their assignments, even if it involves occasional smudges, curved lines or wrinkled edges.

***

The student whom I met with in that classroom in June had dutifully participated in class, had worked with their peers every day and had wanted to do exceptionally well when they turned to Gemini.

Yet what they learned in that unit overall is of far more importance to me than whether or not the end product sounded like it was written by a generic college student. That polish was worth far less, pedagogically, to their teacher or our school than what it cost to produce environmentally.

AI will become more efficient, and scale will help, but good habits should be developed now. As educators, we must model more effectively as the technology and its potential emerges. Let’s use AI more conscientiously to preserve cognition and the environment at the same time.