The truth about AI’s environmental impact: Finding your ethical stance as an educator

Is AI using a bottle of water every time you make a query? Are you a bad person if you use it in your classroom? Should schools ban it entirely—or go all-in?

If you’ve felt confused or conflicted about AI ethics, this conversation is for you.

I’m sitting down with Dr. Karen Boyd, an AI ethics consultant who works with schools and nonprofits, to get real answers about the environmental impact of AI—and to talk through the much bigger ethical questions educators are wrestling with.
In this episode, we cover:

• The truth about AI’s water and energy use (spoiler: Netflix is way worse)
• Why “just don’t use it” isn’t realistic anymore in 2026
• The spectrum from AI enthusiasts to conscientious objectors—and why most of us are somewhere in the middle
• Six strategic stances beyond refusing: wait and see, constrain, compensate, rethink the work, and shape the ecosystem
• How to identify which specific values feel threatened to you (intellectual property? authenticity? effort and craft?)
• Practical ways schools can build ethical AI policies through knowledge sharing instead of top-down rules
• Different ways to use AI beyond shortcuts: as a thought partner, adversary, assistant, or accessibility tool
• Why understanding how AI works matters even if you choose not to use it

Karen offers a nuanced, inclusive approach that validates different perspectives while helping educators move from “this feels icky” to “here’s exactly what bothers me and what I can do about it.”This isn’t about convincing you AI is good or bad. It’s about having the informed, thoughtful conversation we all need to be having.

Listen to episode 343 below, or subscribe in your podcast app

Sponsored by 40 Hour AI

So Karen, can you begin by sharing your personal stance and philosophy on AI use?

I think if you were to distill it down as simply as you can, you can put everyone on a spectrum from enthusiast, like, “Oh my gosh, this is the best thing ever, “I can’t wait to show you what I did,” “This is saving me so much time every day”, to like conscientious objectors, “This trains on artists’ work without its permission, therefore I will not use it for anything”.

And I think that the distribution is probably kind of lumpy, like some people sort of on either end. But there’s a lot of us in the middle, like who are concerned and aware of the issues and trying to find ways to use it that bring real value, that benefit, they save time, they improve the output, but don’t conflict with our sincerely held values. And I think that I’m, yeah, I’m somewhere in the middle there.

Debunking the Water Use Myths

So let’s start with the environmental piece because this is a piece that I feel the most conflicted on in terms of just not being very well informed. I just don’t feel like I have the truth. I see rumors out there like, Oh, every time you do a query with ChatGPT, you’ve used a whole bottle of water. And so I’m wondering, is that true? Help us understand what’s really going on with data centers, water use, and the environmental justice piece.

Great. I’m so excited to talk about this. I also hear that bottle of water quote, and fortunately for all of us, it is not true. To start with, it’s a misreading of the original. The amount of water that’s used per query changes quite a lot depending on where you are, what the energy sourcing is, and what the water supply looks like there. But across the US, on average, it’s 29.6 queries per bottle of water. So already we’re quite a bit lower than that.

And it’s also worth thinking about what that estimate includes. The reason why water is used in AI at all or in data centers, to your point, at all, there are two big ways. One of them is cooling the data center, right? There are bunches of computers in there. They’re not like my laptop, where they’ve got all the space in the world, they’re all crunched together, they don’t have monitors, they get hot because they’re all in there together. They’re working really hard. And so they’ll circulate water in there to cool the data center down.

And the other type of water use that’s included in these estimates is hydropower, which is actually a nice, clean use of energy sourcing. And so we don’t necessarily—it helps us get a better picture of overall water use. So that’s why the study includes it. But it’s not something we probably want to penalize people for doing.

So that’s another, that’s one of the ways that I think we can help with the water use component, right? Using water doesn’t—the water doesn’t disappear and go out of the water cycle at all. But there are some harms that can happen.

One is if you take potable water, water we can drink, and make it not potable anymore. So then you’d have to expend energy to make it drinkable again.

The other thing that can happen is that water can get trapped. Like if you buy a bottle of water and bury it in your yard, keep it sealed, that water is out of the water cycle now, it’s trapped there. For long-term use, you actually kind of want that water to leave the data center at some point, so it goes back into the cycle.

And then the third issue is more of this environmental justice problem, where if you have a data center in a place that doesn’t have a lot of people nearby, suddenly you’ve got this enormous water consumer in a place where the water table, the water infrastructure, isn’t prepared for that. And the cost of utilities, both energy and water, for residents around there can go up really high.

And so that’s why we have people advocating for different rates for data centers or requiring the data centers to contribute to expanding that infrastructure, which I think is pretty reasonable. Like, you’re the one creating all this demand. Why are we going to then spread out the cost of that infrastructure across all of these people in this rural area who had no real say in whether this thing came here?

The alternative would be air conditioning, which is more energy use, which is probably worse in terms of environmental impact, because our grid in the US is not all wind and solar and hydro. There’s a lot of coal and all this other stuff in there.

Okay, so we’re sort of debunking this idea that it’s terrible in terms of water. What about the electricity? And how much worse is this than anything else?

Because I’m thinking about how we stream Netflix for hours and hours and hours. If you think about it, every American household is probably streaming from multiple screens for multiple hours a day. And yet I’ve never heard a statistic about like, Well, did you know how bad that is for the environment? And yet I imagine it is using a lot of fuel to do that. A lot of carbon emissions are coming from that. Is that right?

You’re exactly right. I think when AI starts a conversation about the environmental impacts of data centers, that is fantastic. But it is not just, not only an AI issue, but it’s barely an AI issue compared to Netflix. The water use, electricity use, and all the other data center impacts of Netflix dwarf AI by a long shot.

And the studies that are looking at AI use, I mentioned that most of that water use is actually for electricity generation and not for data center cooling. The other way you can separate is looking at training versus use. When you and I sit down to do our query, it’s like a milliliter or two. Most of that water use is the water used in training spread out over all that time. So if we can reduce the impact of the training, that’s where the real leverage is. Not necessarily you and I changing our individual use.

So the first step is like zoom out. This is a data center problem, and this is a systemic problem that can be solved by companies or the government more than it can be solved by us.

I do not want to say don’t talk about this. I think this is really important, and I’m glad that AI is starting the conversation.

But there are four overall environmental problems with data centers:

The first and second is — well, we can talk about water and electricity together because they overlap a ton because a lot of the water use is for electricity. That resource use can be changed, right? You can, at least in my area, you can decide I’m willing to pay a little bit more to get electricity from a sustainable source. Companies could be doing that as well. And then by way of expanding the renewable energy available on the grid, which would be great.

The other impacts are the location of the data centers has an impact on what’s around it. I mentioned the utility costs. That’s true for water and electricity. If it’s not already a really heavily populated area, they can really strain the grid and raise energy prices. But they also use generators as backup and those generators use gasoline or some fuel that causes pollution to the areas around there. That doesn’t affect all of us, but it does affect those locations. That’s particularly in the U.S.. Northern Virginia has so many data centers, incredible amounts of data centers. But also Dallas, the Chicagoland area, Silicon Valley, Phoenix, Atlanta.

And then the fourth one we talk about even less because it’s not here is rare earth element mining. In order to make all of these computers, we need rare earth elements. They’re mostly found in China, but there are deposits elsewhere. And the mining for those is very—is a very dirty process. They’re found co-located with radioactive materials and other things that are not great for the people mining them or the ecosystem around those mines. And I think it’s easy to overlook the impact on people who aren’t living in our country or our neighbors. But like that is a real both an environmental problem and an environmental justice issue.

I made this YouTube video expanding on the selective outrage around AI and the environment

Teaching Students About AI Ethics

Part of the reason that I brought you here to talk about this is because I’ve gotten this question so many times from teachers, particularly in response to a unit that I have, it’s a three-lesson mini unit. It’s for middle and high school classrooms and it’s called Stay Human, Protect Your Brain Power in an AI World. And the goal of it is really to support teachers in guiding students to think about the impact of outsourcing their thinking, and their perspectives, and their self-expression to AI.

So that’s a related issue. It’s a separate topic though. And this unit does not cover the environmental impact. And I hear from teachers who say this is fantastic. This really helps my kids think about what they are losing when they are allowing AI to do the thinking and writing for them. But they’re like, if I want my students to be informed about how AI usage also impacts other people, not just themselves. And so they’ll say things like how can I teach my students to care about the environment if I’m using and I’m advocating for this tool that has these environmental costs. And I think it’s a really valid perspective, and I really wanted to have you with your background knowledge give a more deeply informed answer. So how would you respond to that in terms of what you think students need to understand about the ethical dimensions of AI and the best way to go about modeling and teaching that?

First of all, I want to say I love the idea of protecting your brain power. I think it’s critical to talk about different ways that we can use to scaffold our work, not replace our work. I love that framing. So yeah, I want to read that.

I think that for teachers talking to students about the environmental impact, one really interesting angle might be talking about the counterfactual. This is a term I didn’t learn when I was in school. I think I maybe didn’t even hear it until I was getting my doctorate. But it’s really increasingly important, I think. And the idea is, if I wasn’t using this, what would I be doing instead? What am I using this to replace? Or what’s the alternative state of the world that we’re comparing it to?

So in the case of large language model use, people talk a lot about, let’s compare the water use of an LLM or a chatbot query. So talking to Claude or ChatGPT or Copilot to ask a question, compared to a Google query. And that is a really good comparison if the thing you’re doing is asking like, What’s the capital of Mongolia? Because it’s a question that has a single answer. By the way, Google is better suited for that because it’s not predicting the next word, it’s literally querying. That’s why we call that a query and chatbot submission a prompt. Because Google is actually getting the information for you. But leaving that aside, if you are—

Just to point that out, though, just to bring that to light, you’re right, because a lot of times people now are just ChatGPTing it and they’re not really thinking about it. And that’s a good distinction for students to learn to make. When do I need a search query versus when do I need a large language model prompt?

Yeah, yeah. And if you’re just going to get a single fact-finding question, it’s worth considering. It is comparing one LLM prompt to one Google search. But if it’s doing something like here’s the list of sources I want to use for my essay. What sources am I missing? That’s going to be several. That would take me several Google queries to figure out because I’m going to search a bunch of different combinations of keywords. I’m going to look at different types of literature, or maybe psychology versus sociology. There’s a bunch of different Google searches, or I should say web searches that are involved in that. But I might be able to accomplish that with one LLM prompt. And so now that comparison isn’t perfect.

At my work, I will often, if I run into an IT problem, I used to just call IT on a video call, which to your point, video is a huge data file and streaming video uses way more resources and data center sort of concentration than an LLM query does. So if I can, instead of calling my dear friends in IT on a video call and going back and forth for 20 minutes, if I can solve that by chatting back and forth with an LLM, that’d have to be a really long chat to make up for that. So thinking about what we’re comparing it to might be a really interesting way, specifically for the environmental side of things.

For other ethics, broadly, I think there are lots of trade-offs and nuances. Just like in environmental protection, I consult with nonprofits, including school districts and schools, about AI use and specifically trying to make sure their use aligns with their ethics. And I’ve got a list of 13 ethics that people talk to me about all the time, and I wrote a chapter on each of them so I could talk about any of them for an hour.

But just to give you a taste of what some of those other ones are, in addition to sustainability, you’ve got privacy ethics, equity and justice, information integrity and information literacy, ownership and intellectual property, authenticity and trust. There’s a bunch of them. But none of these chapters are like, none of what is contained in there is don’t use it because this is always bad. Right? There’s nuance, there’s trade-offs, there’s ways of mitigating risk. There are ways that AI can help support each of those values as well.

And a lot of that depends on how you’re using it. What specifically are you doing? Are you trying to use it as a shortcut, or are you trying to use it as a scaffold? But also what the context is. I think we sort of intuitively can imagine, okay, a teacher using AI to create an image that they can’t draw for themselves for a slide to make a point is different and will have a different reaction than a nonprofit that supports the arts using AI images in their marketing material. Right. The context of how it is used really affects humans’ reaction to how much of a conflict it feels like.

I don’t mind if my bank has my financial information. Of course, they do, because I know what they’re doing with it. If Google Maps had my banking information—oh, no thank you. And just as much, I don’t want my bank to know my location. But Google Maps has to know, right? So it’s that contextual part of it, I think really matters, and that context-dependent side of things can make, I think, thinking through the ethics of this really interesting and instructive both for us as users and for students, I think.

So it might be interesting to have a project around, okay, what is the use case, the values involved, and what context is it in? And having them think through that and that practice will serve them well as they go out into the world and use this at work, presumably.

Oh, I love that. My mind is just spinning with all the different possibilities for how that could be done. Let me talk a little bit more about your work with consulting with schools. Because you mentioned at the beginning that people’s stance, their personal stance on AI is on a spectrum. And I think a spectrum is such a useful way to think about all kinds of different things. So I really like that framing here.

And we can have districts, school districts, and schools on one end refusing to use it, not letting teachers use it, not wanting students to use it, to others who are like, we are going to completely redo the way we do education. We are going to reimagine education, we’re going to go AI-first, like the Alpha schools are, and have students spending lots of time being tutored by AI bots. And then they’re going to spend the rest of the day with hands-on group work, cooperative learning, getting outside movement, all that kind of thing. So we’ve got like this huge spectrum.

And you’ve also mentioned that some schools are just sort of drifting, not making the decision at all. So can you walk us through what some of these stances look like within the educational system?

No time to finish reading?
Download the ep and listen on the go!

Beyond Refusing: Six Strategic Stances on AI Use

Definitely. I like to think about these stances because it reveals a lot of other options besides the natural one, which is refuse. Right? This is violating a value of mine. I cannot use it. This has an environmental impact. Or this is using content from artists without their permission, therefore, it is bad to use it. And I do think, by the way, that there are contexts in which refusing to use it is absolutely the right call. So I name that as a stance because I do think that’s one of the ways that we can respond. That makes a lot of sense in some cases, but it isn’t the only one.

Another option is to wait and see. I can look at my environment, and I know the school doesn’t have the right policies or infrastructure. My students aren’t ready, I’m not ready. Or I don’t have access to a technology that has the right ethical safeguards for me to feel comfortable letting my students use it. But I’m going to keep paying attention and wait until these circumstances change. And I’ve got a point at which I’m going to stop waiting and seeing and dive in in some other cautious way. That is a more active stance than just, I will not use this.

Another option is to constrain. So I could say students are allowed to use AI to have it critique the outline of their essay, but they’re not allowed to use it to write the essay. Or as teachers, we could decide we’re okay using this to create writing prompts or translate high school materials into middle school reading level, or something like that, but we’re not comfortable using it for grading. If we have that conversation about really concrete, specific tasks that we’re okay with it being used for, or even writing prompts like in a prompt library so that we can use it all consistently, constraint can be a really cool option.

Another option is compensate. This is inspired by people buying carbon offsets for their flights. We could buy carbon offsets for our AI use. We could also, maybe if we’re saving a lot of time at the administrative level, we could use the time and money that we are saving to reinvest in a value that we’re worried our AI use might be compromising. We could commission an artist to paint a mural in our foyer, or in the district headquarters, or something. That kind of reinvesting that time, I think, is a really underexplored way of handling values conflicts with AI.

Another is to rethink the work. When you identify an area where there’s a conflict between the technology use and your values, you can take a step back and rethink what you’re doing. So I do not know how long it takes to write a curriculum, but I’m going to assume it’s more than a day, maybe some number of weeks or months for a human to write some curriculum. Now, ChatGPT Pro, even if you’re using the longest thinking model, probably takes half an hour maybe.

But the first thing I imagine that might come to your mind, and your listeners’ minds, is, Well, there’s some quality problems here. I care about doing a good job for my students. I’m worried about the quality. There is a conflict here. What you could do is say, “Okay, we’re going to rethink our process”. Instead of just replacing a human who takes months or weeks with an AI that takes half an hour and doing that a thousand times in a row and now we have a thousand curricula, which I don’t know that we need, we could say, all right, we’re going to have it generate and then we’re going to create a quality assurance process after that. Right. We’re going to change our whole workflow around mitigating risks to that value that’s being threatened there.

And then there’s shape the ecosystem. We talked a little bit about this in the environmental impact part where the companies can make a huge difference. Way more of a difference than we can make.

Yes.

But when we’re organized in an organization like a school or a school district or, you know, if you’re in a large campus or a large district, you’re a big customer for these vendors, you can call them and say, you know, hey, we’re considering using your product. But before we do that, I’d really like to know, where do you get your energy from? What’s your energy sourcing policy? How are you mitigating this value that we are worried about?

Now that isn’t because the salesperson is going to go to the CEO and go, “Oh my gosh, we’ve got to change our energy sourcing”. But if enough people are doing that, they’re going to realize we need to have an answer when people ask this question, and the answer probably shouldn’t be burning a lot of coal or whatever the thing is. And so there’s the option of leveraging your power as a customer, or leveraging political power is also an option. If you’ve got a lot of trust with parents, you can talk about, listen, we’re really interested in teaching your kids how to use AI. But here are concerns. You can have that conversation with your community, potentially.

But the thing that you don’t want to do, although I think is pretty common, is to your point we mentioned earlier, is drift. Right? You notice an ethical conflict, but you go, Look, I can’t really afford to refuse to use this. I don’t have the power to fix it, so I’m just going to ignore it. And that leads to, in addition to actual threats to those values, creates a lot of cognitive dissonance for the user. Like, that doesn’t feel good. Every time we use this, we feel guilty. That’s not helping anybody, is certainly not helping us. And so I recommend any other stance but that one.

So as I think about what your own personal recommendations would be for schools around AI, I’m definitely hearing a lot of nuance. I’m hearing a lot of context, a lot of situational types of things, which is harder. Bureaucracies and institutions like to have clear do this, don’t do that kinds of things. And just to give rules from on high, we’re not doing this. That’s the end of it. You know, no kind of discussion. I’m wondering what kinds of things have you seen in schools that seem to really be working well, that seem to be making a positive difference in involving these stakeholders in the conversation in a way that you think works well? Like what kind of recommended guidelines or policies have you seen?

Yeah, I think that a big—a lot of the conflict that comes in any organization when it comes to AI use is having really different perspectives. Like I mentioned, I’m kind of in the center. I’m somewhere in between an enthusiast and a conscientious objector. But when you’re looking at an organization, you’ve got all of those people in there. And in a school, there are also teachers and parents, and there’s hierarchy in the education system. There’s a lot of different perspectives, and it will be very hard to—I mean, you’ve got to learn how to have a discussion across those boundaries. And that, to me is like the first hurdle to overcome.

So I was recently in an organization where I had an executive who ran a team. The executive is a super enthusiast. This is the best. The first thing that comes up when anyone has a question of him, like, you know, I’m stuck. What should I do? And he’s like, ask ChatGPT, right? And now he has power in this team because he is the boss.

But there’s also a person on this team who’s a conscientious objector, right? In her mind, she’s like, she has a PhD. I learned how to do this process. The process of science is sacred. Why are we putting a black box in the middle of this process? This is so offensive. Absolutely not. Will not use it.

And what I recommended that they do was once a week at their team meeting, just talk about, how did you use AI this week? What that led to was a lot of knowledge sharing. And you might think this might lead to the enthusiast just talking all about how he’s using AI all day and everyone else going, Okay. But in fact, when I talked to the enthusiast, he thought, like, I learned so much every week from this. Because all these people are coming toward the technology with really different perspectives, right? They’ve got lots of little tricks that they use to make sure that they’re being thorough.

A good one here is “Ask me a question one at a time to get all the information you need to do this task”. That’s really useful if your task needs a lot of context, but you don’t know what context it needs. That wasn’t something that he had learned how to do yet. Because for him, he’s already dove in. He’s all the way in there, right? But to hear someone say, “You know what, I read this somewhere. I started to use it. It really helped me feel more comfortable”. He tried it, and he’s like, not only does this help them feel more comfortable, that’s great. Also, I get a better answer when I do it. So it helps the enthusiast, too.

And when I spoke to the conscientious objector, she told me, you know what? I still don’t love this, but she’s no longer having a really strong emotional reaction when someone brings it up, like, “Why don’t you use ChatGPT?” And now she is able to see it in a more like, there’s a lot of ways of using this technology, and she’s found some ways to use it that she feels comfortable with. And that saved her some time.

One of those was that they needed a bio for the website, and she was like, I don’t want to do this, right? Like, I’ve got a CV, I’ve got a LinkedIn, I’ve got all this. I don’t want to write another bio for another audience. So she threw a bunch of information into one of these chatbots and said, write me a bio in this format. Like, here are some examples of what other bios look like. Please do it. And she’s like, you know what? I was not looking forward to that task, and now it’s done, right?

And that, in addition to, I think helping all of them improve the quality of their work, it helped them improve their ability to work together across really different perspectives on AI. And I know that isn’t an answer that includes nice black and white rules, so I will give you an answer for that also.

One thing that I found, especially if you’ve got a lot of people who are very concerned or a lot of conflicting interests in an organization, which I might expect to see in education, is to create a very conservative policy that has, whitelisted tasks or whitelisted workflows where, like, you can use it in here and if you’re doing it for anything else, come talk to us first, right? So we can work together to create a workflow that stays within our ethical boundaries, our quality standards, and so that we can monitor it. Right? We can come and talk to you next week and say, “How’s that going?”

We want to make sure that you’re not losing time, or we want to make sure that these particular outcomes are not happening. So we can set up a way of tracking that, like some kind of measurement and then talk about it in a week and say, Okay, how many times did something bad happen? How much good happened? And learn in context, absorb that context into the policy. And then, you know, if enough people want to do it, we can build out a standard operating procedure and a policy and make that all happen in a way that we could learn safely the first time. It’s a tiny, tiny little pilot that can help people, help organizations build policies that are based on their real context and needs, not something that they found online or something that seems right in theory, the school board thinks but they’re, you know, they’re not in the classroom and they’re not—like that’s maybe not the best. It’s too much distance. And this is a way of building really contextually grounded policy.

What I really love about everything that you’re sharing here and your approaches is how very inclusive it is of different perspectives. Because that’s something that’s really important to me. Whenever I’m doing training with teachers on AI, there’s always the conscientious objectors. And at the very beginning, I usually ask some sort of question that kind of opens up the dialogue and I can tell that when they share their perspective that they think this thing is rotting kids’ brains, and we should not be using it, I validate that. And they seem surprised. I’m like, I’m actually not here to try to convince you to use it or that it’s a good thing. I have no way of actually knowing if it’s a good thing. Like, I’m like you Karen, like I’m kind of in the middle. I have objections, but also I do use it, and I do see use cases for it. And I think an informed usage is the most important thing.

What we don’t want to be doing is just, you know, just open it up and use it and have no idea what the impact is or the implications or how it’s working or what the cost is to society. I think we just need to be educated about it and have these different perspectives. And so I really like that you’re doing that.

I personally struggle to have a good response to the conscientious objectors at this point now that we’re moving into 2026, because when ChatGPT was released, which was November 2022, and it was still possible in 2023 to be a conscientious objector. Like I’m not using this. Now it’s in literally everything. I open up Instagram, and it’s wanting me to—I can’t just do a regular search. It’s an AI-assisted search. You know, the search engines all include it every single time. I mean I was trying to pay a water bill for the house the other day, and there was an AI assistant and I’m just like, I don’t know if it’s possible to even use the internet anymore without using AI. So is it reasonable for an individual to decide they do not want to use AI? What would that look like when it’s embedded in basically every single online tool and app now, and so much of our lives are online?

Can You Really Avoid AI in 2026?

Just like you, I would love to validate the desire to abstain is so reasonable. Whether it is possible I do think is a different question. I think that one point to make here is that generative AI isn’t the only type of AI. We’ve been using AI in all kinds of things for a long time.

When you go online shopping, they’re not presenting the socks to you in alphabetical order or in the order where they were, like chronological order. They’re using a recommender system to say people with similar shopping patterns to this person prefer these socks. So I’m going to put those first because that’s our best shot at selling socks to them. Not because they care that much about your socks, frankly.

So could you drive to the store and purchase socks instead to avoid AI? Well, not in my car. I just got a new-to-me 2017 car, and it has lane assist and automatic parking, which scares me. I don’t do that. I’m not ready to do that. I don’t do that yet. But you know, the smart cruise control and the warnings before you’re gonna run into stuff, that’s all AI-driven stuff for the most part.

I never even thought about that, and I use it all the time. I love it. I love the smart cruise control.

Yeah, me too. It’s so—I thought I didn’t need it. I also thought I didn’t need a backup camera. What was I talking about?

Right.

I love a backup camera. Yeah, big surprising upgrade to me. And similarly your email, if you’re using any popular email system, it has a junk filter. It used to be those junk filters were looking for strings of text like ‘Nigerian Prince’ and throwing out those strings. But when they developed an AI system to use it, that’s a classifier, not a recommender system. When they developed a classifier for junk mail, it got way better. If you’re old enough and used to use the old ones, they were bad, and now they’re so much better.

So could you avoid that by sending a physical letter to someone? You cannot. The United States Postal Service is one of the oldest adopters of AI. They used it to—originally it was optical character recognition for the zip code 92101. It’s all handwriting. It’s not written in some kind of computer readable thing. So they used optical character recognition, which is a type of image classifier, to identify what the zip code is, to automatically sort the mail.

So it would be very hard to completely avoid using all AI. Can you avoid using generative AI? Which I think is where a lot of people—I will say there are ethical concerns with all of those, with recommender systems, classifiers, clustering algorithms, depending on how they are used. We’ve got the uses in policing, which are terrifying. So what I’m not saying is generative AI is worse, just that it is different and new. And sudden. Right? Like the rest of this neural network, machine learning based stuff has been—I mean, the original paper was written in 1945. My dad was building these on punch cards in the 70s. They’ve gotten way better in the last 10 years. But 10 years is a lot different than to your point, two years or three now almost, hey, 2026, look at that.

So it would be very difficult to avoid all AI. It would be pretty difficult to avoid generative AI. It’s a little easier at the moment. You can log out of Instagram and LinkedIn and you can turn off the thing in Word and you’ve got to do some work to avoid it, but it would be possible.

The risk is that if productivity expectations in school and work and the rest of our lives expand to assume that we are using generative AI to take some of that labor off our plates, that we will then start falling behind. And that risk also exists for our students, right? If they enter into a workforce where generative AI use is very, very common and their productivity expectations assume that they’re using this stuff, they’re going to be way behind if they haven’t learned how to use it by now.

And the other point I want to make is that even if you choose not to use it, it’s still really worth knowing how it works. Because your students, job applicants, if you’re in a job where you’re hiring people, your competitors, your administration, your lawmakers, perhaps like a lot of other people around you, are going to be using this stuff. And so understanding how it works, how to recognize its weaknesses and strengths, will help you navigate a world in which tons of people are using this thing.

For the teachers who are feeling conflicted about this, which I think is probably most people who are listening, I think most reasonable people feel conflicted on some level, and they’re sort of torn between these benefits and the concerns, what would you say to them? What’s a helpful framework to think about your own personal philosophy towards this, and when you want to use it and when you don’t?

Finding Your Personal AI Philosophy

The first thing I would say is that whether you use it or you don’t use it, someone out there is going to judge you about it. Right? Someone’s going to be like, you’re leaving all this productivity on the table. I can’t believe you’re not using all the resources at your disposal to make sure that kids are getting taught the best or whatever. And there’s also going to be people who are like, if you’re using AI, it means you’re bad at your job and a bad person. So you’re not going to please everyone.

My encouragement would be to learn enough about the technology, how it works, and think enough about your own personal ethics as how they apply to the technology so you can at least please yourself. That would be my first step there.

And I also want to encourage people to not think about it as a tug of war between productivity and ethics. It’s not a zero-sum game where, in order to get productivity I have to make a sacrifice on ethics. Think about it more as a Cartesian coordinate plane. Right? There are two factors here: productivity and ethics. And there’s a way to do both of them worse. There’s also a way to do both of them better. And so looking for ways to, in your own life and in an organization, practices and policies that encourage a values-driven AI culture, that knowledge sharing is a big part of that.

Another thing is to think through your own values. And where is the conflict happening? I think a lot of people, for example, look at AI art and go, “Ugh, like that’s icky. I don’t like that”. And that could come from a lot of places, and it probably comes from more than one. Common places for that conflict to come from are concerns around ownership and intellectual property. These models scraped other people’s content without their permission. Now they’re making money off of it. That’s gross. Or to many people that feels gross.

Another part of it is, is this taking away artists’ jobs, or will it soon? Another is, is this violating the value of effort and craft? Right? There’s value for putting forth human effort for its own sake, to learn a craft and execute it, and express yourself. AI art isn’t doing any of that. And another I hear is authenticity. Right? There’s something that feels really inauthentic about AI images. Even calling them AI art feels like an offense to the value of authenticity.

And so finding out which values are in conflict, and that’s why I separated the middle part of the book into these different ethics, is so you can identify which one is feeling threatened and make more precise choices about how to constrain, compensate, rethink the work that actually address that value instead of going AI art is yucky, therefore I don’t use it. Which again, [a] very valid point. In fact, that one is probably fairly easy to avoid. But if you can identify what value feels threatened for you, this is like a personal question. And maybe in an organization they have organizational values and could do that, but identifying that values threat is the first step.

Learning what you can about how the technology is weak and strong, and different ways of using it. It doesn’t have to be a shortcut. Although I gotta say I was not a good student, and I didn’t get good grades until I was getting a PhD. Who knows how that all happened, but I absolutely would have used it as a shortcut to be lazy. But that’s not the only way to use it, right?

Yeah, I list a couple of different ways. We talked about a search engine also not being a great way to use it. An oracle is also not a good way to use it. Tell me something true about the universe, or predict the future, or reveal what somebody’s face looks like behind a mask. It cannot do that. But people were doing that on the news yesterday. Like as we were talking yesterday, they were trying to see behind someone’s mouth covering. It can’t do that.

There are other things it can do, right? It can operate as an assistant. Here’s a concrete task with goals, context, a specific format that I want, go do this component part of this task, and then I will integrate it into my work.

You can use it as a thought partner. If I had had any sense of doing this at the time, I could have said I’m writing a book about ethics and AI in mission-driven work. Please give me seven options for an outline. By the way, I suggest picking a high number more than you think, because when you’re scraping the bottom of the barrel, then you know you’ve got all the good ideas out of there.

Or what I did use it for was I hear people talking—when I talk to people in organizations — I hear them talking about AI as if it is cheating. So there’s clearly a value at work here, but outside of an educational context, cheating doesn’t make a lot of sense. So what’s going on here? And I chatted back and forth with it until I figured out both authenticity and trust and effort and craft. And I was able to separate those two. When things are murky, it can be a good thought partner.

Adversary is one of my favorites. Here’s my argument. What are the best arguments against it? What am I missing? All of these things. The reason why they’re sycophantic, which is a great vocab word I needed to learn for the SAT, and maybe kids need to learn a little earlier now—they’re all aligned to help the user. But one thing that users really like being told is what a good idea, you’re so smart. And so setting up the chatbot to have usefulness, be disagreeing with me, is a really good way to get around that sycophancy issue. And it can be useful instead of just echoing back to you what you’re thinking to have a separate set of—you can’t see maybe, but big quote marks on this one—eyes on your work can be helpful.

You can use it as a shortcut. There are some cases where that is fine. I need a refrigerator. My food is currently going bad. I have this size, space, and have this much money. I don’t need to know the history of refrigeration. I don’t need to know all of the different companies that offer refrigerators. I just need to know a refrigerator that I can buy today.

There are accessibility tools, right? You can use it for speech-to-text, text-to-speech, and image recognition. I was talking to someone who uses it to transcribe meetings because they have really struggled with executive function, and they’ll get really distracted, and then they’ll come back and be like, Oh crap, I missed something. Well, they’ll either be lost, or they can scroll back in the transcript, and then they can stay on track with the meeting.

Use it as a tutor. So don’t solve this problem for me, but write practice problems for me. Or as a coach, I would say be very careful with this, as in a self-improvement world. But if it’s very tightly scoped, that can be helpful. I’ve got—I’m unfortunately kind of a negative and pessimistic person by nature, and there’s a limit to the amount that you can go to your boss and say here’s all my negative pessimistic ideas about what we’re doing here, and I’m afraid I have crossed that limit.

And so I created myself a little GPT. ChatGPT has a tool called GPTs that’s like you’re fine-tuning your own tiny little model. And I gave it Learned Optimism by Marty Seligman and other papers and books that I have read and I like. So it’s language I know, and I use it. Okay, I’m going to come to you with a thing. I want you to reframe it to me in a more positive and more optimistic way. And that’s kind of adversarial, but it’s also, it’s in the direction of a personal development task. Do not use it as a therapist, though, is my recommendation.

So thinking about different ways to use it and then finally creating or finding learning resources and channels to share across other people who you’re working with across your classroom or across your school or your district to build best practices and learn from other people who are using it in a bunch of different ways.

And set up guardrails or a prompt library like here’s a prompt that I use if you’re comfortable creating a rubric with it. It’s, you know, three paragraphs long, much longer than you were thinking, but it really does come up with a really good rubric. If you add, the assignment and three samples, you know, one sample for each level or whatever, like a checklist and a prompt library can really help bootstrap high-quality and ethically compliant outcomes when we sort of work together as an organization or a team.

That’s really nice as a goal. I feel like you’ve given a really broad set of examples there for ways to address—to really get to the bottom of, okay, what is the conflict for me in this moment? Why am I resisting this idea of using AI? And then I think once you understand which aspect of, you know, which ethic you feel like is being violated, then you can kind of work from there and figure out, okay, what would make sense to do.

I love the focus on individual actions because I think that’s much more empowering. I also don’t want to leave out the piece you were saying earlier about how this really, there needs to be some things happening at systemic levels, too. Is there anything that we should be advocating for in terms of our lawmakers or our companies, our institutions, our schools, the tech companies that we partner with in schools? I really liked what you said earlier, by the way. I want to make sure folks got this: to be asking, you know, if you’re going to be partnering with this specific AI tool, what are you doing to mitigate the environmental costs before we spend thousands of dollars adopting your tool, what are you doing? Because that’s where you have your leverage.

Yes, definitely. I want to acknowledge right up front that our governments and individual companies, less so the latter, are somewhat restricted in what they can—like the government can’t regulate China. So there’s an issue of local impact, but there’s a lot of local impact that’s really important, like the data center locations. And so that is something to think about when we talk about regulation, that there’s a limitation to how much that tool can help.

But right now, the biggest tools and the most useful tools mostly are located here. There are some really powerful Chinese models that exist as well, but they’re not as good, currently, as ours. And so advocating directly with the companies can be more effective currently than I think working with the government.

I do have on my website, drkarenboyd.com, a workbook called Mission First AI Starter Kit, that’s got a vendor rubric in it, that’s got questions that you can ask, and then sort of some prompts to help think about what other questions you want to ask and how to set up a rubric.

And this can help not only get the people in your organization on the same page about what kind of technology are we acquiring and what are our priorities ethically across the organization, but it can give people who are calling a script, right? Like, make sure you ask about this and this and this. And I think that can both help your organization make a choice that feels aligned with their goals and ensure that these companies are getting these questions from a lot of different folks.

Yeah, I think that’s great. I think that is the value of having these kinds of conversations because sometimes people can be like, Well, what’s the point in even talking about it? Well, because our discussions about it shape consumer use. Consumer use shapes the way that capitalism runs. And if the people refuse to use something and the people are asking questions, if it starts to hit the bottom line, it makes a difference. And so I think we actually have quite a bit of economic power as consumers to think about where we’re spending our money and how we’re spending our money and what we’re saying to these corporations. So that is awesome. So we can go to drkarenboyd.com to get those. And then you also have your book.

Yes, it is out on March 20th. It’s available for pre-order on Amazon. Yeah, excited. Right now the ebook is for pre-order, but once it’s released, there’ll be a paperback and an audiobook version as well. The audiobook version, I cloned my voice and had it digitally narrate the thing, which was a really controversial choice that I thought about a lot, and I wrote a whole blog post on and how I decided that. I think it was a good little microcosm of making these trade-offs on a one-person scale.

But I will tell you, it was so much more work than I thought it was going to be. And the technology is not quite there. So you’re sitting there talking to a microphone, like coaching it through its emphasis for hours. But anyway, there’s such a thing as a good audiobook narrator, and that’s not me.

And I’m not good at reading. You have to read all the words, not just some of them or the ones that come to mind. You’ve got to read the ones that are there. And yeah, I really struggled with that, but at the end of the day I thought it was more important to have an audio version out there.

I guess what I’m not wanting to do is normalize this idea that we just have AI do everything and everybody’s constantly questioning is it or is it not? I really want to see the labeling. The tech companies, you know, the social media channels, don’t seem very excited about doing that.

Anyway, I agree with you, and I look forward to seeing what that answers. But I think we are very early on in the development of norms around disclosure.

Yeah, we are. And you’re pushing my thinking because you’re right, like I’m wanting the label so that I can avoid it. So I would see the AI-narrated audiobook, and then I would never listen to your book. And meanwhile, you spent more time on training that bot. You’ve put so many human hours into making this really good. And actually, what I would want to do probably, if I didn’t know you, I’d probably be like, Oh yeah, I don’t want to listen to that. And is that fair for me to do? I mean, certainly it’s my right to do it, but what am I missing out on? Particularly since most things have at least some AI involved in them.

It’s more complicated, I’m realizing, as I hear your response, than what I was giving it credit for. I guess I’m thinking of spam accounts on YouTube who create like a hundred videos a day, and it’s all AI-generated versus an actual creator who maybe has some AI images in there, used AI to help a little bit with the script or whatever, maybe, you know, some clips. Totally different thing. And yeah, how do we manage that other than as we’re saying here, like, let’s have the conversations, let’s think about it, let’s reason about it, let’s not let billionaires make these decisions. Let’s not let the people who stand to profit from AI decide for us what our norms are going to be. Can we talk about it?

Very true. Can we get some incentives in there? For sure, yeah.

Like, let’s interrogate that. You and I, as we—you specifically and I are both content creators and content consumers. We are both making things. We have our own intellectual property we put out into the world, and we consume other people’s. We know what that’s like, and we’re reasoning about it, and now we’re reasoning about it here publicly for other people to also consider. And I just think there’s so much value even in the conversations, even though we’re not leaving folks with like, okay, do this, don’t do that. Here’s your quick, easy answer. Just don’t use AI, or just don’t worry about AI. It’s fine. It’s not impacting the environment. It’s not that bad. Don’t worry about it, guys.

Yes, totally agree.

The reason conversation I think is important.

Yes. It is so tough for me to watch the conversation about sustainability and AI online because it really seems like there are two options. One is, ‘This is ruining the environment, and you’re a bad person if you use it’. And the other one is, ‘Don’t worry about it’. And neither of those is, in my opinion, reasonable at all. Actually, you know, maybe one more than the other. But I just—yeah, it’s hard to have a nuanced discussion on the internet. And now we have a lot of conversations on the internet.

Yes.

And so I think that, I mean, I think that is a big part of why I advocate for lots of feedback and sharing in organizations. Because when we’re talking to each other face to face, like, I know the people that I work with, I know they’re not trying to get out of work. Or if there was someone I thought was getting out of work, I wouldn’t trust their output anyway. So whatever. That facing each other as humans is going to be a really important part of moving forward into a future of work and life that includes AI.

A Takeaway Truth About AI Ethics

I always like to close out the show with what I call a takeaway truth. It’s something that I want folks to remember in the week ahead. What is something that you wish every educator understood about AI, sustainability, and ethics?

I think for sustainability and AI, and actually all AI ethics, I would encourage people to remember that it is complicated. Reasonable people can disagree. It’s not black and white, but it’s a lot more interesting than that.

Yes, it is. And I like the framing of it as an interesting conversation because I think some folks see this and are like, I don’t want to think about this. This is too complicated. It’s changing too much. I can’t understand. It doesn’t make sense. But it is interesting. And we have the opportunity to shape the future of our entire society. Those of us who are informed about this.

Especially those of you who are teachers.

That’s right. That’s right. And it’s an opportunity. We don’t have to just be passive and just say, It’s inevitable, whatever’s going to happen, I can’t do anything about it. Yes, you can. Yes, you can. You can be talking about it, you can be reasoning, you can be thinking. And I love this framing of it’s, you know, it’s something interesting. It’s an interesting problem for us to solve. It’s an interesting new territory for us to explore. And I think we’re going to come away on the other side with a new understanding of what it means to be human, because now we have something that’s really challenging that.

Yeah, I really do think there are skills that are going to be so much more important in the future. And one of them, fortunately for me, is being critical and negative. So, like, read one of these outputs and go, No, this is wrong. But like, moral imagination, these things don’t have that, right. There’s a lot for us in the future, but it requires, I think, really talking to each other as people and giving each other the benefit of the doubt and trying to come to an agreement across some different perspectives in order to get there.

Get the sustainability chapter of Karen’s book for free at drkarenboyd.com/freechapter. No sign up is required, but you can get updates on AI in mission-driven work in your email about once per week if you select “sign up for news and updates” there.

You can also download her free vendor rubric for schools called the Mission-First AI starter kit.