## Avoiding Death by Meat-Proxy: Moral Ownership and AI in Schools Since term started in September, I have seen an abundance of shameless unacknowledged AI usage by other teachers at my school. Leadership members have shown us PowerPoints with obvious AI-isms; departments have presented assemblies with AI-generated images as central illustrations; and AI-made posters abound. The kids hate it. They see AI and they roll their eyes. They describe it as 'slop'—soulless, lazy, and without care. Nowadays, there is no harsher critique than 'that's AI'. Increasingly, I find myself agreeing. Over the last few years, I have used AI extensively. I have used it for a wide range of tasks: making websites, editing writing, creating worksheets, and as a discussion partner to think things through. Yet, when I see many of my colleagues use it, I groan. There is a place for AI in schools—but this place is not *everywhere*. On some level, the kids need to get over themselves and accept that not everything AI is bad. We also need to get over ourselves and stop using AI for everything. We need to restore and build trust with our students by demonstrating integrity and transparency in our work. This is vital as we have a moral duty to model ethical practice. With questions looming about what the future of work will look like—whether capitalism will crash, and whether we are all going to die—the most sensible thing to do is to establish a framework to navigate the unknown. A firm grounding in ethics and morality, tempered by pragmatism, will hopefully provide a stable platform from which we can face the future. ### Moral ownership as a framework What does it mean to own an idea? Can you own something created by AI? Does ownership even matter? None of these questions are new, but the ease of generating content and abundance of AI-made material has placed them slap bang in the middle of every classroom across the world. Addressing these questions is necessary to decide where the use of AI in schools is appropriate, and where AI should stay well away. The idea of ownership has always been messy and is made even messier in the world of AI. Modern AI models are trained on the open internet: they are an "alien"[^1] synthesis of the sum of human thinking and creativity. They exist in the shadow of all human creation that came before them—as do we. Everything we think, write, and do happens in the shadow of our lived experience. Our lived experience is the sum of everything we have read, heard, watched, felt—all we have ever consumed from the world around us. Our world is shaped by the works of billions, and we stand on the shoulders of giants. Nothing is new under the sun. When we create something using AI, we mix our own bastard intellectual inheritance with an alien imitation of human thought. Depending on the prompts we type and the thinking we do, we might feel a greater or lesser claim to ownership of the output. If we use AI to create something novel, we might likewise think that we own it—and we might. The only legitimate basis from which we can claim a moral stake in a creation is doing the work of thinking ourselves. When we do the work of thinking, our creations become part of us. We inhabit them and they inhabit us. This is **moral ownership**. Moral ownership is not objectively measurable; it's rooted in the feeling of inhabiting something and being inhabited—the feeling that "I made this". We can interrogate the legitimacy of this feeling by checking our understanding, and assessing our ability to explain and convincingly defend the work without the chatbot present. The evidence we might find is not, however, proof. While I might think my defence robust, you might think it meek; while I might think I understand, my understanding may be false. ![[the-internet-comic-anthony-clark.png|Immoral ownership|300]] *Immoral ownership.* Comic by Anthony Clark, Nedroid Comics[^2]. We can ask ourselves if we did the work of thinking, but we must question our reliability as witnesses (particularly in our own trial). Can we be honest with ourselves? We probably can't. Not 100%, anyway. In this mess of determining ownership, there's not a neat answer. We must settle for good enough—finding *enough* evidence that we did the work of thinking, convincing ourselves *enough* that we are being honest about our contribution. Doubt is good because it keeps us honest—and honesty is vital. #### Responsibility and accountability It is possible to be responsible for something and have no moral ownership of it. When a manager directs their team to complete a task, the manager might have no involvement in the work of their team beyond the initial direction. The manager is responsible for their team's work—they own the correctness, quality, and consequences of it—but they cannot account for it as they don't own the thinking behind it. As they cannot account for it, they cannot claim moral ownership of it—doing so would be dishonest and erase the efforts of their team. When a student claims the work of others as their own, we call it plagiarism. We must hold ourselves to the same standards as our students and not take credit for work that we did not do. Just as with the manager directing their team, we are responsible for the output of AI when we make it do work for us. However, when we are hands-off and let AI do all the 'thinking', we have little claim to moral ownership of its work. When we put something AI-generated out into the world and take credit for it either explicitly *or by omission*, we are no better than our cheating students. Holding a manager like the above accountable for their team's work is difficult because they have done none of the thinking required to understand it and cannot account for it. Had they done the work of thinking, they would be able to explain and convincingly defend the team's output. One can only account for, and therefore be held accountable for, the substance of a piece of work if one has moral ownership of it. When a student submits AI-written homework and we grill them, they are unable to explain or defend the work. The student cannot account for the work they didn't do, though they are liable for dishonesty. Sometimes, accountability is necessary. When making decisions that affect others, we must be accountable for our choices. When writing policies and inventing initiatives to address problems at school, we must account for our actions. Accountability is important because of the impact our actions have on others. The problem comes when humans are removed from the picture and AI does all the 'thinking'. If there's no human behind the curtain to be accountable for decision making, what do we do when things go wrong? If nobody is able to explain and defend decisions about the way our school is run, the integrity of the decision-making process disintegrates. We can interrogate the leadership team as much as we like, and cut their heads off if we find them liable, but if AI is taking the lead, we'll get nowhere. We can't put AI on trial—and we shouldn't have to. #### Writing is thinking Writing is an excellent mechanism to test claims of moral ownership and hold ourselves accountable. When writing, our understanding (or lack thereof) is laid bare before us. We can't trust ourselves to be honest, but we can trust the blank page in front of us. When we let AI write for us, we destabilise our claims to moral ownership—but there are degrees to this. Consider the roles of writer and editor. Often, the writer takes the lead with their ideas and the editor supports them. A good editor, however, can end up just as accountable for the final text as the writer. Different writers and editors will contribute varying amounts: a bad writer might be saved by a good editor, or an overly zealous editor might quash the voice of the writer. I regularly use AI as an editor when I write. I use it to check my grammar and identify clunky phrasing. I also use it to suggest alternative words and phrases when I can't quite find them myself. I am still doing the work of thinking, but I let AI handle some of the implementation details. AI is a source of inspiration which gets filtered through me and put into my own words. When writing, I also use AI as a discussion partner. It challenges my reasoning and pokes holes in my arguments, which refines my thinking in the same way a debate with a friend would. There is a difference, however, between using AI as a partner to refine thinking and letting it replace the act of thinking altogether. [[Nowhere to Hide|Writing is how I think]]—if I outsource my writing, I lose moral ownership. For me, as a writer, my ability to claim moral ownership is important because I want the thing I inhabit to be mine. It's a matter of pride in the work that I do. I can't be proud of work that isn't mine. The flip side to writing, of course, is reading. Reading is about the meeting of minds: human connection that elevates the understanding of writer and reader[^3]. This presumes that there is a mind to meet—rather than an alien simulation of 'thought'. If minds can't meet, there's little point reading. If there's little point reading, there's little point 'writing' with AI. The potential cost of letting AI write on one's behalf is the worth of the words themselves. ### Application of the framework #### 'Writing' I had a colleague who used AI to write all of his emails. It was done so blatantly and lazily that my respect for him plummeted. His emails would say something, but I never had confidence that it was what he really thought. I never had confidence that he thought at all. Messages were exchanged, but minds never met. With current models, it can be easy to spot AI writing: gratuitous emojis, 'it's not *x*; it's *y*', and em-dashes[^4] are a few common indicators. It's not clever; it's embarrassing when AI use goes unacknowledged—and when its influence is bloody obvious. It's hard to have confidence in a message when it blatantly hasn't originated with the person communicating it. In the future, I can imagine a world where AI writing is harder to spot, but harder detection doesn't address the absence of moral ownership. We should be accountable for our words, which means we should do the work of thinking to shape them. With better models, confidence in the message may not be shaken as badly as it is today, but the integrity of the message remains questionable. Honesty about the involvement of AI in creating a piece of writing helps by dispelling doubt that otherwise taints the message—assuming, of course, that the writer has moral ownership of their writing. In the absence of moral ownership, transparency about AI use is a handy signal that we can save some time. My view is that, if you can't be bothered to write, you shouldn't expect me to read. #### Moral ownership doesn't always matter The problem with using AI in the kind of writing described above is that the writing's purpose is to communicate—it's about the meeting of minds and human connection. Not all writing, however, has quite such a romantic goal. A worksheet question, for example, is judged on its effectiveness at testing understanding rather than the integrity of its author's voice. Not all writing requires a meeting of minds; not all tasks require doing the work of thinking. To determine whether moral ownership matters for a given task, we can ask two questions: *Does it matter who is speaking?* and *Does it matter that I made this?* The first addresses the importance of voice; the second, the importance of inhabiting the creation. If the answer to either of these questions is yes, moral ownership matters. A more practical approach is to group tasks by their primary purpose: **function** or **communication**. Functional items—worksheets, quizzes, spreadsheets—succeed based on utility, regardless of who wrote them. Communicative items—emails, assemblies, staff presentations—rely on the authentic presence of the speaker's voice. While this boundary can blur, the distinction gives us a practical way to think about where AI belongs, and where it should stay away. #### Function When something's primary purpose is its function, AI is well-suited and (probably) appropriate as the driver of its creation. We see the same principle already in use with bought-in lesson resources: whether a resource is from PG Online or Dave down the street doesn't matter, as long as it's good. Using AI to create functional items fills the same gap as buying in resources. AI compensates for our lack of time and (maybe) creativity, offloading the hard work of implementation and thereby freeing us up to think about the big picture of our teaching. We don't need to worry about the exact wording of a question or scrape the bottom of our imaginations for contrived examples when this work can be outsourced. In the time it takes me to write one example paragraph, I could describe five and have AI write them for me. Rather than worrying about the exact wording of each paragraph, I can think instead about their shape: what misconceptions do I want my students to be aware of? How should they structure their answer to ensure maximum marks? I would much rather spend my time thinking about this big picture stuff than worrying about the weeds. The speed and ease of AI-generation allow me to be much more responsive to my students' needs than bought-in resources ever could. If PG Online's worksheet doesn't test something my students are weak on, there's little to be done short of writing new questions myself—except I don't have to because, if I want, every worksheet could be tailor-made for my class. To be clear—this expectation should not be normalised, but it's freeing to be able to respond so quickly to my students' needs. 'Adaptive teaching', anyone? It would be a mistake to assume that just because something is functional, I get a pass to not do any of the thinking behind it. I am still ultimately responsible for the work I set my students and I have a duty to ensure its accuracy and quality. I should be accountable for the decisions I make in my teaching. This means that, while making worksheets and so on with AI is fast, it isn't instant, as verification of their suitability takes time. If I evade accountability, my students' learning will suffer and I will undermine the respect they (may) have for me. Functional work doesn't require personal authorship, but it demands professional judgement. It would also be a mistake to not be honest with my students about when I have used AI. We expect them to not pass off the work of others as their own, and we should be held to the same standard. As part of our duty to model ethical practice—and to cultivate an environment of trust—we must be honest with our students. A huge danger to the relationship with our students is us getting caught in a lie; they can probably tell when we've used AI, so denying it is stupid. Regardless of whether they can spot it, we should not lie to our students! ##### Gradebook: navigating moral ownership Moral ownership can matter for functional things, which complicates the practical test of *Is it functional or communicative?* Sometimes we need to step back and ask *Does it matter who is speaking?* and *Does it matter that I made this?* At the moment, I am using Claude to build some functional grade tracking software, *Gradebook*, and it matters to me that *I* am making it. My role in Gradebook's creation is important to me because I've done a huge amount of thinking for it. I feel that I inhabit it, and it inhabits me; however, I haven't done *all* the thinking. Gradebook shows how doing the work of thinking is made messy when using AI and how the answer to our two touchstone questions can be different; it also demonstrates how pride and perspective play on claims of moral ownership. Claude has built me grade tracking software that does everything I dreamed of and more. One of my favourite ideas that Claude implemented is automatic question extraction from exam papers. The questions get tagged by topic and command word, and are then presented in a Google Form for students to enter their raw mark, which takes all the admin work out of doing QLA. Creating Gradebook has taken a huge amount of time because it demands huge amounts of my judgement. Imagining the shape of the software and iterating with the right prompts is arduous. At a software company, my role might best be described as the Product Owner. I have done the work of *some* thinking: I identified the issue and had a vision for a solution, but I wrote no code. The process of creating Gradebook has been intensely iterative. I start with a vision for a new feature or functionality and describe it to Claude; I often tell Claude to "talk before you build", which leads to it interrogating my vision and asking questions. Usually, it offers a series of options and, nine times out of ten, I am happy with its default choices. Sometimes, it comes up with new features on its own, or does the work of figuring out how to make something work when all I know is the problem I want to solve. Through endless iteration, the lineage of ideas becomes an untraceable mess, which mirrors our own bastard intellectual inheritance. While we could trace the chat log, we could not determine at what point my ideas became Claude's and Claude's 'ideas' became mine. Given this mess, to what extent can I be accountable for Gradebook? I understand and can explain all of its features, but I can't say exactly how they work. In earlier iterations, I could, but it has grown so much in complexity that I can't keep it all in my head. Equally, I can defend design choices, though I didn't make them all myself. My claim to moral ownership is at best partial—yet my claim is important to me as Gradebook is something I'm proud of. A user of Gradebook probably doesn't care who owns it or how it was made; they care instead about how long it takes to access the data they need for writing reports—so my feeling of pride is private and only my concern. The significance of moral ownership depends on perspective: it can be both critical for the creator and irrelevant for the consumer. I am comfortable with how Gradebook was built because of its primary purpose as a functional tool; I am comfortable taking responsibility for it as I have tested it myself. Partial ownership is okay for functional things. AI use is fine as long as we take responsibility for the output. #### Communication as connection Partial ownership is not okay when the voice of the speaker is important—when it matters who is speaking and that they own the message. When we want students to understand the negative impact of drugs, we don't just do an assembly. We get an outside speaker with lived experience, who presents the anti-drugs message in a way that is more meaningful than if it were just delivered by a teacher. The voice of the speaker matters because their voice, in this case, is informed by their lived experience. The speaker's moral ownership of the message also matters as the assumption that the message is theirs is foundational to the communication. This importance of voice and ownership would likewise apply to any other communication where the meeting of minds is the point. The appearance of moral ownership could result in an effective delivery of the message. AI writing that passes wouldn't undermine the message in the eyes of an unknowing audience, just as a hired actor could convincingly and effectively deliver the anti-drugs message. If the actor were good enough, students wouldn't suspect anything was off. Difficult-to-detect doesn't mean okay. The same words, spoken by a liar, are worth less than if they were spoken by someone with integrity. If students learned that we had hired an actor and hid this from them, or we hired an actor so awful that their inauthenticity was blatant, our deception would become the headline rather than the anti-drugs message. Inauthentic communication is less valuable whenever there is an expectation of good faith. In schools, we should be operating under the expectation of good faith whenever we communicate with students, staff, and anyone else in the wider community. As teachers, we have a moral duty to model ethical practice, and working in good faith is part of this. When our voice matters, AI has no business speaking for us. We must avoid death by meat-proxy: becoming a front for a chatbot that did all the 'thinking' for us. #### Institutional integrity As my former colleague's emails demonstrate, unacknowledged AI usage poisons the message regardless of its primary purpose. When presenting a new initiative or delivering staff training, moral ownership of the message is important because the voice and credibility of the speaker determines the worth of the message. Using slides with obvious, unacknowledged AI-isms—garbled images, citations with no bibliography, AI sentence structures—is embarrassing. As a professional addressing a room of professionals, the standards for integrity should be higher than the standards we hold our students to. Further, failing to acknowledge the role of AI is dishonest and undermines your credibility and the integrity of your work. AI undermines the integrity of some work whether it's credited or not. Some work needs to be done by humans because humans need to be accountable for it. Writing a school-wide policy on AI usage, for example, should be human-led. In this particular example, the irony of using AI is problematic in a way that undermines the resulting policy. More broadly, it's problematic because AI cannot be held accountable. When accountability is needed, AI must take a back seat. Beyond just accountability is the question of the suitability of AI-led solutions. As teachers, we know our school context better than anyone else and we know that just because something works in one school, it won't necessarily work in another. Who is in a better position to imagine initiatives to address the problems in a school: the teachers who work there, or an AI model that may not have even been trained on the school's website? "Because AI said so" does not automatically make an idea good. We must be responsible for the work AI does for us, so we should be judicious when directing its work. Let's please avoid further bullshit like learning styles and make sure our decisions are grounded in either the reality of our school context or research evidence. This is not to say that teachers and school leaders should not develop their ideas with AI as a discussion partner, just that the resulting work should be owned by a human who can be held accountable for it. #### The poster problem Posters plaster the corridors. They are frequently created by AI without care or verification. Two amusing examples from my school are a poster with a hallucinated school motto: "Faith, community, excellence" (we are a secular school) and another with a series of AI-generated images with obvious issues: a plane with two cockpits, imaginary sign language signs, and worse. The most egregious example, however, was at our recent Sixth Form open evening where every sign showing the different departments had a hallucinated version of our school logo. That none of these issues were caught before the posters were shared is embarrassing. We must take responsibility for verifying the correctness of our work. AI-made posters become a broadcast of institutional laziness that invites ridicule rather than engagement. It is shameful to cover the school in slop, which students take as evidence that we don't care. Posters are a prime example of things that blur the line between function and communication. Most posters are functional in that they convey a message whose author is irrelevant: "Maths revision at lunch". However, for posters to succeed, they need to first engage which is why they must also succeed at communication. As the example of these posters shows, the function of a thing can be degraded by its presentation. Students perceive posters as proxies for communication about how much we care. If they sense AI has been used, they see a negative subtext and instead read the poster as a message that "my teacher couldn't be bothered". These concerns go beyond posters, and apply to worksheets and lesson resources too. ### Student sensitivities On one level, students need to get over themselves. They need to learn that there is nothing inherently bad about things that were generated by AI, and that it is possible to retain moral ownership of things implemented by AI. We must try to understand, however, why students might react so negatively to the smell of AI. I want to do more to understand student perspectives, so please bear in mind that what follows is based on my conversations with students so far. Students tell me that they worry about AI-made resources being lesser in quality and containing inaccuracies, or not sticking to the scope of the specification. They think that we are shirking our duty as their teachers by being lazy and not putting enough effort into teaching them. Students trust us to do the right thing by them; they trust us to act with integrity. When we breach their trust and use AI-made resources without telling them, the foundation of our relationship is shaken. In my experience, if students feel like they have been deceived about AI's involvement in a resource's creation, they respond extremely negatively. Related to this broken trust is the perception of hypocrisy: why is it okay for teachers to use AI, when it's not okay for me? We might want to dismiss this and say "well I've already got my GCSEs", but this attitude is insulting and delegitimises concerns that are extremely real for our students. The biggest concern, for some, is existential. Amidst the climate crisis, the rise of the far right, and post-Brexit economic instability, students are intensely aware of P(doom)—the non-zero chance that AI leads to human extinction. As teachers, we have a moral duty towards our students, and using—therefore tacitly endorsing—technologies that some students think are evil is understandably problematic for some. If I were a teenager today, I would be terrified for my future and I wouldn't want to think my teachers were speeding along the end of the world. Some of these concerns are easily addressed. If we take responsibility for ensuring accuracy and correctness of AI-made resources, students shouldn't have to worry about it. The threat to trust comes from deception—and from students wrongly suspecting us of using AI. The best way to combat this is always disclosing when we have used AI. In the past, we never felt the need to say that a worksheet was made by a colleague, or from an online resource bank, because human ownership was the only possibility. Today, the world is different. To model ethical practice and to cultivate genuine trust, we have a duty to be honest with them about where we have used AI. It is equally important to use a mix of both AI-made and human-made resources—if for no other reason than to combat the growing misconception that usage of a semicolon—or em-dash—means AI was involved. By being up-front with students, the knee-jerk assumption that correct punctuation or (heaven forbid) good writing means AI can be defused. The question of laziness is more interesting as it's grounded firmly in student perception, rather than anything we can directly control. In my current practice, I explain to students why I have decided to use AI-made resources (usually the stuff we had before was rubbish) and I ask them if they would rather I spend hours making a worksheet or planning the rest of my lesson. I am also very quick to point out to them just how little time we get each week for PPA, and challenge their assumption that I would spend every evening of my free time working. While I can't control their perception, I can explain and defend my own. Accusations of hypocrisy are harder to deal with, particularly in the absence of a coherent school-wide approach to student use of AI. The reality is that, as teachers, we are working in a different context to our students. The standards we are held to are lower than the standards we hold our students to—and this is wrong. This discrepancy speaks to a broader issue with education in 2026: we are teaching students to complete tasks that are irrelevant in the future we are heading towards—arguably, tasks that are irrelevant at any point in human history, but more acutely and visibly so today. I think we should be honest and pragmatic about this reality, explaining that students need to work in such a way that will help them pass their exams, but taking opportunities when they arise to demonstrate how AI can be useful in the 'real world'. When I was at school, one of my teachers would always tell me to jump through the hoops and play the game. Our students need to play the game, while we need to think about how the game should change. Regarding student concerns about existential threat, I have nothing to say other than, yeah—we're probably fucked. We should probably keep playing the game and see what happens. Perhaps elect better leaders. We're counting on you, America… ### Conclusion Moral ownership matters in schools because, as teachers, we have a duty to model ethical practice. While we do not need to be able to claim moral ownership of every worksheet, we absolutely must seek to claim moral ownership when we communicate—with each other, and with our students. Human connection in communication is important now, more than ever, as we face the prospect of death by meat-proxy. We need to be intentional in using our voices where they matter. When our voice is intrinsic to the message, AI has no place writing on our behalf. AI has a place in schools, but that place is not everywhere. ### Postscript: Workload creep The obvious argument in favour of AI-made posters is that of workload. As teachers, we regularly don't have the time to make posters, so AI seems like the perfect solution. However, cheap does not mean free: there is a time cost to using AI for making posters. Someone needs to spend the time writing and tweaking the prompt, verifying the output, and so on. If I don't have the time to make a poster without AI, I don't have the time to make a poster with it. We must be firm in drawing the line here—one poster will not be the death of us, but you can only fit so many posters on a camel's back before it breaks. The fundamental issue here is admin tasks. Union guidance is crystal clear that teachers should not be expected to do them. Ideally, support staff should do this kind of work—but schools are broke, and support staff scarce. Given the capabilities of AI, asking teachers to use AI for admin work might seem ideal, but the risk of workload and expectations creep is concerning. If we normalise the expectation that teachers use AI to do admin tasks just because AI makes them cheap to do, we normalise making teachers responsible for admin tasks. This is wrong and removes any incentives to address the underlying problem of a support staff shortage. If I am not expected to make a poster without AI, I should not be expected to make a poster with AI. AI empowers us to be more efficient with the tasks we are reasonably expected to do, but it would be wrong to use it to justify teachers becoming responsible for admin tasks. AI does not resolve the structural issue of fundamentally unmanageable workload. It must not be used as an excuse to expect more from us when it's not our job. Even when it is our job, we should not normalise the expectation that every lesson has fully scaffolded individualised worksheets with Frayer models and dual coding with extra learning journeys and a side of slop. Maybe one day the time cost will be close enough to zero that this kind of hyper-personalisation for every student is viable, but the problem of accountability will remain. ### AI disclosure I used Claude, ChatGPT, and Gemini while writing this to both edit my writing and refine my thinking. I additionally spoke to Martijn Arts and James Cussen to help develop my arguments. ### Notes [^1]: Eliezer Yudkowsky and Nate Soares, If Anyone Builds It, Everyone Dies (Vintage, 2025). [^2]: Anthony Clark, "The Internet", Nedroid Fun Times (Tumblr blog), January 30, 2013, https://nedroidcomics.tumblr.com/post/41879001445/the-internet. [^3]: Mortimer J. Adler and Charles Van Doren, How to Read a Book (Simon & Schuster, 2011). [^4]: Amusingly, my favourite punctuation mark—as you can probably tell.