A College Course Class Policy Turned Manifesto
This post originally appeared on my Substack in December, 2025.
I haven’t stepped into a classroom since 2017, and now, preparing to teach a composition course this spring semester, I feel like I’ve time travelled. I missed teaching through COVID, with the rise of remote learning, and I never needed to think about how generative AI impacts classrooms and education.
In thinking about it now, I’m finding myself a little overzealous, or what the kids might call “unhinged.” What should have been a paragraph about whether students are allowed to use this technology or not in the classroom, became a lengthy manifesto about how we need to resist harmful technologies and futures we do not want.
I’m publishing this policy for a composition class I am teaching this spring semester because I think we should all be required to think about how and why we use generative AI.
Thanks for reading! Subscribe for free to receive new posts and support my work.
The Policy
This policy is more of an essay or manifesto. It’s also a work in progress. After we discuss it in class, I may add or take away from it.
I know it is atrociously long. Still, I hope you will read it through and deeply contemplate your use of generative AI in this class and any aspect of your life.
Generative AI, different from traditional AI, is a type of artificial intelligence that creates new content—such as text, images, music, or data—by learning language patterns from existing, human-created examples. Whereas traditional AI follows explicit algorithms and rules to perform specific tasks (think: the chatbot that coordinates a customer service chat or call), generative AI distinguishes itself by innovating material, which is why it sounds human. With this distinction in mind, when I refer to AI in this policy, I am referring to generative AI.
To sound like us, large language models (LLMs) are fed online content directly sourced from the internet. Our emails, YouTube videos, and Reddit and social media posts have all been used to train LLMs.
AI companies, such as Scale AI, also employ humans, known as data annotators, who train LLMs.1 Typically, these workers are underpaid and exploited, living in the Global South. They perform tedious tasks, such as identifying visual data that the LLMs cannot recognize. For instance, a data annotator might sort images of pants, shirts, and accessories from a commerce site for hours. This has been well documented in articles and books.
Humans are also needed to regulate violent, unsafe, or illegal content. Journalist Karen Hao has written for The Wallstreet Journal about workers in Kenya who have been traumatized by sexually and violently explicit images and videos while working to produce models like ChatGPT. And, since language and culture are always in flux, data annotators will continue to be needed to train and distinguish harmful content for LLMs into the foreseeable future.
Unsurprisingly, considering LLMs learn from internet content, generative AI reproduces our culture’s racist and sexist ideas. In an article from Science News Today, the editors write: “In a strange and unsettling way, AI holds up a mirror to society. And in that reflection, we see not just our intelligence, but our inequality . . . AI can be racist. It can be sexist. Not because it chooses to be, but because we have not yet chosen to be better.” This learned bigotry impacts society as it proliferates harmful ideologies. For instance, when AI is used to sift through resumes, it discards non-white and non-male applicants.
On top of reflecting back to us our own biases, AI has not yet learned how to distinguish facts from fictions. Its errors are called “hallucinations.” Instead of getting better as AI technology develops, though, these hallucinations are getting worse, and companies like OpenAI are unsure why.
Recently, a lawyer used AI to file a court case, which was found to be riddled with hallucinations, including made-up cases.
So, even though it may seem counterintuitive, using generative AI for school or professional work often means working harder, not smarter. To avoid a similar humiliation, you need to verify all AI claims, which means that using AI to research and write your papers often results in more work.
Copyright and plagiarism issues continue to arise as well, since, again, LLMs need human-created material to learn from. AI reproduces our writing, art, music, voices, and physical appearances. Think, for instance, about AI bands on Spotify, AI film and television actors, and commercial designs sold, like T-shirts and tote bags, with AI-generated images printed on them.
Scarlet Johansson recently threatened legal action against OpenAI when its chatbot stole her voice. They had based their chatbot on her character Samantha from the film Her, and she was understandably upset about it.
Let’s not forget the negative effect AI has on the environment: The creation of expansive data centers across the globe comes with high carbon costs. These sites replace farm land, neighborhoods, communities, and ecosystems with industrial sites that suck up resources and produce tremendous levels of noise and light pollution. In drought-stricken Argentina, for example, local activists vehemently object to the staggering amount of potable water these data centers require, which takes away from the local drinking water.
Another harm proliferated by generative AI includes the dissolution of human-to-human connection. Think about a situation in which a teacher creates her Composition I course entirely with the help of AI. In the class, the students write papers, which they use AI to write. Then, the teacher grades those papers using AI.
What would be the point?
You might be asking: If AI is so bad, why do we have it? Who benefits from AI?
As Karen Hao argues in her book Empire of AI: Dreams and Nightmares in Sam Altman’s OpenAI (2025), billionaires in Silicon Valley lead a for-profit AI industry, amassing their wealth and prestige at the world’s expense. If we define “colonialism” as a wealthy entity’s control over resources as it expands its authority into other territories (which companies like Google and Meta do as they outsource low-wage labor and mine precious minerals and resources for computer chips and data centers), then Hao is correct that US-based generative AI companies clearly position themselves as neocolonialists.
Generative AI CEOs like Sam Altman promise us that the benefit of AI outweighs the cons. He has made claims that AI is the key to administering universal basic income (UBI) once AI takes over our work, decreasing our need for employment. Instead of launching a company that would help him realize this future, though, he launched Worldcoin, a company that collects people’s biometric data—using an iris scan—in exchange for cryptocurrency.
In other words, instead of tilting the world toward betterment, Altman and other Silicon Valley AI tech CEOs create dystopian applications that surveil us and collect our personal data, which they then sell to advertisers and state governments. These technologies also advance facial recognition capabilities, used to police citizens, monitor political dissent, and enable the activities of ICE. If you are unfamiliar with the company Palantir, I suggest looking them up.
TL;DR, the bottom line is: Because OpenAI and enterprises like it usher in a new era of US imperialism that is neither good for the people nor the planet, LLMs are unethically produced and used toward unethical ends.
So, when you use generative AI to write your paper, correct your grammar, or plug in correct MLA format, you are paying a high social and ecological price to do so. Is it really worth it?
As I hope this course demonstrates, writing and communication are useful skills in all aspects of your life—personal and professional. Without good communication, for instance, we cannot forge deeper intimacies.
Additionally, when we use AI, we also lose our confidence and sense of value in the things we produce through our own, unassisted labor.
Perhaps most significantly, learning to read analytically and think critically for yourself may be the only defense you have while sorting the “real” from the AI scam. One of the biggest of these scams includes that we need to use generative AI to better ourselves and society.
To navigate the road ahead, we need our critical thinking skills to really assess if a future with AI slop and technofeudalism is one we want. We also need these skills to find alternatives to harmful technological advancements. White it is not AI that is innately bad, it is the way it is wielded as a precious resource by the people in control of it, people who put capital over people and the environment. Finally, we need our critical thinking skills to determine how to resist and decolonize the technological and economic systems that greatly shape our lives.
I acknowledge that AI, to some extent, will be part of our future and that it has the potential to be a helpful tool that could lead to a better society if the people and economic system behind it were truly non-profit and better regulated. At the end of her book, for instance, Hao cites researchers and activists in New Zealand developing AI technology for the non-profit purpose of preserving Māori language.
But the imperialism of generative AI is not the only reason I want to discourage your use of it in this class.
Frequently, in the workplace and in your communication with peers and colleagues, you won’t always have AI to rely on. There are many, many contexts in which you will need to be able to write and think without it.
Do you really want all of your email exchanges and text messages with colleagues and friends to consist of LLM output on both ends? Again, what would be the point?
And while writing assistants, like Grammarly, can be great tools for proofreading, I want to see your errors. Your errors confirm your humanness. Additionally, I want to help you identify weaknesses, overcome challenges, strengthen your writing, and feel confident. None of these things can be accomplished with an assisting tool.
After you establish your writing and analytical abilities, you may eventually turn to generative AI tools—your job may even require you to do so. But in this context, I want you to embrace that being a writer—whether a novice or expert one—is a process in which you fumble, grow, and always remain vulnerable to critique.
Plus, how can I get to know you and your unique voice through the words of a machine? What value will my feedback have for you—or me—if it is directed to a LLM?
In this class, the quality of your ideas counts the most, and you will never be penalized for struggling through the elements of the conventional college essay (including structure, format, style, citation, and grammar). I only ask that you work to improve these skills over the semester, letting go of the idea that mastery is the end goal.
To summarize (and this is a joke!): There are more ethical ways to cheat than using AI.
But, not joking: Cheating or plagiarism of any kind is not permitted.
Additionally, any use of generative AI for any assignment or activity, unless designated otherwise, is considered misuse in this class. These and other types of misuse may be considered academic misconduct and consequences will follow college policies.
1 I briefly worked as a data annotator for Outlier (part of Scale AI) after I was recruited through LinkedIn in 2023. At the time—and I believe currently, too—the company was looking for creative writers and people with advanced degress, particularly PhDs, to train LLMs. They lured me in with the promise of high pay—at $45 an hour—before dwindling that rate down to $15 (also before quarterly-paid independent contractor taxes, which subtracts about 15%).
Training included unpaid exams that you had to take and pass before work would be assigned to you. Then, the work was infrequent, and I often faced an “empty queue” of tasks. If I was assigned tasks, I had to work on them right away before they ran out, taken up by others eager to make money from the platform while they could.
Since we all worked remote and were assigned into ever-shifting teams of people—while hundreds of new employees were onboarded on a daily basis—we were alienated from our thousands of coworkers. Slack channels were filled with disgruntled and confused workers, often new to the platform, and the moderators of these channels were unhelpful. Our questions, critiques, and complaints led to more dead ends as the blind led the blind—we took it upon ourselves to respond to our colleagues posts on issues we also had little clarity on.
Algorithms assessed the quality of our tasks, monitoring how much time we took per assignment and how our work was rated by other people. The people who rated our work were other data annotators who didn’t understand the standards any better than the person they were rating. So, poor evaluations were not only demoralizing but rage-inducing as you could not contest, start a dialogue with your evaluator, or get any transparency from anyone about your poorly graded tasks.
Tasks were also timed, and the system often glitched—if it did, we would have to submit “tickets” to explain the glitch, using screenshots as evidence. If we forgot to track our own time and take screenshots, we wouldn’t get paid.
The tasks themselves were varied, but mostly I either rated LLM responses to human-generated prompts or wrote a response to a human-generated prompt as if I were the LLM.
Rubrics and policies changed continually, sometimes daily. Additionally, I was often bumped off and assigned to new projects that held similar but different tasks. So, every time I got the hang of a particular work flow, I was moved to a new project with different standards, rubrics, and an insurmountable learning curve that I’d never master.
For example, one day I was told to use wordy, human-like greetings and other conversational fluff in the responses I would write, and the next day I would need to get straight to the point. Or, sometimes I would receive praise for crafting lengthy assessments of why I rated one LLM response higher than another and, in another project, I would be penalized for producing the same quality of work. Moreover, any mistakes you made while doing this work, with its constantly changing expectations and technical lingo, resulted in less tasks assigned to your queue.
It was a Kafkaesque nightmare run by people much younger than me who didn’t share or care about my field of expertise. Yet, somehow, even though they needed me more than I needed them, and even though they ran this platform inefficiently and unethically, they succeeded in making me feel useless and stupid. So, when they cut my pay rate by more than half of what I had been promised, that was the last straw, and I quit.
I also realize now, too late, that they were using people like me to circumvent blocked access to copyright-protected content. In other words, since using copyrighted novels, books, and databases with scholarly articles like JSTOR and ProjectMuse were forbidden, they used the authors of such content to generate new content for LLMs to copy from.
Thank you for reading! Consider following me on Substack or following me on Instagram, @gina.yo.gina.
Leave a Reply