(General note: throughout this post I am saying “AI” for short, but I mean large generative and LLM models specifically, not other forms of machine learning. I’m sacrificing some specificity for flow, and I’m okay with that.)
This post came out of a conversation I had with old friends on Discord, about a month back. It gave me words for the finer points of how I interact with the current landscape of generative AI/LLM tools.
My friend Adam Morse later wrote up a list he made, of the categories of reasons people do not use these tools, in a post called “AI Vegetarians, AI Vegans, and AI Users.” I talked about this briefly on Bluesky a few weeks ago, and I’ve continued to talk to people about the categories he came up with. My assessment of myself has turned out to be remarkably stable over time, in my little case study of myself, the past month. That has given me a lot more confidence in describing the exceptions to my overall policy of low-to-no generative AI use, at the same time as it has helped me to reaffirm my core reasons.
When Adam originally shared this list with friends, I immediately classified myself as strong on seven of the eight categories of reason for AI refusal. (Adam hadn’t written #9 yet.) The holdout is #6 about threats to humanity from the creation of more and more powerful AI models, up to and including artificial general intelligence or AGI. I have higher priority things to do than sort out my feelings about that one.
Of course, September 2026 brought a huge wave of worrying about AI risks of all sorts, from unexpected agent behavior to patient privacy violations in healthcare to what if it gets SMART and decides to WIPE US OUT??? (Easy there, John Connor.) I have learned more about about this than I wanted to, and I am taking refuge in dark humor. First there’s the usual grift as Tochi Onyebuchi described it:
Then there’s something I wish I’d learned earlier: what the TESCREAL ecosystem of people are, and how much of a risk they actually are to our world: “The ‘AI Safety’ Movement Is Making AI Less Safe” from Reason is a good rundown. (Disclosure: I rarely agree with Reason, but I went to Yale University at the same time as the current editor. A lot of people at Yale tried to make me more conservative or libertarian. It slowly did the opposite. Thank you to the Southern Local Teachers Association, and to the Yale unions, who taught me so much about solidarnost.)
But to drag myself back on topic, existential risk itself is still my lowest priority reason, even though I want to pop popcorn, dunk on Effective Altruists, and read more of Timnit Gebru and Faine Greenwood’s Bluesky posts. (Look both of them up, for more on just how concerned we should be about this food web of people.) I would say my main reasons are still labor solidarity first and foremost, unwillingness to let these tools dull my own skills second, and its poor-quality-to uselessness for the writing, indexing, editing, and researching tasks that are the core of what I do in the world, third. #1, generative AI is useless, is so often entirely true that it’s a main driver of why I’m so not worried about reason #6.
Even though I don’t want to see your latest prompt results, I am willing to talk about these reasons with other people. I would call myself an AI vegetarian, rather than an AI vegan. Now that Instagram has integrated Meta AI into their search, I do still search to find the post it just refreshed me away from before I was done (again!), and haven’t ditched the platform in disgust, because I follow so many makers, writers, and musicians there. Likewise, I don’t click on “Ask Youtube” when I’m searching, but I’m sure I’m contributing to training data. I have AI search turned off in my most frequently used browsers, but I run unmodified Chrome when I have to in order to make someone’s fancy web page work.
What this list has done for me, in conjunction with the “AI vegetarian” and “AI vegan” language, has lowered the level of defensiveness and desire to “win,” in conversations about how all of us use generative AI tools. It has made me more confident in the moments where I do need to interact with the output of generative AI, such as the example result from Indexia AI Book Indexing that a colleague showed me last Friday. I know people who wouldn’t even want to have that file on their hard drive, and we talked about how to check first, to see if an instructor we know would want it as an example, but not to presume about her response. I also love that generative AI vegetarianism/veganism stacks up against actual vegetarianism and veganism in having so many shades, definitions, beliefs, and reasons behind it. In some places a soup with a meaty broth is still called vegetarian, fish is categorized interestingly in a number of faiths, and I’ve had at least one friend who would only eat “fallen fruit” meat by finishing your leftover chicken dish or taking the leftovers home: frugality and avoiding waste were now the priorities, after you’d purchased and eaten your fill. That friend could still digest meat, but many of my veggie friends cannot, while still others have had to stop being vegetarian because of other dietary needs.
It matters to me to coexist with many kinds of people, and to have conversations with my friends that leave room for our choices and practices to be different, and room for us to have to exist in institutions that may be requiring us to use tools. This framing is also great for encouraging others to ask rather than assuming, about each other’s generative AI use. So you bet that the next time someone sends me a generative AI video out of the blue without asking first, I’m gonna call myself an “AI vegetarian” and see how it goes. It doesn’t have to work perfectly, either, as long as it helps the conversation to happen. Give Adam’s framework a shot, and see what it nets for you.