There are many variations of “artificial intelligence,” from those annoying bots that pop up on websites offering to “help” you (I don’t need a robot assistant to help me buy a pair of socks, thanks) to Arnold Schwarzenegger as the Terminator (which doesn’t exist … yet.)

These days, what most people think of as “AI” are large language models, or LLMs — computer programs trained with vast reams of data for tasks of language and information processing, as well as language generation. Essentially, complex text prediction programs. Have you started to type “thank y” and your computer or email program pops up with the suggestion that you say “thank you very much”? LLMs are a bit like that.

The most famous of these is called ChatGPT, run by a company called OpenAI. Most people seem to use this program as either a more wordy Google search — that is, to pull and present information off the internet — or to write emails, notes and other things. Personally, I have never had any interest in this sort of AI. I don’t need a machine to do my reading for me, and I sure as hell don’t need a machine to do my writing. The whole point of being a writer, which I very much am, is doing it myself, setting my own thoughts and conclusions to paper (or, in this case, to Word document). I am confident enough in my own intelligence to not need to hire out my thinking and processing.

I have to admit I’ve been intrigued, though; mostly after reading about how ChatGPT has caused a tidal wave of nearly undetectable cheating in high schools and colleges across the country. I can only imagine the reaction my parents would have had if I’d passed in a paper spat out by a word generator (I’d probably still be grounded) but it’s hard to fault young people when business, the media, everyone is saying AI is the future; their teachers may even be using it to make lesson plan or grade their papers.

So. As an experiment, I opened up ChatGPT and gave it the following prompt: “write a 750-word newspaper column in the style of the Maine Millennial who writes for the Portland Press Herald.”

I’ll be honest, the result was a little humbling. The AI was very good at aping my style: my flow, the parentheses, my little aside-jokes, even the thing where I stick a couple of factoids in the piece to back up my opinions. (I learned to do that writing for school, and I guess the AI learned by … copying me.) It used more clichés than I would, and the jokes didn’t really land (in my opinion) but if you weren’t super familiar with my writing, you might fall for it.

Reading it was weird, like what I imagine watching a method actor play you in a movie might feel like. You recognize your actions and mannerisms but it’s not you there. For the subject of the column, ChatGPT wrote about moving back in with my parents, very much a topic in my wheelhouse.

This little experiment didn’t make me want to jump aboard the AI train, but it certainly made me want to improve my writing.

I’m not against AI completely — there are various artificial intelligence models being used in medical research and training that seem to be more helpful than harmful. But the general-use, publicly available AI is not for me. If anything, this has made me even more against LLM usage, in part because of the steep environmental cost of artificial intelligence usage.

ChatGPT uses an enormous amount of energy to process its information and huge amounts of water must flow through the data servers to keep them from overheating. According to a study by the Washington Post and the University of California at Riverside, generating a 100-word email with ChatGPT requires about 519 milliliters of water, a little more than the average bottle.

By that estimate, my 750-word “column” consumed about 7.5 bottles of water. I keep myself well hydrated but it doesn’t take me that much water to write a column. The electrical power required to generate a 100-word email is reportedly about 0.14 kilowatt-hours of electricity, which is enough electricity to power 14 LED light bulbs for one hour. My generated column would have used enough juice to keep 105 LED light bulbs on for an hour. When I’m writing at home on my phone or laptop, for the record, I usually have two bulbs on, max.

Peddlers of AI software claim it will soon replace millions of various low-level employees. I doubt this. Just the other day, I did a Google search for “Nero guard dog U.S. Mint.” The “AI overview” at the top of the results, which I still haven’t figured out how to get rid of, did give me a brief paragraph about the animal in question (a dog purchased by the U.S. government in 1793 for $3, which was quite a bit back then). However, it picked an image of a marble bust of the Roman emperor Nero to accompany this fact … about an 18th-century American canine. So it hasn’t quite figured things out yet.

The data LLMs are trained on is often stolen. A couple of months ago, the Atlantic Magazine did an investigation into LibGen, a database of pirated books that Meta (the parent company of Facebook) used to train its own AI. Lawsuits have revealed that OpenAI, the maker of ChatGPT, has also used LibGen in the past.

Online, the Atlantic made it so you could search by author to see if their books had been pirated and used to train artificial intelligence without their permission or compensation. All of my mother’s published novels — in English and in German — were in that database. I don’t care how convenient it is, I’m not using tools that were built off my mom’s stolen labor and that burn up the environment to boot.