No AI

What follows is an excellent argument against generative AI from Tumblr user mikkeneko. Due to the unstable (and largely illegible, particularly on mobile) nature of Tumblr as a platform, I'm archiving it for future reference.

nonbinaryelphaba says:

generative AI literally makes me feel like a boomer. people start talking about how it can be good to help you brainstorm ideas and i’m like oh you’re letting a computer do the hard work and thinking for you???

headspace-hotel replies:

There are many difficult things that were replaced with technology, and it wasn't a bad thing. Washing machine replaces washing clothes by hand. Nothing wrong with that. Spinning wheel replaces drop spindle. Nothing wrong with that.

Generative AI replaces thinking. The ability to think for yourself will always be important. People that want to control and oppress you want to limit your ability to think for yourself as much as possible, but continuing to practice it allows you to resist them.

mikkeneko continues this thread:

“This tool replaces thinking,” is a technology problem we (humans) have faced before. It’s a snark that I’ve seen pro-AI contenders take as well: I bet these same people would have complained about calculators! And books!

Well. They did, at the time.

We have records from centuries – even millennia back – of scholars at the time complaining that these new-fangled “books” were turning their students lazy; why, they can barely recite any poems in their entirety any more! And there are people still alive today who remember life before widely available calculators, and some of them complained – then and now – that bringing them into schools dealt a ruinous blow to math education, and now these young people don’t even know how to use a slide-rule.

And the thing is:

They weren’t wrong.

The human brain can, when called on, perform incredible feats of memorization. Bards and skalds of old could memorize and recite poems and epics that were thousands of lines long. This is a skill that is largely lost to most of the population. It’s not needed any more, and so it is not practiced.

There is a definite generational gap, between the people who were trained on slide-rules and reckoning and the generation that was taught on calculators. There came a year, when that first generation grew up and entered the workforce, when you suddenly started encountering grown adults who could not do math – not even the very basic arithmetic needed to count down from one hundred. I would go into a shop, buy an item for sixteen dollars, give the cashier a twenty and a one because I want a fiver back, and have them stare at the money in incomprehension – what do? They don’t know how to subtract sixteen from twenty-one. They don’t know how to calculate a fifteen-percent tip. They did not exercise the parts of their brain that handle this, because they always had a calculator to do it for them.

Nowadays, newer point-of-sale machines compensate for this; they will automatically calculate and dispense the change, no subtraction necessary on the part of the operator. Nowadays everyone carries a phone, and every phone carries a calculator, so if you need to do these calculations, the tool is right there. As more and more transactions go electronic and card, and cash fades further and further out of daily life, these situations happen less and less; it’s not a problem that most people can’t do math (until it is.)

The people who complained that these tools-that-replace-thinking would reduce the ability of the broad population to exercise these cognitive skills weren’t wrong. It’s simply that, as the pace of life changed, the environment changed so that in day-to-day life these skills were largely unnecessary.

So.

Isn’t this, ChatGPT and Generative AI, just the latest in a long series of tool-replaces-thought that has, broadly, worked out well for us? What’s different about this?

Well, two things are different.

1) In the previous instances of tool-replaces-thinking, the cognitive skill that it replaced was a discrete and, on a day-to-day basis, unnecessary outlay of energy. Most people don’t need to memorize thousands of lines of poetry, or anything else for that matter. Most people don’t need to do more than cursory levels of math on a day to day basis.

This, however, is different. The cognitive skill that is being obsoleted here is more than “how to write essay” or “identify what is the capital of Rhode Island.” It encompasses the entire field of being able to generate new thoughts; of being able to consider and analyze new information; of being able to follow logical trains to their conclusions; of being able to order your thoughts to construct rational arguments; or indeed of being able to express yourself in any structured way. These cognitive tools are not occasional use; they are every day, all the time.

2) In the previous instances of tool-replaces-thinking, the tool was good at what it did.

Calculators may have replaced reckoning, but calculators are also pretty good at what they do. The calculator will, as long as you give the right input, give the right answer. ChatGPT cannot be relied on to do this. ChatGPT will tell you, confidently and unhesitantly and dangerously, that 2+2=5, and it will not care that it is wrong.

Books may have replaced memorization, and books certainly could be wrong; but a fact, once in a book, is pretty stable and steady. There is not a risk that the Guy Who Owns All The Encylopedias might wake up one day and decide – to pick a purely hypothetical example – that the Gulf of Mexico is called something else, and suddenly all the encyclopedias say that.

Generative AI fails on both these counts. It fails on every count. It’s inaccurate, it’s unethical, it’s unreliable, it’s wrong.

---

I remember some time ago seeing someone say (it was a video about medieval footwear, actually) that “humans have a great energy-saving system: if we can be lazy about something, we are.”

This is not a ethical judgment about humans; this is how life works. Animals – including humans – will not do something the hard way if they can do it the easy way; this basic principle of conservation of resources is universal and morally neutral. Cognition is biologically expensive, and though our environment is not what it once was, every person still goes through every day choosing what is valuable enough to expend resources on and what is not.

Because of this, I don’t know if there is any solution, here. I think pushing back against the downhill flush of the-easy-way-out is a battle both uphill and against the tide.

So I’ll just close with this warning, instead:

Generative AI is a tool that cannot be trusted. Do not use it to replace thought.

Tags: scrapbook, no-ai

Last modified: May 13 2025 11:07:04 EDT

AI

💡 Synonyms

diffusion, genAI, GPT, LLM

Nothing on this site will ever knowingly be made with the use of generative AI. All of the text, except when quoting someone else, was written entirely by me. I occasionally use permissively-licensed resources from sites like openclipart and OpenGameArt. To the best of my knowledge, nothing I use was made by generative AI. If you know of any such material here please let me know.

My work is permissively licensed for use by other people. Nothing on this site may be scraped or used in any way for the purpose of training large language models or other machine learning applications. Permission will not be given.

My position on generative AI boils down to two simple questions: (1) what does it do that is good for people, and (2) does the good outweigh the harm it causes? If the answer to 1 is "someday it'll XYZ..." then: cool. Let the research continue in safe, controlled environments. Like all scientific research, it should only be done if it can be done ethically. If "someday" arrives and it can do all these great things for people while minimizing harm, then you can release it.

Even if we accept that generative AI will be a technological revolution bigger than the printing press that will lead to new horizons in human consciousness-my opinion is that it won't, but even supposing it will-we still shouldn't let it keep hurting people in the meantime. If the printing press had some technical fault that made every third one blow up and kill whoever used it, Gutenberg wouldn't have let people use it. He would've kept working on it until it was safe to use, no matter how big of a sea change it would eventually be for human knowledge and communication.

I have deeper philosophical objections to generative AI. Tumblr user mikkeneko wrote an excellent post about this, which I have archived here; but these arguments should be unnecessary. The current state of GenAI should be obviously bad for plain, practical reasons. The harms are many and the benefit is theoretical. It is a scam.

Tags: topics, no-ai

The Price of Eggs, the Value of Taste

Thursday March 20th, 2025

Tags: archive, cats, currents, food, music, no-ai, personal, politics

One of my coworkers has an entire carton of eggs in the shared fridge. I don't know what they intend to do with them. We don't have a proper kitchen here. Maybe they have a little electric hob or griddle they plug in at their desk. It seems like an awfully inconvenient way to eat eggs. Maybe the jacked-up cost of eggs has given them a perceived value beyond their actual worth. They must be worth all the hassle if they cost nearly $8/doz, right?

Or maybe they're not here to be eaten. Maybe whoever it is is simply trying to keep the precious ovoids away from whatever greedy hands are fixing to crack them.

Me, I don't really get it. I like eggs fine, but their main value to me is as a cheap source of protein. If they're no longer cheap, I can get protein from other sources. Eggs have a little vitamin D, some B6, a tiny bit of iron, but there are plenty of other cheap ways to get those too. I guess I'd be upset if eggs were some incredible delicacy, but I don't think they're that tasty. To me they're mostly a vehicle delivery for salt and pepper. To each their own, I suppose.

I'm not saying the skyrocketing cost of eggs isn't annoying, I just don't think it's as big a deal as a lot of people seem to think. For example, I don't think it justifies allowing a fascist takeover of my country's government. If the choice is paying a lot for eggs or voting in the fascists—it wasn't, but even if it was—I'd be perfectly fine eating more chicken salad until prices go back to normal.

There's No Accounting for Taste

This is one of those clichés that's been repeated until it's lost all meaning, but I think it's a good way to describe the ways recommendation algorithms and "AI art" leave me wanting. Computers are good at dealing with quantifiable information, and the qualia, the subjective internal experiences that define our artistic tastes, are unquantifiable. There are no formulae a computer can understand that will communicate the reasons I like or don't like a particular piece of art.

There are a million examples I could pick, but let's take music. I intensely dislike the band Rush. On paper, one might think they'd be in my wheelhouse. I like progressive rock. I like complicated guitar stuff. I like vocalists with unique and unpolished voices. But I can't stand Rush. They never did anything for me. My dislike could stem from the fact that they have right-libertarian beliefs and wrote an album inspired by Ayn Rand, but I didn't like them before I knew that about them.¹

A computer could look at the fact that I like classic prog like Pink Floyd and Jethro Tull, and epic guitar/synth shreds from jam bands like Phish and The Disco Biscuits; and it might conclude oh, of course this person likes Rush. There's no way they couldn't. These elements add up to a perfectly Rush-shaped sum. But I don't like Rush. There's no accounting for taste.

There's no accounting for why "AI art" makes my skin crawl. A computer can look at a bunch of disparate elements and add them up into what its programming concludes is the most logically perfect distillate of art. But that has no bearing on whether it's something humans want to look at. There's no function that defines what makes people feel things.

Now, I'm a materialist.² I don't think anything about this process is the result of a soul, or a god, or anything supernatural.³ But that doesn't mean I think all of our subjective experiences can be quantified and codified. I think there are processes in the brain that we don't understand on a technical level, and are probably unknowable. And that's cool. It means philosophy and art can never be "solved". Our quest to understand and express ourselves is a constantly-evolving journey of learning and introspection. If it were possible to program a computer to make perfectly enjoyable human art, we'd lose an important source of meaning in our lives. Thank goodness it's all bullshit

Cat Food Flavors

I don't know why they sell cat food in flavors meant to appeal to humans. "Mixed Grill"? "Salmon Dinner"? "Prime Filets"? I think my cat would enjoy it just as much even if the can was labeled "Poop". And, well, from my perspective that would be a more accurate way to describe how it smells. Sunny will still eat it up, though. She likes other cheap proteins, like chicken and tuna, but she'll also go bananas for various whimsical flavors of nauseating meat paste. There's no accounting for taste 🦝

Like 👍 Share 📣 Save 🏛️ Reply 💌

Like 👍️ Share 📣 Reply 💌

IP Theater, Congrats to Rabbit, Bloggodynamism

Thursday May 01st, 2025

Tags: blog, copyright, correspondence, crafting, meta, no-ai

I was thinking about the art I use from openclipart and opengameart and it made me feel bad about publishing with a noderivs license. Having public domain art to draw from adds a lot of life to this website and my other projects, and it would be cool if someone liked one of my stories and wanted to turn it into a comic, or read it on a podcast, or use it as inspiration for a game. I'd be proud if something I made could contribute to open culture, too; I just put the noderivs clause on there as a feeble protest against AI scraping. But this is kind of silly. Technically the "non-commercial" clause should protect me, now that OpenAI is no longer a "non-profit" there's no legal argument that their activity is non-commercial, but the sad reality is that corporations are going to do whatever the fuck they want and I as an individual have no legal recourse, because the system is designed to protect capital and nothing else. So fuck it, I'm going to switch to CC-BY-SA. The only thing ND does is discourage the people who might actually get some value from my work and use it ethically. I'll keep the no-ai disclaimer even though I don't think it legally does anything; I can't legally do anything, anyway. Maybe someday I'll be part of a class-action brought by the EFF or something, that's the best I can hope for. Maybe I should just go full public domain with everything. In practice this is all just intellectual property theater. Legal cosplay. I'm not a member of the class that's allowed protection so I get exploited. That's the game I'm forced to play, so I might as well try to help my team.

Congrats to Rabbit!

Rabbit at jackalope.city finished 100 days to offload! Completing a challenge to blog 100 days in a year doesn't sound that hard on paper, it only requires a 27.4% posting frequency, but if you're not used to that amount of output, it can be quite an adjustment. It takes moxy to stick with a new habit. It takes grit to stay self-motivated. It takes courage to write even on days it doesn't come easy. Well done!

I finished the challenge in July 2022, and in typical me fashion, I didn't tag any of my posts or really talk about it much until I was done. I don't know why I'm so resistant to self-congratulation; I have this knee-jeek reaction that I would be perceived as conceited or narcissistic, but I don't think it's conceited or narcissistic when other people are proud of their accomplishments. Maybe I should try the Be Kind To Myself For 100 Days challenge.

My 100-day challenge was the self-imposed kick in the butt I needed to think of myself as a writer again. Although my output slowed down in 2023 and significantly in 2024, the bug is still with me, hopefully for good.

I'm not doing any kind of official challenge this year, but I'm well on my way to 100 posts again: 40 posts already and it's only May. I think I'm going to be more motivated to write now that I'm doing it on my own website, too. I don't have to worry about spamming any feeds, because I'm not sharing it with anyone else. I have free reign over my domain! Oops, there I go thinking like a landlord. It's hard not to when most of the rest of the internet treats us like digital serfs.

Bloggodynamism

So, I went kind of ham with the custom dynamic variables: I have not one but three custom emoji: pat, nb and lander give you ¹

I'm sure I'll add a few more, but I'll try to use them sparingly in the blog: they probably won't show up right in the feed, because they rely on external stylesheets for the formatting. You can't just drop an img src tag into the middle of a sentence and call it good, you have to make a tiny div that uses the image as a background if you want to be able to size and position it properly. Way too much hassle to do it manually, but a piece of cake with the power of PHP.

But, it kind of leaves feed readers out in the cold, so I'll try not to rely too heavily on them. In fact, most of this stuff is only going to be relevant to page visitors, sorry! Feed readers can skip this section.

I created variables for each of the buttons at the bottom of the blog post: likebutton, sharebutton and replybutton. Just in case I ever want to use them individually. Hey, it could happen! But I also made pagebuttons so I can drop in all three at once, which is what I'll be doing all the time.

This wasn't that much friction before, because I could include the code in my post template, but now all that ugly HTML is hidden away in the guts of the computer, instead of messing up my nice orderly markdown files. Plus, if I want to change something about them, I can do so on every page at once! If I decide I don't want them, I can replace them with an empty string. Anything I want repeated over and over, it's always better to have a function rather than doing it by hand.

I also created blogtop, which generates the entire top section of the blog post: the h1 title (including the pagetitle id attribute that makes the share button work), the block quote div, the date and the tags. For pages, there's pagebottom which shows a blockquote div with the tags and the date the page was last modified. Ideally, there shouldn't be a single page that you look at and wish you knew when it was from. I find that sort of thing frustrating. There are pages where it's not necessary, I don't think anyone is going to look at the contact page and think "what's the context for this??!", and it seems excessive to use the full pagebottom on every single topic page. But all the main stuff should have some form of date stamp.

I also cleaned up the buttons: the share button previously broke everything if it was on a page without a pagetitle id, but now if that attribute doesn't exist, it'll just copy the URL. To encourage inter-blog communication, the reply button will now show you a field to enter a URL to ping me from your own site (a great idea shamelessly stolen from Rabbit), with a link to the contact form if you want to send a message. I'm not using webmentions or anything, it just uses the same script as the contact form and guestbook to send me an email. Webmentions would be cool if every blog was running them and set up to automatically ping every other blog that's mentioned, but since that's not a thing, it might as well just be an email. Maybe the glorious webmention future will arrive someday.

Rather than post all the new code, I'll just link to the repository I set up as a site backup. It's a snapshot of all the HTML generated in static mode, all the markdown files and all the bits of custom code in conveniently-labeled folders. The idea is to do backups on the last day of each month, so if anything happens and I lose my webhosting again, I won't lose that much history. I kinda wish I was doing that with my oddmuse site, there's probably some stuff in there that I would like to have kept maybe I'll try to do weekly backups of my markdown...

Anyway, there are only two more features I want to hack in to call it "done":² I want the "what's new" section to automatically display the dates and titles of the three most recent blog posts with links to view more,³ and I want the /blog landing page to automagically show links to all the blog posts. I use a lot of images and footnotes, so I don't think doing a simple tag sort will be the best way to browse the entire blog, especially once I have hundreds of posts here.

I'd prefer if tag search worked more like bear, where it just shows you links to everything with that tag instead of full pages, and I'd like it if I could have a page that shows every tag. There's no easy way to do that with kiki. Which isn't a big deal right now, but you know, I'd like this to be my forever home. Someday, muse willing, this site will contain vast multitudes. I think I'll be able to hack all this in, I feel more confident after messing with dynamic variables and looking at the pagegen code. It'll be a lot more complicated, but it's doable. I know how to do a for loop. I can see which part of the code looks at the page tags and how. I know how to put a div together piece by piece. The main problem is figuring out at which point in the program the variables I need to work with are in scope. It's challenging to figure out by reading the code which variables I can use where.

I know I can do it, but I really, really need to take a break from programming now. I love it but it has taken over my brain. I have a site that I'm really happy with, it's time to use and enjoy it

Like 👍 Share 📣 Save 🏛️ Reply 💌