I've been seeing a lot of anti-AI sentiment which blames people for wanting what generative AI promised but never actually intended to deliver. These assertions tend to ignore that it was and remains entirely reasonable to want a lot of things that the people promoting generative AI keep insisting it can do. The problem isn't that people are bad for wanting these things, but that it doesn't actually work most of the time, and that when it does, it does so by stealing other people's work and by requiring water and power in quantities which are contributing to climate change and the destruction of the environment.
"what do you mean you can't write an email"
It's not ableist to be against generative AI, a thing which is expensive, resource-intensive, and doesn't actually do most of what it's marketed as being able to do.
It is frequently ableist to assert that anyone who *wants* any of what it promises is bad for doing so.
As someone with a cognitive disability which puts a 30% pass-filter on my writing, yeah, I actually would (and do) benefit from something that guesses my next word when I'm writing a work email. Every little bit that I draft while picking from a binary tree of options lets me write without forgetting as many of the pieces.
I make templates for myself of text I'll need more than once, with highlighting so I don't forget to address someone by name or say thanks.
I use text-to-speech and then clean up the (rampant) spelling errors and incorrect guesses in the transcription. I read things out loud, to myself and to other people, to find where I had the idea when I was typing, but my brain didn't let it get out in the moment and it takes several more reviews to figure out what I missed.
When having parallel conversations with friends, I copy and paste from one message thread to another, using my own words repeatedly instead of typing them anew.
I go away, I come back later, I dictate incoherent paragraphs and then tease them into something resembling the complicated thought I had in my head which dysgraphia stripped down into a vague idea that there was enough to fill a list, I'm sure there was.
I write things out of order because I don't know in the moment what the full structure will be in the end, except that the best I can do, even with all these compensating strategies, is 80% of what I wanted.
But it's better than 30%, or than giving up when I see something someone wrote and I want to say more, I want to respond to it, but I don't have the spoons, they turn to sieves in my hands and everything I wanted to say fades, shattering into pieces that don't feel worthwhile.
Emojis, gifs, and memes make things easier. A single symbol is a whole concept, de-compressing some of what dysgraphia crumpled.
Except if I post an image it needs alt text, requiring writing at the very moment that I'm least able to do so. This isn't negotiable either, because just as I need the accessibility of reaching for a meme to say more than nothing, blind or low-vision users deserve the reciprocal courtesy if at all possible.
Even there, a meme is better than nothing, because I don't have to describe the whole shape of my reaction when I can say that it's the "surprised pikachu" meme with a certain label, or the "is this a butterfly" meme with a twist. Even if they don't know what it looks like they'll know (or can learn, just as I did) what it *means*.
When people say that no one wants predictive text or to have something write a paragraph for them, or that they're bad for wanting those things, that's just not true. Something that offers up generic email language for your generic office email, drawing solely from the hundreds of thousands of previous, generic, office emails it has, that thing is useful. Autocorrect where you can adjust your personal settings to tell it you do, really, swear that much, or that a specific word isn't a misspelling and you do want it added to your computer's personal dictionary, or even templates online that give you the format and language examples for a specific kind of formal letter that most people write only a few times a decade, at most... these are useful things that people aren't bad for wanting.
Templates, models, resources, examples, drafting guides, they existed before, if you knew where to look.
Generative AI and models like ChatGPT appeared, backed by grifters and con artists, to promise people who didn't know where to look that all they needed to do was write a few words and press a button.
We weren't bad for wanting it to be real.
Generative AI is ruining the tools which used to help me with accessibility. The things I used to use to help me are getting worse, because they were made for specific scenarios and worked fine, but now they're getting replaced or melded with something that doesn't understand context and isn't proscriptively bound in lieu of that understanding.
My generic work email isn't improved by drawing from everything ever said on Reddit. Predictive text which pulls from everyone instead of what I write frequently no longer knows how to spell my last name, it offers that perhaps I meant "Robin Hood" instead of myself. Putting generative AI into formerly useful tools and ruining search engines with incoherent text that mimics the average internet post doesn't help in a reliable way. The only thing it does reliably is steal and pollute. The easiest way to not sound average is to ask the black box to copy someone in particular, doubling down its thievery.
It's okay if you have trouble sending that email. You're not bad for wanting something that would help you. But ChatGPT and its various clones/embeds aren't that help.
Comments
Post a Comment