I'm curious, what's your stance on generative AI? I know we in the fandom community often talk about it in the context of AI created fanart, but I'm talking more in the context of the uses generative AI has in the realm of general work productivity, like what Microsoft is trying to do with their new CoPilot program.
Well, the ethical issue is basically the same as it is for A.I.-generated images—but for some goddamn reason, people don't like to think of any kind of writing as a form of propietary "art" the same way they do about visual arts, so it's garnered FAR less attention.
But as far as their usability goes? As someone who writes documents for a living, I can see these programs being potentially beneficial for creating early rough drafts, but that's the extent of what it's good for now: They can make outlines. BUT! You could get the same outline from a template or from an online boilerplate, so is that even worth anything? Once you go beyond an outline, any text generated from these A.I.s always needs heavy revision, reorganization, and editing. I'd spend less time just writing it from scratch.
Currently, generative so-called "A.I." programs that are designed to assist with writing text are based upon predicting what they think the user is requesting or desiring. They set out to give you what they believe you want, and accuracy is NOT part of the equation. This might not be as big of an issue if you're trying to make a book report on a classic novel, because there are probably enough examples of reliable web coverage on the subject it could reasonably generate something that's at least usable. But outside of doing some of your homework for you, how useful is it?
It can certainly bullshit some generic blather to fill space in a paper, or it could spew corporate-ese for the purpose of drafting a mass company email... but can it announce something new to your staff or the press in an accurate way? Nope. Can it reliably create copy or a script for advertising/marketing? Not if you want your ad to actually be true, let alone unique. :P
If you're doing something fairly rote like taking existing legal documentation to create a new, similar legal document for a different usage? You're better off just having a template on-hand with editable sections to revise; that way, the A.I. won't attempt to "improve" the legal text in a way that fucks you over. And if you're asking the A.I. not to edit that text in the first place... well, then why are we using this A.I. when we already have templates?
If you're hoping to create some kind of instructions, maybe a "How-To" book or a manual for something? Just forget it. It doesn't matter how much documentation on the subject you feed into that A.I. Ultimately, it will preconcieve how the process COULD work or what the program/device/person MIGHT do, and then it starts going off on bizarre claims/tangents that are wholly imagined. The longer the document you want, the worse the amount of nonsensical bullshit gets.
But even if you're just trying to get it to reduce a massive document down to like, a single page that covers the basics? It has no real system for judging what "the basics" are. You can try to specify to the A-not-I what you need to include, feed it the original document... and still wind up with a combination of falsehoods and excluded requirements. This won't necessarily happen every single attempt or in every single paragraph, but it'll definitely happen enough times to make it more trouble than it's worth. Still... this kind of thing — i.e., revising a single existing source into a different format or length — is probably the area that's the most promising application for these programs in the near term. It should be possible to "teach" to the programs in question, and it handily skirts past most ethical questions about the sources behind its knowledge.
What I said about falsehoods and skewed info/inaccuracy is also why search engines that have incorporated A.I. have gotten LESS reliable. Generative A.I. isn't truly "Artificial Intelligence," because it can't make any kind of judgment. It doesn't have a clue how to deem something true or false, and it's really fucking hard to build that into a program. Because ultimately, what do you ask it to do? How do you explain that to the program in a logical fashion? You can't just say "only believe the sources I give you/tell you to trust," because it only generates based on tons of pre-existing examples that it's observed. It only exists at all because of those examples, which is always going to cause these issues.
....and that last point ALSO raises the same exact ethical questions already brought up by A.I.-generated imagery. What right do they have to use these sources? Where are they getting them, etc.? And now I'm back where I started.
Suffice it to say I'm not a fan. Although I do, of course, have skin in this game, so I acknowledge that I'm definitely biased.
