Icon unrelated, I just like it and haven't used it in a while

[dhampyresa]

I was reminded to ask about this by a locked post and schneefink's recent post.

I have opinions about using AI [1] to produce stories and/or visual work, but they're not as supported/sourced/considered as I want them to be.

So I ask: What are your opinions? Have you read/watched/listened/etc to anything you thought was interesting on the subject, even ones you don't agree with?


[1] It's machine-learning, dammit.

Comments

[isis]

The Ezra Klein Show (a podcast, available wherever, transcripts on the NYT site) has been covering AI a lot over the last few months, talking with a lot of guests who are involved with AI/large language models etc, and I've found them super interesting. One point that was made (by Adrian Tchaikovsky, SF author) is that creators make things because they like to make things, and that joy in creation is not going to go away, so therefore, people are going to continue to write and create art even if AI-type things are producing stories and art. And that part of the joy of creation is that it's a conversation with the audience, and so readers approach stories written by people they know (and can have dialog with) differently from those written by people who are harder to dialog with (like pro writers, when you'd have to write them a letter), differently from those written by people who are no longer living (like Voltaire). The reader/viewer brings part of the conversation to the table and so that will be different for machine-created works.

[china_shop]

Oh, cool! Those episodes sound really interesting -- I'll check them out. :-)

[muccamukk]

As I said over at Schneefink's blog, I have profound concerns about the ethics of how the information is processed by underpaid and unsupported workers, and therefore if they should be used at all without that element being remedied. I have not used text generation for that reason, and don't plan to start.

I also agree with the linked post there that using a text generator at least somewhat negates the community aspect of writing fic, in that I see creation as a way to communicate with each other.

If point #1 can be fixed (which I don't think it can be), I don't object to them as prompt generators, or ways to get past writers' block, or tools of a similar level, creativity aid if you will. I also hear they're decent at writing for a niche kink if you just want smut to get off to, but again, point #1 with extra attention to the YIKES of all that.

[senmut]

Someone on tumblr commented that they will be glad to read machine-learning based stories when it is the machines telling their stories about their experience and perspective.

They, nor I, have interest in distilled formulas from bro-tech algorithmed chat programs set as "stories".

[kimaracretak]

I despise it to the point that I find it actively hard to talk about, honestly. There's nothing intelligent about it and honestly I think even calling it machine 'learning' is being overly charitable - there's no retention even if statistical models improve over time. It's ethically awful (plagiarism! for one!) and environmentally disastrous (see: bitcoin) and imo any potential benefits for individual creators are massively outweighed by what corporations + people acting in bad faith can and are already doing. Like I've worked in tech for over a decade, I know how shit the ethics here are overall but - jfc.

About content generation in uni classes, not fandom, but Bret Devereaux's essay on how text generation is harmful to students is my go-to thing to link. Long but so, so worth it.

Anyway, submission of works made by content-generation tools is fully banned in any fanwork exchange I run.

[china_shop]

I posted about my experiments using ChatGPT to generate short stories for personal language learning (under "Korean" in this post), but I didn't know about the human workers used to process stuff, mentioned by muccamukk. That definitely puts a different spin on things. /o\

[muccamukk]

The initial source on that: https://time.com/6247678/openai-chatgpt-kenya-workers/

I haven't been keeping close track of how this situation has developed since then. It's possible they've found away around human content moderation, but frankly I highly fucking doubt it.

[china_shop]

Thanks for the link!

Yeah, it sounds from the article like someone has to teach the algorithm what is taboo and what isn't... /o\

[melannen]

I think the work being done in getting an AI to generate a story is fundamentally different from the work done to write a story without one, and it's dishonest to claim authorship as if you'd written it. (But there's a fair amount of wiggle-room in between, it's a gradient not a line.)

I think there are major ethical (and legal!) issues with how the current large machine learning models are getting and processing their data and how they're making them available for use and all that's going to be a mess for a long time.... but that we're not far off a situation where there will be one that doesn't have those issues and is still capable of doing something like writing fic, so it's not an answer to the deeper question.

I do think it's a bit pointless to post AI-generated fic to something like AO3, because presumably if someone wants an AI generated fic for their fandom, they can generate their own? Like, I feel like, once readable AI-generated fiction becomes easy to get, it also becomes disposable in a way human-made isn't. There will probably be a transition period while people figure that out, though. (If someone puts enough human work into an AI-generated artwork that the average schmuck can't get something just as good with very little skill or effort, though, the answer's a bit different.)

[china_shop]

You don't think there'll always be a place for human curation? Idk. At the moment, ChatGPT is (in my very limited experience) producing stuff that's not very good, so there's an obvious niche there for someone to generate a ton, pick out some good ones and post them (ideally NOT to AO3!!), and for that curated content to be a bit like a tumblr feed.

In the longer run, even if machine learning can reliably produce "good"/satisfying stories, I feel like that just means our expectations will go up. There could still potentially be a blog or platform or whatever where people post a daily "best of". And people will read some of them and go, "Huh, what's so great about machines? I mean, I could have written that." You know?

It's a bit like digital photography, maybe? There are now a billion billion photos in the world (I'm guesstimating), and most of them represent reality very impressively/convincingly, but that doesn't mean every one of them is interesting to look at (especially out of context). Or that they're interchangeable. And the idea of sorting through for an interesting one is... I mean, I really appreciate other people going to that trouble so I don't have to!

[princessofgeeks]

I am not interested in reading/watching AI generated "art". I do not expect to change my opinion.

I am way more interested in supporting/watching/listening to actual people who have a hard enough time making art that gets any interest at all.

[ellenmillion]

I have strong opinions that I haven't distilled yet.

[sineala]

Coming in late, but if you haven't read Ted Chiang's article for the New Yorker about ChatGPT, I think it's worth a read. Also, I recommend the book You Look Like A Thing And I Love You, which came out before ChatGPT was unleashed on the world but I think it's still good, by that researcher who has that blog where she's been posting things like AI-generated ice cream names for years.

I have a moderate -- though outdated -- amount of academic experience with machine learning, and I have a lot of mostly negative opinions about the uses people are putting AI to and the ethical concerns involved in creating and curating the massive data sets involved here.