TVTropes Now available in the app store!
Open

Follow TV Tropes

Following

Artificial Intelligence discussion

Go To

With how much artificial intelligence has been improving, in many areas such as text reading/generation, picture reading, picture generation, convincing voice synthesis and more, I think there's a lot that can be discussed, about the effects that this technology will have on society.

I'll start off with one example.

I'd been thinking about the enshittification cycle of tech, and I think it's coming for Google hard. The search engine just isn't so great at finding what you actually want, and I think that's gonna leave a big opening for Bing with their use of AI. If the AI can sift through the crap and actually find what you want for real, due to its understanding of language, it'll actually make searching super useful again.

In the pre-Google internet, search engines used to search only for exact words and phrases, which had its uses, but also meant finding a lot of sites that simply crammed in a lot of popular words and phrases to get visitors. Google cut through the crap with a better understanding of how to "rank" sites relative to how relevant they are, and even find sites that are on the topic you were looking for without using the same exact words.

But Google started to become more advertiser-friendly, then later, more shareholder-friendly. There's a limit to how much one can make their product built entirely around shareholder growth, so as it turns to crap, it leaves an opening for a competitor to show up.

Since Bing/ChatGPT (which Bing is plugged into now) understands the use of language, it can actually understand context and determine relevance based on that. And that'll make it huge, I think. Context-based understanding of web pages can potentially do an excellent job of finding what people actually want, in a way that goes way beyond Google's page ranking systems, or the examination of exact words.


There is also a more specific thread for the legality, ethics and nature of AI art. Posts on this aren't off-topic here, but may sometimes be more appropriate to the other thread.

Edited by Mrph1 on Jun 22nd 2024 at 11:53:33 AM

Zendervai Since: Oct, 2009
#276: Feb 20th 2024 at 7:56:29 AM

[up] The niches it's useful for are repetitive iterative tasks that humans don't have the attention spans for. Like "we need to identify every form this particular protein can take". That can be done with AI easily and it means that you don't get duplications or skipped iterations because the human literally can't focus enough long term on the job.

An AI can also run physics or chemistry simulations much more rapidly than a human can because it's easy to put in a "this is the result we want" condition and the AI just keeps going until it's achieved the result or hit the number of tests chosen.

These tasks being automated doesn't actually risk any real jobs because they tend to be dumped on unpaid interns anyway and if an AI is quietly handling them in the background, the interns can be doing stuff that helps them actually learn and the researchers can focus on the results and not spending 9 months just folding the same protein in seven million different ways.

It's also good for clickbait.

It's not much good for long-form content or for anything that needs to be factual or that needs to have a specific perspective. And it really does seem to be true that the hallucination or distortion problem is never going to fully go away. Because both require the AI to really understand what it's talking about and, to be blunt, that's not going to happen anytime soon, because the AI having true understanding isn't necessary and no one's really putting any effort into it.

Edited by Zendervai on Feb 20th 2024 at 10:59:03 AM

Lyendith Since: Mar, 2011
#277: Feb 20th 2024 at 7:58:10 AM

[up] Now that’s more interesting and useful. I hadn’t thought of that.

Zendervai Since: Oct, 2009
#278: Feb 20th 2024 at 8:00:06 AM

A lot of the problem we're having is people approaching the situation backwards. It's not "hey, here's a thing that could really benefit from AI, let's make an AI for it." It's "we developed an AI, where can we hammer it into place so that people will give us money for it."

BonsaiForest Since: Jan, 2001
#279: Feb 20th 2024 at 8:08:07 AM

Sometimes things are just invented that way. "Here's something we can do. Let's see if we can create it. Now that we did it, we need to find a use for it."

I don't think AI should be crammed into everything, but I do think more uses will be discovered over time.

Zendervai Since: Oct, 2009
#280: Feb 20th 2024 at 8:12:23 AM

Maybe, but I think we've legit kinda maxed out where generative AI can go. Because like..."this thing can approximate what you want in a vague sort of way and it makes a lot of mistakes that require editing" isn't particularly interesting.

It also runs into the problem of this question:

"Why should I care about what you created, if you don't care enough to actually do it yourself?" Especially since I've never seen anyone be actually passionate about a specific AI output. It's so disposable that it's hard for even the loudest and most obnoxious pro-AI people to care about a specific output for more than five minutes.

Lyendith Since: Mar, 2011
#281: Feb 20th 2024 at 8:18:05 AM

[up] That’s the thing, yeah. Why would I care about art that no one actually put any thought into? To analyse and criticize art, you need to be aware of the author’s intent and how well it’s executed. If there is no author intent outside of a vague concept, then all you can judge is the "wow" effect of the technological achievement, and that will wear off very quickly.

My fear is that… that would be enough for a lot of people.

And as a professional translator, I’m afraid that GAI could produce literary translations that are just passable (if bland) enough that some publishers would feel they don’t need humans like me to do it anymore.

Edited by Lyendith on Feb 20th 2024 at 5:19:38 PM

Zendervai Since: Oct, 2009
#282: Feb 20th 2024 at 8:24:07 AM

[up] Well...there's been some companies trying to use those AI translators on a professional level, and it's gone badly.

Like, there's a manga called Centaurs? It got an AI translation and not only does it have constant problems with grammar (it seems to be trying to replicate Japanese grammar in English but also doing proper English grammar at the same time, so half the sentences sound bizarre and the other half have no personality), it's also translating terms wrong because it's not catching obvious context. Like, IIRC, the Japanese word for foal and calf are the same, and the AI is like "a baby centaur must be a calf" which...no. And it's repeatedly getting the main character's name wrong because it latched onto the wrong meaning of the characters in her name. I don't really get how these problems can be fixed if the company can't be bothered to edit it.

And AI trying to translate anything involving wordplay is just completely hopeless and it's incredibly unlikely that hurdle will ever be overcome, because it'd require the AI to be able to recognize that the word it's translating isn't actually the word it's translating and that it's a different word.

I'm actually kind of curious about what an AI would do with the Abbott and Costello "who's on first" routine if it's translating into a language that doesn't quite fit the necessary structure.

Edited by Zendervai on Feb 20th 2024 at 12:02:47 PM

Imca (Veteran)
#283: Feb 20th 2024 at 9:13:26 AM

Even in artistic feilds AI is at an industry adoption rate of 31% of people using it on the last survery.

While artists who mostly go from commission to commission are panicking, the ones who do industrial scale tasks have adopted it for things like shading and repetitive background objects like distant billboards where the framing and theming matered more then the content of a fake add that no one can even read(both uses I have personaly seen first hand from my coworkers)

Then there is the aforementioned use of it in spiderverse.

The thing is that like once your dealing with that kind of work people generaly just be quiet about it, it's another tool in the closet rather then something to male a fuss about.

PS: This is why steam droped the ban on AI game content, there audit showed that over 40% of products made in the last year used it at some point in the pipeline.

It's not just a hit button get finished item thing like CEOs want it to be.

Edited by Imca on Feb 21st 2024 at 2:24:16 AM

Zendervai Since: Oct, 2009
#284: Feb 20th 2024 at 9:19:54 AM

Although, it is worth noting that the 31% rate of adoption doesn't actually match the 5% of revenue each company is reporting, which actually matches the revenue rate from before these programs became available. It's not actually saving or making these companies any money.

Because it turns out that it's not really a good way to save money and a lot of uses are counterproductive, like the repeated attempts to use a chatbot as customer service.

RainehDaze Nero Fangirl (4 Score & 7 Years Ago)
Nero Fangirl
#285: Feb 20th 2024 at 9:24:14 AM

Why would the expectation be an immediate cost saving or increase in revenue? If you make people long term more efficient with a new tool, that doesn't mean you're going to see a big change in only a year or something. Even if you now have extra capacity...

Imca (Veteran)
#286: Feb 20th 2024 at 9:26:01 AM

[up][up] Not exactly surprising to me, especially since again this is anecdotal but at least in my work environment, all uptick of the tools have been by the workers themself, not from management down.

Bottom up adoption tends to prioritize making the workflow more comfortable, not to make the project more money.

Edited by Imca on Feb 21st 2024 at 2:26:39 AM

Zendervai Since: Oct, 2009
#287: Feb 20th 2024 at 9:30:37 AM

[up][up] It is interesting though that there's so much hype and the actual impact on the bottom line of these companies appears to be nothing, which kinda indicates that the AI assistance is just replicating programs the company already had with no impact on the output or the workforce.

This also explains why none of the AI companies seem to have any meaningful success stories about corporate AI use to point to. Because on the company side of things, effectively nothing happened.

archonspeaks Since: Jun, 2013
#288: Feb 20th 2024 at 9:34:44 AM

Maybe look into the specific example I mentioned there? It's almost like I said a specific movie series does that in places because a specific movie series does that in places or something. Also, the CGI equivalent of smear frames are a thing now. How do you do that if you aren't constantly resculpting the model?

Spiderverse, the specific example you mentioned, made their signature style in part by making each line in an effect or on a character its own model with unique rigging. They used AI to control the motion of all these models, something that would otherwise normally be prohibitively labor intensive. They also used AI to procedurally simulate the variation in hand-drawn lines.

Smear frames are done in CG by stretching a model [1] in between keyframes, not by “remaking” a model frame to frame. In the time since that paper I just linked was written, AI has allowed for convincing procedural generation of smear frames, with tools to do so being integrated into several cinema-grade 3D suites. The animation field has been making extensive use of AI for years now, so I’m not sure where the resistance to AI in animation is coming from. It’s certainly not something most animators would agree with.

They should have sent a poet.
RainehDaze Nero Fangirl (4 Score & 7 Years Ago)
Nero Fangirl
#289: Feb 20th 2024 at 9:39:10 AM

[up][up][up] Mmm, over here there's definitely some higher-level buy in for testing it, but that's probably also why it's a limited-scope trial to see if it does make a difference. Or at least helps developers enough it's worth using.

[up][up] I mean, from a programming perspective the purpose seems like every other bit of automation, suggestions, and generation: you're not going to see a huge difference, but it does make doing repetitive tasks less of a pain.

Edited by RainehDaze on Feb 20th 2024 at 5:40:53 PM

Zendervai Since: Oct, 2009
#290: Feb 20th 2024 at 9:43:41 AM

[up][up] The one where I was talking about the resculpting wasn't Spider-Verse. It was the other one I mentioned.

[up] I do think it's telling that the really hyped up stuff appears to be going nowhere, and it's the really quiet stuff where it's being accepted. Like, OpenAI can't advertise by going "it makes the lives of some of these people a little bit easier!" because that doesn't hype up the investors. So they just have to keep hammering the art and text generative AI even though it really does seem like it has no future as anything but a toy. For it to really go anywhere and have any commercial viability as a big money maker, it has to do more than just sorta approximate what the user wants via prompts. I also think with Sora, there's a very big question that OpenAI is very clearly trying to dodge and really doesn't want to answer.

How long does each prompt actually take to generate? And if it really is being generated in super low resolution and then gets really heavily upscaled, how long does each step take? There's speculation that a Sora prompt might well take significantly longer that what might be considered acceptable, and if the output basically has to be hugely upscaled to be viable, that's much less impressive. (I'm pretty confident on this, because upscaling artifacts are pretty distinctive).

This is the one product OpenAI has put out where all of their examples are internal. They're usually really happy to let people test it before release and let them make videos that get released at the same time as the product, but here, the company is doing some real work to prevent anyone from being able to pin down anything about the production pipeline. They might release it to the public, but I would not be surprised if it just kinda fails to materialize beyond the occasional hype video.

Edited by Zendervai on Feb 20th 2024 at 12:54:37 PM

RainehDaze Nero Fangirl (4 Score & 7 Years Ago)
Nero Fangirl
#291: Feb 20th 2024 at 9:55:42 AM

That describes basically every incremental technological advancement ever. The futuristic vision is still years off, but the foundation for it is still solid and useful (although the integration is sometimes a bit lacking). AI is special in this regard, apart from the eternal resurgence of arguments it will never amount to anything when the stuff it's doing more quietly is already making significant inroads.

Zendervai Since: Oct, 2009
#292: Feb 20th 2024 at 9:57:41 AM

I think a lot of the problem is just that you really do get people being really, incredibly unpleasant about the art/text generative side of it and it just...not being very good at it.

There's also stuff like some recent buzz about how people not confident in English are writing papers and then are attempting to revamp the papers with ChatGPT and then the result is basically incomprehensible. Or that students are failing by using AI generated assignments because it turns out that chatbot generated essays can be really easy to pick out, but also because they're not reading the output properly and aren't catching that it goes totally off topic halfway through.

AI has its uses, but the things that it's being championed for are probably not where it'll stick in the long run. It's just, like I said, that the stuff it's actually useful for doesn't sound exciting or interesting to the investor class, so these companies have to hype up the features it kinda sucks at.

Again, it's really hard to come up with viable major use cases for "a product that can sorta approximate what you want". Especially lucrative ones.

Edited by Zendervai on Feb 20th 2024 at 2:22:59 PM

RedSavant Since: Jan, 2001
#293: Feb 20th 2024 at 11:16:49 PM

Unfortunately, there's enough of a culture war around translation and localization that I imagine we are going to see consumers declaring that they'd rather have janky translations that fit their own pre-determined ideas of what a text says rather than translations that have been written in a naturalistic style. In fact, we already are seeing that. Too many readers nowadays think that the mark of a good translation is that it sounds kind of busted and still pointedly foreign, rather than that it's written in fluent and characterful English.

Hard to say what kind of impact it'll have on sales, though publishing companies' decisions to use gen AI in translating official manga releases like Ancient Magus Bride is a worrying sign.

Edited by RedSavant on Feb 21st 2024 at 4:17:53 AM

It's been fun.
Cordite-455 (Troper Knight) Relationship Status: Complex: I'm real, they are imaginary
#294: Feb 20th 2024 at 11:30:20 PM

I honestly think AI only does harm when used for art and culture things like literature, painting, music, and any other creative stuff

AI should just be used for working out stuff like cure for cancer and how to create fusion, if that.

My honest opinion is that we should just ban AI entirely, because it'll ruin us in the long run. Whether that be us growing too reliant on it and it destroying our culture, or in a sci-fi way where AI genocides humanity for inefficiency. (I know it's an extreme position, but it's genuinely how I feel about this whole AI thing.)

RainehDaze Nero Fangirl (4 Score & 7 Years Ago)
Nero Fangirl
#295: Feb 20th 2024 at 11:37:44 PM

What.

No, we should not throw out our single best tool for almost any data heavy processing because of some nebulous risk of "culture loss" or fretting about paperclip maximisers. Do you have any idea how many things that would already set us back on? Like vaccines for that whole global pandemic we had?

Protagonist506 from Oregon Since: Dec, 2013 Relationship Status: Chocolate!
#296: Feb 20th 2024 at 11:43:11 PM

To put it this way: The COVID-19 Vaccine was created in large part using Artificial Intelligence.

Leviticus 19:34
archonspeaks Since: Jun, 2013
#297: Feb 21st 2024 at 12:02:12 AM

I’ve seen several arguments from fine artists that AI may actually enhance their creativity. The argument goes that because fine artists are essentially beholden to the gallery system, they are required to make whatever collectors want to buy and not what they want to make. By automating the production of gallery art for collectors, the artists would then be free to focus on their own passion projects.

The one where I was talking about the resculpting wasn't Spider-Verse. It was the other one I mentioned.

If you’re referring to Hotel Transylvania, the animators used extreme rig deformation and squash and stretch effects on textures to create their signature look. In no way were rigs or models being “remade frame to frame”. [1]

They should have sent a poet.
Xopher001 Since: Jul, 2012
#298: Feb 21st 2024 at 1:08:23 AM

[up]That's certainly an interesting argument, but it assumes fine art collectors would see value in something generated by an algorithm. From what I know people in those circled have a stigma against 'mass produced crap'.

RedSavant Since: Jan, 2001
#299: Feb 21st 2024 at 1:15:00 AM

[up] At least until someone attaches a random number generator output to each one and sells them as NFTs. The fine art world seems like a ripe sphere to target for that kind of scam.

It's been fun.
editerguy from Australia Since: Jan, 2013 Relationship Status: You cannot grasp the true form
#300: Feb 21st 2024 at 2:48:43 AM

Unfortunately, there's enough of a culture war around translation and localization that I imagine we are going to see consumers declaring that they'd rather have janky translations that fit their own pre-determined ideas of what a text says rather than translations that have been written in a naturalistic style. In fact, we already are seeing that.

I suspect that would be a vocal minority. For example, Haruki Murakami translations are bestsellers because people want to read international literature written with evocative prose. I can't see that being replaced by AI anytime soon.

Unless I'm misunderstanding something, I think the purists who go for amateur translations probably aren't actually consumers of well-written literature anyway.


Total posts: 1,707
Top