Is the Second Life Controversy Over Gen AI Mainly About SL Images That No Longer Look Like Second Life -- Not Over AI Per Se? (Comments of the Week)

Since writing about Primfeed's survey on whether AI-enhanced images should be allowed on the social media site, the vote continues to lean heavily on No. 

One challenge, however, is that "AI" is still broadly or loosely defined. As SL creator  Jennifer Fluffington comments:

We need to define more what is considered 'AI' to be allowed in the public forum. I.E., I use Photoshop Generative AI to make minor repairs that would take me 5-6 steps and layers to do in one click.

Am I lazy? I don't think so, I being smart with my time spent on a project that I am making pennies on the hour for.

Yes I enjoy my craft, but let's face it, we all have limited time in a day and many things to do, so if I can shave 15-20 mins off 3 different defects on a project that almost an hour saved in my day.

So, because of those repairs that don't alter the overall aesthetic of the photos, are my photos going to be rejected from the public forum? This is a discussion we need.

Fair question! One possible answer is implied in a comment from Kate Nova -- when an avatar image no longer looks like the avatar themselves does in Second Life:

I agree that allowing AI images and tagging them as such is fine. I don't need to have AI images banned, even though I personally dislike them.

What AI images mean in First Life is its own question. For me, in Second Life, they take me out of the experience. I guess AI users feel they look "better" (mutant fingers and other small details notwithstanding) but to me they look "Not SL".

Certainly lots of SL-photographers including me buy newer and better skins and bodies/heads over the years. We do a little or a lot in Photoshop. Yes, slippery slopes everywhere. Still, an SL image, even touched up and Photoshop uber-saturated still looks like SL avatars. My photo still looks like me.

What I don't like about the "pretty" AI-SL pix is that my image doesn't look like me. The AI-art-bots aren't trained on SL avatar images, they're trained on First Life fleshvatars. When you AI process an SL image it becomes more FL-looking than SL-looking.

This isn't "evil" or "wrong". But for me, it's just not SL.

To Kate's point, years before Gen AI platforms existed, I've seen many SL images that have been Photoshopped so aggressively, I wondered if they should be categorized as "Second Life" at all. (Cajsa and I sometimes get into that debate over various images she features in her highly excellent column.) It's also become a pretty standard part of the Second Life Flickr community culture to tag an image as being heavily post-processed if the alteration goes beyond surface re-touching/cleaning.

Maybe the solution is to put Gen AI-based alteration and post-processing alteration on the same spectrum, and allow Primfeed users (in this case) to vote, "Is this image too altered to be considered Second Life-based?" And if a super-majority votes Yes, flag the image as Heavily Altered. Framing the question that way gets us away from the contentious holy war over AI, and puts the focus on what actually matters.

Comments

[psychee]

Simpler, and here speaks a professional illustrator: ban the use of AI. Period. Because everyone abuses it totally unreasonably, without shame or responsibility, so the diplomatic way doesn't work.

[Dolly]

Hi, the controversy over GenAI isn't just about images that no longer look like SL. That is a factor, yes, but not the main one at all.

First and foremost is the fact that the vast majority of GenAI models were trained using data and images stolen from artists and creatives for the express purpose of imitating their styles, to take money out of their pockets and put it back in the hands of corporations. Any time you generate an image using AI where you would have paid an artist if AI weren't available, you take food out of that artist's mouth. It is predatory software made by corporations off of the backs of starving artists around the world.

On SL, GenAI assets also compete directly with the hard work and efforts of real creatives and artists for residents' hard-earned lindens, while offering worse products to those who don't know better. AI-generated models generally offer only one face for texturing, high-polygon counts that generate more lag, and because they are often generated by people with zero modeling knowledge, they generally cannot be fixed and updated if a customer finds problems with them.

SL was built from the ground up by creative humans and artists into what it now is. Many of us worry when our hard work will be the next to be harvested and fed to an AI model without our consent or knowledge, to directly compete with us. Our chats already have been, so that Linden Lab could partner with a 3rd party company to provide an AI chat-bot for customer service. For some, that may not be a problem yet, but for others, we are left wondering if that's truly the end of it.

Those choosing not to engage with it, who want it out of their SL-adjacent spaces acknowledge these SL and real-world issues, are often doing so on principle. If the creation of a GenAI model was only possible by way of theft, then ANY use of that GenAI model is unethical, and encourages the theft necessary for its creation.

MANY people, both on and off of SL, have been speaking quite loudly about the theft issue with GenAI for a long time on a variety of platforms, including a Second Life feature request that generated hundreds of comments worth of discourse on the topic. It's disappointing to see a platform this large, a trusted source of information about the virtual world, minimize the issue as merely one of vanity and nostalgia for the aesthetics of Second Life, and continue to push pro-AI narratives by only platforming opinions the author agrees with.

[pro-facts]

Dolly wrote

First and foremost is the fact that the vast majority of GenAI models were trained using data and images stolen from artists and creatives for the express purpose of imitating their styles, to take money out of their pockets and put it back in the hands of corporations. Any time you generate an image using AI where you would have paid an artist if AI weren't available, you take food out of that artist's mouth. It is predatory software made by corporations off of the backs of starving artists around the world.

Sorry, but this is a fallacious argument on so many levels. And it's parroted over and over on social media. In reality, Any time you generate an image, you probably wouldn't have hired an artist anyway and more likely you would have downloaded an image from the Net. Or made something disappointing.
These are the only options most people have, even if they have creative ideas: most people aren't skilled with professional tools nor have the money to hire a professional artist for their little things. If you are lucky, you have a skilled friend that has some time to spend for you.

I've seen places in SL where the paintings were AI-generated and fitted well, customized for that specific virtual scenario, which is more original and so much better than the usual generic stuff SLers usually just download from the Net. However, you don't have full control on what's generated, it's improving, but there are still many discarded generations between the images you eventually use.
If you are an artist, you will make something better... by taking a longer time. Also some artists now use AI-gen, either as part of their work with Photoshop or as a base or to brainstorm new ideas.

If you have the time and energy to become an actual artist, good for you. And there is nothing wrong in learning from other artist's styles: it's what artists do since art exists. Images aren't stolen: stolen is when you don't have them anymore. By the way, that same 'stolen' and 'theft' argument is frequently used by the film and music corporations. And here it is even less valid, because the results aren't copies but derivatives and styles are often mixed and blend together for a more unique result.

At most, a valid argument may be a license, to get paid for training. However, if you publish your images and other content on the Net, you should read the TOS of the platform you are using, Instagram, Facebook... and also Second Life, because usually you grant them a license to use your content, display, distribute, and even modify it.

The 'MANY people' crusading against AI-gen sounds like the typical falling tree that makes more noise than a growing forest. But I'm sure many jump onto the hate train, seeing that "everyone" hates AI. In reality, many more SL profiles than them are using AI-gen. ChatGPT is on the top ten visited websites now and AI apps are popular.

The fun thing is, the ones really at risk are the scripters (and to a lesser extent, the actual coders). Do you see them protesting so much?