Generative AI - Policy Discussion

Generative AI - Policy Discussion - Printable Version

+- Eagle Time (https://eagle-time.org)
+-- Forum: Administration (https://eagle-time.org/forumdisplay.php?fid=2)
+--- Forum: Announcements (https://eagle-time.org/forumdisplay.php?fid=29)
+--- Thread: Generative AI - Policy Discussion (/showthread.php?tid=7143)

Generative AI - Policy Discussion - btp - 11-11-2023

Hi all!

In the time since ET went down about a year ago, the accessibility of generative AI has become more prolific than ever.

As many members of our community spend countless hours and effort drawing, writing, and creating, we need to address the question of what place we want Generative AI to have in our community.

For clarity, Generative AI refers to any "prompt-driven" natural language model capable of generating or assembling a creative work including, but not limited to, art, stories and music.

Some examples of these Generative AI's include ChatGPT and LLama2 (text), DALLE and StableDiffusion(images), and Elevenlabs (audio).

Some points for discussion:

To what degree do we wish to restrict or allow the presence of works created using Generative AI on the forum?
Should all areas of the forum contain similar restrictions on Generative AI?
Do we, as a community, place a higher value on content created without the use of Generative AI?
What ways, if desired, should we support community members in their goals to create independently of Generative AI?
(on the opposite end) Are there areas where we wish to have more works of Generative AI on the forum?

Thank you to everyone who participates in this discussion.

-------

While we are figuring this policy out, I have added some bbcode to allow for voluntary flagging of AI Generated content.

You can now use the

Code:
[ai][/ai]

tag to wrap any content you have posted that was fully or in part generated by AI.

This will wrap your content in an "AIContent" tag (which is visible to the browser), and add a label beneath your content that says "made using generative AI" (which is visible on the post).

You can specify the message under your post by adding a modifier to the "ai" tag:

Code:
[ai=Image]Content[/ai]

Will display the message "Image made using generative AI"

For example:

Code:
[ai=Avatar][img]https://eagle-time.org/uploads/avatars/avatar_316.png[/img][/ai]

Looks like:

Avatar made using generative AI

As a courtesy to the members of our community, I'd request that anyone posting content made in part or in whole with Generative AI, use the ai tag to tag their content as such for the time being.

RE: Generative AI - Policy Discussion - Ixcaliber - 11-11-2023

I feel like there was a time where I was like 'oh yeah this is a fun tool to make weird little pictures' and that was like when dall-e mini was just becoming a thing and god was that last year? ANyway my horrible grasp of the passage of time aside. I feel like AI art is one of those things that can't really be consumed ethically any more. Sometimes it can still produce funny little images but I do get kind of uncomfortable about it nowadays.

RE: Generative AI - Policy Discussion - Schazer - 11-11-2023

Personally, I do not truck with the stuff and would be quite happy if we had a clearly-established culture, if not quite a rule across the forum, that the usage of AI art without further "human" manipulation be discouraged

Additionally, if there is any way to prevent our textual content being further scraped and dumped into AI projects without the express permission of the users, I would love to see that implemented

RE: Generative AI - Policy Discussion - btp - 11-11-2023

(11-11-2023, 08:14 PM)Schazer Wrote: »Personally, I do not truck with the stuff and would be quite happy if we had a clearly-established culture, if not quite a rule across the forum, that the usage of AI art without further "human" manipulation be discouraged

This is certainly a possibility. It would be a rule similar to our "no one under 18" where enforcement is primarily based on community request.

(11-11-2023, 08:14 PM)Schazer Wrote: »Additionally, if there is any way to prevent our textual content being further scraped and dumped into AI projects without the express permission of the users, I would love to see that implemented

The simplest way to do this would be to disable forum visibility to anyone who isn't logged in. This would prevent bots and webcrawlers from viewing data unless we expressly authorized a account, or if a bad actor wanted content bad enough to create an account for this purpose.

Unfortunately, with large language models connecting more and more frequently to search engines, normal webcrawlers can become active dataset builders, any content on the internet can be used to train language models.

RE: Generative AI - Policy Discussion - Qaxoasax - 11-11-2023

(11-11-2023, 08:14 PM)Schazer Wrote: »Personally, I do not truck with the stuff and would be quite happy if we had a clearly-established culture, if not quite a rule across the forum, that the usage of AI art without further "human" manipulation be discouraged

Robot Devil's advocate here. What would constitute "further human manipulation"?

You can spend hours figuring out how many times you have to seed "ultrahd 4k" into your image prompt before you get the desired result. I could sketch out a pose and then have ControlNet overlay a completely different style on top of it. Basing the use criteria on the amount of human effort it requires is an empty standard.

RE: Generative AI - Policy Discussion - Reecer6 - 11-11-2023

if it's for the purposes of shitposts and more or less laughing at the generative ml algo rather than with it, i have no problem with it. any usage of it that doesn't fundamentally rest on its provenance as generative ml feels wrong to me. i can't say it has no serious artistic purpose, but it really has to look and feel more like the stuff we were seeing a couple of years ago, right? something that makes it feel only possible by the ai and its strange and alien way of analyzing and creating images or text or whatnot (and thus we can say isn't just copying a few artists)

that's about as liberal as i will get in its favor though; if there were an outright ban on generative ml material, i wouldn't miss it at all.

RE: Generative AI - Policy Discussion - Cauchemar - 11-12-2023

I'm with Ixcaliber and others on the matter.

In the current state of things, machine art is a poison, it vampirizes human production AND uses this recycling to rob people of their job.

It is frustrating because like everyone, I see the endless possibilities, but it is currently not possible to use it ethically.

RE: Generative AI - Policy Discussion - CSJ - 11-12-2023

1. It's not 'AI', no matter how people claim it is and I will continue to mock people that act as if that label is in any way legit
2. Procedural generation is still capable of being used for positive ends, but is deeply tainted by the unlawful reliance among most such tools on data gathered without valid consent
3. The distinction between generated art and human art is currently still significant when you discount the utter behemoths of Big Data that are currently emerging, but is likely to decline over time
4. Because of the highly dubious but still untested legality of its datasets and the increasing pressure it is placing on the value of traditional art and art education, it is preferable to regulate its usage here until such time as the true scale of its impacts emerge
5. Sexually explicit AI art should be thoroughly yeeted as there is literally no way to validate if models capable of such art is not constructed without the use of child pornography, a risk further magnified by the current proliferation of CP content and tools producing CP content across the scene right now (civitai is going to be screwed the moment US legislators look at it)
6. The power dynamic of procedural tools is spit between leaked and otherwise 'free' models and those that are fundamentally gated behind processing and power requirements that limit it to being the domain of vulture capitalism which is fundamentally opposed to the values and livelihoods of eagle time's members

For the above reasons, I condone the following uses of procedural generation;
1. Shitposting - The alien and black box nature of precedural tools leads to some wacky results and an immense capacity for rapid production of low-quality memes that do not fundamentally cause harm
2. Intermediary tool for non-commercial purposes - Rather than being the raw output, the use of procedural generation to come up with ideas and inspiration can be useful as part of a creative process but not its sum total. An example of this would be generating an image of a bird on a surfboard, copying that image, then replicating it by hand in a way that is sufficiently transformative; perhaps your version of the image is an eagle with large chest muscles and a police uniform, an Officer Big Pecks if you will. Artists steal from each other all the time and I have no qualms about artists stealing from AI to the capacity their existing skills allow where this does not place the livelihoods of fellow artists at risk.

I would specifically suggest prohibiting the use of procedural generation for the following
1. Grand Battles that do not specifically permit it
2. Forum Games where procedural generation may provide a sufficiently unfair competitive advantage (eg; a drawing contest)
3. Forum Adventures where the generative content is the total or majority of its content AND it is not a parody work in which the use of procedural generation is central to its humour (see; Shitposting exemption above)

I am unsure about the use of procedural generation for
1. Tertiary content for Forum Adventures, tagged within the thread title (I would 100% use it for this if I could devise a means for doing so that is not disrespectful or demeaning towards my artist friends)
2. Threads in the Projects forum (I think this can be fine but should be a topic of further discussion as to what projects and works we are ok with being housed here that use procedural generation)

Quote:You can spend hours figuring out how many times you have to seed "ultrahd 4k" into your image prompt before you get the desired result

An artist spends an inordinate volume of time and resources to learn how to produce art which vastly dwarfs that required for procedural generation. As much fun as I have with generative tools, I would never, ever want to even suggest that what I do is comparable to someone that's had to sit in classrooms and listen to lectures about this shit. If I did not compile the dataset myself and personally craft all the assorted LORAs and other components required for non-shitty outputs, I have no right to consider comparing my labour to that of a digital (let alone traditional) artist.

I was an amateur musician growing up and seeing visual artists also struggle through the pressure of not feeling 'good enough' is common to that experience. A child interested in artistic expression now has to compete not only with the pressure of being 'not good enough' compared to other artists, but also 'not good enough' compared to an automated process.

RE: Generative AI - Policy Discussion - CSJ - 11-12-2023

Quote:non-commercial purposes

Just to highlight this bit, I would not want to see any commercial use of procedural generation (nor the use of commerical tools for procedural generation) on Eagle Time itself, ever. The moment monetization enters the equation, it is untenable and toxic to the people that made this site what it was (and is). I think it is reasonable for the community to develop guidelines and exceptions for certain non-commercial uses (and allow anyone still put off by it to filter it out of their forum experience) but this is a red line that everyone should be able to agree upon. I am happy to play around inside the bones and skeletons left behind by the san jose behemoth but it must not be fed.

RE: Generative AI - Policy Discussion - Kaynato - 11-13-2023

(11-12-2023, 06:28 AM)CSJ Wrote: »...

I'm very much with you on these points, but I want to make a distinction between general procedural generation (i.e. broader category containing things such as "made with handcrafted algorithms and not including external data") vs statistical data-driven models (neural networks and statistics models fed external data).

Using AI for shitposting though - I wonder if the time for even that is at an end. There was once a day where there was a fascination with the surreal quality of what GPT-2 would produce, but nowadays the context has changed too much.

EDIT: Also a tag for "hide from non-logged in users" would be very nice.

RE: Generative AI - Policy Discussion - btp - 11-13-2023

You must be registered to view this content.

RE: Generative AI - Policy Discussion - btp - 11-14-2023

Looking through the discussion, there seems to be some general consensus on things, and a few unresolved questions.

Consensus:

AI generated content should be restricted on the forum either outright or in a regulated fashion.
As a community, we strongly prefer content created without the use of Generative AI.

There seems to be a mixed opinion on what would constitute allowable use.

(BTW, I'm not including my own opinions (expressed as btp or my non-admin-alt-account) in this summary - more on that in a bit)

Some users are uncomfortable with Generative AI content in any form considering

the datasets used to create the most effective models are developed without the consent of the creators of that content.
allowing the use of Generative AI on the forum contributes to this ongoing problem
As such, its use inherently infringes on the rights of the original creators.

Others feel that, while the collection of these datasets are an ethical problem, the use of Generative AI tools are acceptable in circumstances where its use is clearly not infringing or harming human content creators such as

in circumstances where the final product is sufficiently transformative
where the work is presented for academic or parody based purposes
where such a work would not otherwise compete with human creators

Personally, I am trying to be very careful with what I say in this regard because it only recently came to my attention that my feelings on this are different from a lot of other folks on the forum. I am the person that's been the most open about using and posting works of generative AI, both here and on discord. I want to be really clear that my current role in getting the forum back up is to do what the community wants, and while most of the time what that entails is pretty intuitive, this has definitely been a blind spot for me.

Anyway, since I currently have access to the buttons that make this stuff go, it's important to me that I make a clear distinction between what I say as the button presser and what I say as just some random guy. With that in mind, unless there is significant objection, I'll probably continue to use Qaxoasax when I feel drawn to say something more confrontational in this context.

Full disclosure: I do pay regularly for access to OpenAI services like GPT and DALLE, and I use those services extensively in my development projects, including those here at Eagle-Time. Resources like GitHub Co-Pilot were created in ways very similar to those used to develop AI Image generators and some of the same ethical concerns may still hold.

Show Content