I would also like to mention the community & sociological impact of AIGAPs. While this does have the benefit of providing a new and useful utility to authors who need very specific images, it also introduces two new issues into the wider Wiki that I believe cannot be reconciled at this time:

1. There is significant disapproval and backlash to the use of AIGAPs in wider spheres that the SCP Wiki intersects, including the art community which we have spent significant amounts of time trying to support and cultivate on-site.
2. The introduction of additional, instant, limitless tools for generating high-quality(?) images locked behind a $30 per month paywall introduces a barrier between a "causal" or "new" SCP author and someone who is willing to spend money on the service for articles.

The first one is simpler to explain. Artists in general have shown distaste for AIGAPs because of both the training data stealing and that it removes a source of livelihood from artists whose clients may now pay money to a machine to do a less decent but faster job instead. While the first is more up for debate in regards to Midjourney, the second is not so; I have spoken to a few active, on-site artists who are incredibly disheartened that the use of AI-generated images is even up for debate on the Wiki in the first place. The mainsite mirror shows similarly that most regular SCP users are not particularly settled on the topic yet, though that may change with time. As such, I recommend from a PR perspective that AIGAPs not be allowed on the Wiki for the foreseeable future.

Regardless, I have an additional hesitation which I outlined above. I believe the introduction of AIGAPs that are only obtainable and usable behind a paywall will create a minor, if notable, schism in the authorship and may discourage new authors. It should not be a surprise to anyone here that an article's image can significantly affect its rating, viewership, and general perception; both commissions of artists and AI-generated images can greatly improve a piece's attraction / atmosphere. However, this also means that anyone who is not willing to pay money to an image-generating service or artist would be put at a disadvantage, needing to rely on websites like Wikimedia Commons (as many authors before have). This may very likely introduce the perception to new Wiki authors that AIGAPs or commissions are necessary or greatly encouraged to publish articles, which may reduce new authorship and mildly tarnish the perception of the on-site / staff further than it already has been.¹

For these reasons, I am against the use of AIGAPs on the Wiki for the foreseeable future. I believe we may likely revisit this at a time where the thread's above concerns and my own two concerns are less prominent, but until then, I think it is better for the Wiki as a whole if we shelf the discussion of including AI art, and default to the position of no.

Whether AI images should be allowed as art pages is a separate issue for a separate thread.

From an ethical standpoint, I find AI generated art to be extremely derivative. This isn't something that's trying to create something like peoples' faces, its scraping art - the long and hard work of many people - and making its own without giving any credit to those artists. Even if they did credit those artists, would most of those artists be happy with that? After reading about how the AI works (Image Abstraction vs. Image Variance), I think the concern of scraping people's art has somewhat been alleviated. I still have concerns though, for example can we really verify that the image sources are all CC BY-SA? What happens if the AI creates something significantly similar to an existing artwork? I'm still not convinced that the benefits of AI generated art outweigh the cons.

I'll put aside all legal arguments, because I trust CityToast and Licensing's judgement more than my own. However, even if its all legally sound, I think this is a case of a thing that can be implemented but shouldn't.

Firstly I want to say I'm very glad that this thread was created, asking the community the question it is, that's very important.

But I hope it goes further.

While the content that we are legally able to put on the site should be determined by Licensing, as that is in part their function and I believe they are the most capable among us, the legality of content does not necessarily mean the community wants it on site.

The community should decide, and I mean actually decide. Not just suggest how staff should proceed, and hope that we listen, which is effectively the case with staff votes, and discussions such as these.

To very briefly touch on the decisions that brought us here today, and resulting community sentiment:

People don't feel they were listened to, people don't feel they were informed. I know that this is true for both members of staff and members of non-staff. (Not everyone, obviously, but I've seen sentiment from people in

This is a case where, to my knowledge, there is no information that is sensitive. That is to say, I, for instance would make no more of an informed decision than any member of the community.

So why not do a community vote?

People would certainly feel more listened to than if the result of this discussion is a staff vote, or no vote at all.

Myself, I trust the Licensing Team made an informed decision on this. Every single time I've seen a reasoning from Licensing that concerns AI art, or an explanation of their process, it's obvious that a lot of time and effort has gone into their work. As such I fully believe them when they say we're alright to allow Midjourney, and am happy to have it as an option, and a possibility of whitelisting more

As an individual,as a member of this community, I would vote for us to allow AI images on the site, as we've been doing, and to stay the course as Licensing intended.

However, in this instance, my words and my vote should carry no more weight than your average John Wikidot member of the site.

TLDR: Good question, let's have a community vote to decide.

These works are not credited to the original artists, nor are most of them derived from CC sources.

The database which was the initial training dataset for Midjourney, and many current AI systems, is a database that is, in fact, CC compliant.

In general, I'm seconding psychicprogrammer. The fact that this is more restrictive than what even Creative Commons recommends and that we can easily remove all AI art if it becomes problematic (this was mentioned before, but an _ai tag would be a good idea for this) makes me think we're safe enough on the legal side of things as well. Although I'd also be fine to waiting until after this general issue went through courts

I am curious about section 4 about requiring a filter for illegal content. In that this will stop things like the local version of stable diffusion or any other FOSS AI artwork generators run on the users machine.

Basically I am not sure why we don't allow stuff that could generate illegal content as long as it does not generate any for the wiki.

It is inherently exploitative. It involves an algorithm parsing countless existing works in an attempt to extract a semblance of originality or "humanity" from a piece of work. It is a process of "cultural analytics", that quite often cannot comprehend what it is looking at and simply groups it based upon visual similarities. These works are not credited to the original artists, nor are most of them derived from CC sources. Stating that something is "reasonably abstracted" is impossible if the source cannot be defined, nor has a definition been supplied by a court of law.

It pushes existing artists out from spaces and niches they'd previously occupied, which is especially concerning for entry-level artists who may be overlooked by companies that view AI-generated content as "sufficient" for their needs. Why pay someone to create renders or concept art when you can have a machine spit them out for you?

It worryingly opens the floodgates to hosting text-based AI-generated content on the site, which links to debates surrouding whether AI-generated content should be considered augmentation, or replacement.

I would recommend that anybody invested in this topic does a cursory search of mainstream media outlets and social media to see just how concerned many artists are about AI-generated content.

In allowing AI-generated content on the site, we would be enabling the exploitation of artists, who are already undervalued and underappreciated by the wider world, and it therefore has no place being on the site.

I've clearly written a lot about the ethics and morals of hosting exploitative content on a creative writing platform, but I'm also concerned about utilising our community as a testbed for the hosting and curation of such content. It simply flies in the face of our site's entire atmosphere.

On a related note, does the above criteria also apply to AI-generated text providers? If so and if we are similarly uncertain as to whether those are on stable legal footing, then those should be disallowed as well.

After witnessing the debacle and controversy that AI art has caused, along with gaining more experience in the past few months with respect to SCP writing, my opinion with respect to this has significantly changed. While I do fully understand the perspective of the art enhancing the article, there are plenty of other alternatives that don't stray into the gray area of CC. I get how painstaking it is to comb through image after image, trying to find the perfect picture to make one's SCP appear all that much better. Or trying to learn photo editing to make the piece seem semi-realistic. AI solves all those problems, but implementing it causes a lot more. We shouldn't need to be consulting lawyers for something this avoidable.

In other words, the risk at this point is not worth the payoff.

Also something to keep in mind is that this is an evolving technology. I'm sure within the next few years, we'll have some sort of CC compliant AGAP that's more widespread and already been "tested". Once that point comes, I'm sure this issue will be revisited. But for now, it isn't something we should participate in.

Supporting the exclusion of AI-based art from the wiki.

Why expose ourselves for almost no payoff? Especially when even the hypothetical payoff — "people can generate art for something there aren't good CC images of" — already comes with its own huge ethical caveat: these AGAPs scrape huge quantities of other peoples' images and do not credit them even if they are CC, which most of them aren't. Even assuming for a moment that AGAPs can be said to 'create art', they do so in an exceptional way unlike anything else in the world, and so I feel comfortable treating them exceptionally: excluding them from use on the site.

EDIT: I should note for posterity that the Wanderer's Library has, due to a combination of the licensing ambiguity and the massive ethical concerns, instituted a blanket ban on AI-generated images. This is the policy I would recommend the Wiki follow.

Multiple discussion threads and informative actions should have been taken as part of this effort, and we have much to answer for both internally and externally on this front. However, the purpose of this thread is not a post-mortem for the above change.

For the last several months the Licensing Team has been evaluating the legal questions surrounding the use of AI generated art. This has included a thorough review of our own license, thorough reviews of licenses from AI providers on a per-case-basis, discussion with the Creative Commons community, and multiple stages of evaluation with legal counsel.

After thorough consideration, the Licensing Team has developed a list of requirements which it believes are necessary for an AI-Generated Art Provider (AGAP) to be compatible with the Site License (CC-BY-SA 3.0):

• (1) The ability to attribute a particular image to a particular AI generation event.
• (2) The ability to attribute it to an individual user (to avoid disputes of ownership).
• (3) Explicit assignment of ownership of a generated image to the person who generated the image.
• (4) A built-in filter or moderator team to prevent the engine from generating illegal imagery such as child pornography.
• (5) A model using an image abstraction protocol, rather than image difference or image variance.
• (6) Even IF it meets all the above, the training data needed to be compliant, too.

To Expand on (5), the difference is essentially that the AI cannot just combine images; it has to have a model that breaks the image data down into something else, and then uses abstract relationships to fully procedurally generate a brand new image.

To expand on (6), Creative Commons suggested that this wasn't necessary, but the License Team insists on this requirement due to the legally untested nature of AGAPs. Many AGAPs do not release what models or training data they use. Because of this, it becomes difficult if not impossible to prove that the training data itself is compliant with our license.

At present, Licensing has identified one AGAP which meets the above criteria and is therefore likely to be compatible with our site license, and that is the version of MidJourney available via a paid subscription. The Licensing Team makes no formal endorsement for MidJourney as a service and plainly states we have no relationship whatsoever with either MidJourney or its related business, we reiterate we have simply identified its likely compliance per the above.

Having said that, the decision the Licensing Team was empowered to make was whether or not there are AGAPs which are likely to be compatible with both our Image Use Policy and the Site License. It is not the Licensing Team's sole discretion as to whether or not this type of content should be allowed - that is to say "an content created by an AGAP" and not just via MidJourney.

The purpose of this thread is to collect staff and community thoughts about whether to allow AGAPs for use on site, and if so, how to ensure such a matter is implemented both ethically and with support and consideration for members of the art community.

I created this discussion with the assistance and input of CityToast. I am one of two current Admin Contacts for the Licensing Team, and CityToast is the Vice-Captain (and currently acting Captain). Me and my previous vacation plans appreciate everyone's patience in allowing me to delay this discussion until today. As this topic is quite large, requires community input, and does not propose a specific policy for immediate consideration, do not feel constrained by the typical "7 day" limit.

Link to main site mirror

This is open to all staff.