I think one of the ways to go about it could be to look for signs of something being an AI translation (assuming AI translations have similar hallmarks to regular AI texts) and just deciding based on that. If somebody uses AI just for help in translation, those hallmarks shouldn't really be there at all.

This doesn't really answer the question that I asked, which was how to write an enforceable rule that would meet your criteria. I can't think of a good wording and was hoping that you might have one in mind.

Regarding the good faith exception, I think I would just cut out the "The image is cited properly when used on the Wiki and appears to be compliant with both our license and the Image Use Policy."

I don't think we should grant an exception to the AI rules in cases where a user is breaking other rules at the same time.

In essence, I think our current methods are only really good at detecting a fully AI-generated text, though I could be wrong here, so they can still be used to determine if something is over the line or not.

Regarding the good faith exception, I think I would just cut out the "The image is cited properly when used on the Wiki and appears to be compliant with both our license and the Image Use Policy." That's already handled by the Image Use policy itself and the first point gets at the essence of it better.

Would changing "The image is not described as having been generated by AI, and the poster has no knowledge that it was AI-generated." to something like "The poster has no knowledge that it was AI-generated after having done a reasonable amount of diligence." address your concern here?

Yep, something like that sounds good to me.

Re. the good faith exception, it's not a punishment for misattribution. It's a limited exemption from punishment for breaking the AI rule. I'm open to suggested changes in wording if you have any.

That said, I don't know what percentage of people would actually go through the trouble of appealing, nor am I part of Disc/AHT, so I'm not totally sure if this is an actual problem.

It's a reasonable thing to consider, but based on the quantity of AI ban appeals we receive I wouldn't expect it to be a problem even if they increased by a substantial percentage.

I do think users should be expected to look for such disclaimers. I also wonder if such a misunderstanding would be something that could just be explained in a user's ban appeal.

It's something we'd consider as part of an appeal, yeah. I'll note that nobody has yet tried to make use of the good faith exception, nor have there been any cases where staff has found that it's applied to a given situation.

Would changing "The image is not described as having been generated by AI, and the poster has no knowledge that it was AI-generated." to something like "The poster has no knowledge that it was AI-generated after having done a reasonable amount of diligence." address your concern here?

I'm fully opposed to a full ban of tools like DeepL translator. I'm a bit biased here since I also sometimes use Google translate if I'm not pleased with a sentence in English, use translate to make it german (my mother language), change it a bit so I think it fits better and then translate it back to english.

I think valid usecases also come up regularly for translations. If I forget the good translation for a single word I can look it up in a dictionary. When that extends to more complex sentence structures or idioms, machine translators are often the better tool to help you. (And in the end it's up to the translator familiar enough with both languages if this is a good phrase to use or not).

I think usage of tools like DeepL translator in this or a similar manner has no effect on how creative of a medium translation (or regular writing, which, as I said, can also be affected here for people whose mother language isn't English) is and don't think of it as making it more mechanical than by using tools like a dictionary.

None of this applies for complete machine translations. Full AI translations are something that I don't think are a good thing to have on the wiki. I think the best place would be for the Translation Policy to be amended to disallow complete machine translations, but leave leeway for normal usage of allowed translator tools.

• Removal of Warning Revocation: This makes sense to me, given the revoke seems to be either ignored or the account abandoned
• Including the Sandbox: I've been slightly opposed to this in the past, because stuff like AI-generated ideas are allowed, so I think storing them on the sandbox. I don't think this is a thing that comes up in practice, so I think this is a good idea.
• deletions rule: support
• good faith exception: I'm unsure how Licensing currently handles attribution mistakes, but this reads to me that not fulfilling all three exceptions means you get a permaban. The second point for example should only be punished to the extent that normal misattribution would be. The third point makes sense. There's no excuse if you made an AI-generated image.
• support for the rest

If not Grammarly, then at least one spellchecker software should be added back on the allowed list, so it's clear that regular spellchecking is on the okay list (though I do think calling out the case of Grammarly specifically might be worthwhile, since we have strict line we draw there on what parts of Grammarly are allowed and what aren't, I believe)

In short:

• Banning AI Translations: Full Support
• Removing Initial Revoke: Almost Full Support; concerned about quantity of permaban appeals.
• Narrowing Good-Faith Exception: Almost Full Support; concerned about good-faith users forgetting to check AI disclaimers on image sites.

In long:
Banning AI Translations: Full support of banning machine-generated translations. It's an art form unto itself and machine-generated translations are disrespectful to the original author as they often do not properly capture their artistic intent.

Removing the Initial Revoke: Almost full support here. I feel like if someone's decided to use AI in the first place, they've either (1) not read the rules, indicating a lack of maturity, or (2) have a mindset surrounding creative work incompatible with that of the wiki which will likely lead to more clashes in the future. A revoke is just a bandaid to either of these problems because they go deeper than just a misunderstanding. That said, I do worry that people banned for AI will immediately appeal, creating work for staff in the long run as they have to process the flood of appeals, so it might be worth adding some sort of provision to filter out or delay or otherwise expedite them. That said, I don't know what percentage of people would actually go through the trouble of appealing, nor am I part of Disc/AHT, so I'm not totally sure if this is an actual problem.

Narrowing the Good-Faith Exception: This one I'm largely in support of, but I do feel like it's worth showing a bit more leniency here than the rules would imply as-written, specifically for cases where someone might've simply forgotten to explicitly check the page for "this was made by AI" disclaimers. I think that's a reasonable mistake to make, especially when the disclaimers on some of these websites might not be in an immediately obvious place, or when one might not have developed the "muscle memory" to consciously look for them while searching for images, because it's easy for one to tunnel-vision and forget these things while in the thick of sifting through tens or hundreds of images for the perfect one for an article. All that said, in the long run, I do think users should be expected to look for such disclaimers. I also wonder if such a misunderstanding would be something that could just be explained in a user's ban appeal. Overall, despite this concern, I'm in support of this change.

Rules text

Current rule

Proposed Rule

1. AI-generated content
   1. The use of generative⁴ machine learning⁵ models to generate, translate, or edit user facing content⁶ is banned.
      • Posting generative machine learning (AI-generated) content to the SCP Foundation Wiki, the associated forums, or any official Sandbox in violation of this rule will result in a permanent ban. This ban may be appealed through the standard disciplinary process.
      • Good faith exception for images: Users will not be banned for making use of AI-generated images from external sources provided all of the following conditions are met for a given image:
        • The poster has no knowledge that it was AI-generated after having done a reasonable amount of diligence.
        • The image is cited properly when used on the Wiki and appears to be compliant with both our license and the Image Use Policy.
        • The image was not created by, for, or at the behest of the user who posted it to the Wiki.
   2. All pages containing AI-generated text will be eligible for summary deletion under the standard summary deletion procedure.
      • All AI-generated images and other media will be removed by staff, irrespective of whether the good faith exception applies.
   3. Furthermore, the usage of tools whose primary use is generative machine learning is also banned for user facing content.
   4. Examples of banned content include but are not limited to
      1. Any ChatGPT-generated text or images, including modifications of existing text or images
      2. DeepL translator
      3. Midjourney/Stable diffusion
      4. AI voice generation
      5. Almost any system which generates text or media in response to a free-text prompt
   5. Examples of allowed content
      1. Fractal generation
      2. Traditional text to speech (if license compatible)
      3. AI-generated CSS code

Rationale

Removal of warning revocation

By this point, just about everyone with access to the Internet is aware that the use of generative AI is widely disallowed. Between that and the multiple mentions in our rules and Join the Site page, we make it clear that this applies to the Wiki as well. In addition, staff has noticed that almost nobody who has received a revocation for AI usage has gone on to produce a successful article legitimately. (They generally either use AI again and get banned or else abandon their accounts.)

In essence, the purpose of revoking on the first use of AI was to encourage users to instead contribute legitimately. This hasn't worked out in practice. Removing the warning revocation may help discourage AI use and will help reduce the likelihood of staff having to spend time dealing with the same user twice.

Including the Sandbox

All mainsite rules already apply to the Sandbox. This is technically redundant but serves as a potentially-useful reminder.

Deletions rule

This rule makes it explicit that AI-generated pages may be summarily deleted. This won't change the existing practice, but it'll help make it clearer to users and staff alike.

Modifications to good-faith exception

This exception is designed to be narrow. It was written to ensure that people who put in the work required to find and properly cite images compatible with our license and rules don't get banned for failing to realize that those images are AI-generated. The changes increase the clarity of the rule for staff and users.

Translations

Translation of fiction is an inherently creative endeavor. By allowing the use of AI for translation but not composition, we imply that translation is mechanical work on par with generating CSS to meet specifications. I don't believe this to be the case. Translations have their own awards and we've long recognized the work of translators by crediting them alongside authors in the attribution metadata. This puts translation in the same level as writing, as opposed to how we treat e.g. CSS.

It's also worth noting that anyone who wants to view an -EN page in another language, or another language branch's page in English, is free to run it through a translator for their personal use. As popular browsers such as Chrome, Safari, and Firefox now have this functionality built-in I don't believe that the additional effort required on the reader's part will be significant.

In addition, the subject of using AI for translation of portions of an article for in-universe purposes has been discussed. (E.g. an article on -EN, written primarily in English, with a section containing a poem written in French.) It's true that disallowing the use of AI for translations would make it more difficult for authors to include text in languages that they don't speak. However, this is already the case for other specialized domains of knowledge. We don't allow an author to make use of AI to transform dialogue into militarily-correct jargon or other vernacular, for example. Disallowing the use of AI for language translation would be consistent, and would also have the beneficial effect of encouraging cross-branch collaboration when an author's vision for a work required it.

If adopted, this change would not be retroactive. Existing AI-translated works would remain on the site.

Removal of allowed list

The list was empty. If there's ever an exception that we allow, it'll be restored.

Removal of "stories about AIs" rule

This was redundant and only produced confusion. It seems better to only mention generation of content, not subject matter.

Discussion and Voting Procedure

There are two questions here:

1. Should the overall rule change above be implemented?
2. Should the use of AI for translations be disallowed? (I.e. should the underlined portions be included in the rule change?)

I anticipate that the second question will be more controversial than the first. It seems reasonably likely that the first question may have consensus in the discussion stage without requiring a vote; on the other hand, I don't expect that to be the case for the second.

If we need to vote on both questions, then the issue of whether AI translation will be allowed will be voted on as a separate question in the same vote. This will affect the underlined portions only. (If the overall modification passes but the vote results in AI translation being allowed, the underlined sections will be removed and 'DeepL translator' will return to the list of examples of allowed content.)

If the overall modification passes without need for a vote but a vote is required for the translation question, then the new rule will go into effect with AI translations permitted in the interim. Should the voting results lead to translations being disallowed, that modification will go into effect following the conclusion of the vote.

Edit log: incorporate suggestion based on radian628's input.