04 mirror: https://scp-wiki.wikidot.com/forum/t-16479190/discussion-broad-based-banning-of-ai-generated-content
Previous discussion: https://05command.wikidot.com/forum/t-16451655/discussion-ai-but-again
Introduction
As per the previous discussion a move to ban a wider range of AI generated content has widespread support from both the staff and community. This then leaves us with the very thorny question of defining what exactly that means. We want to ban things like Midjourny and GPT generated text but do not want to ban DeepL or Grammerly.
The boundary between Photoshop’s de-noising/image enhancing tools and latent diffusion models is very very thin when it comes to how it works.
As such I am proposing that we add the following text to the site rules
The use of generative1 machine learning2 models to generate user facing content3 is banned.
Furthermore, the usage of tools whose primary use is generative machine learning is also banned for user facing content.
All exceptions to this policy are on the following white list:
- This list is currently empty
What this means
This would ban the following things:
GPT generated text
Midjourny
Stable diffusion inpainting (not generative but )
Auto captioning (we would need a policy for an exception here)
AI based auto summarization/excerpt extraction
AI voice generation
This policy would allow for the following things:
Grammery, DeepL: Machine learning, not generative
The 001 image generation software: Generative, not machine learning
Algorithm curated recommendations: The generative part is not ML, the ML part is not generative
Smart layer blending/Smart infill/ AI image upscaling: Again not banned as not generative, despite being machine learning based
AI generated CSS or Java script: not user facing content
The music content that was discussed on the 04 mirror
Traditional text to speech
Automatic transcribing of text
Idea generation
Artificial general intelligence (though the age limit would still apply here)
Writing about AIs
As a note there are certain types of AI image generation that would pass this test, however any method of image generation that requires the construction and training of a custom neural network requiring a master's degree in machine learning is A: Not going to happen in any case and B: Probably involves enough human creativity to count as art anyway.
Nuances and expansion of purview
Given the complexity of this and the very fast movement of the field, I would also like to propose that the tech team can ban any machine learning tool that they deem to be sufficiently like a machine learning tool. The tech team can also approve a banned tool under this policy for a specific use (like automatic alt text generation for images). The licencing team can overide this aproval on the grounds of licensing reasons.
This discussion will go on for one week