The Society for Artistic Integrity states that training methods for generative AI systems – that scrape the internet, require compliance in Terms of Service without opt-out or compensation, or otherwise steal from or manipulate a creative human author/owner – are unethical. While the intended outcome of the AI technology may claim to enhance society, the methods used to train the datasets are a violation of collective trust and are harmful to the creator/author/owner of the original works.
Recognizing that the lack of consent is theft, we established SAI to allow the creator utilizing the benefits of this new generative AI technology to disclose their use or non use by designating their work with personalized SAI Seals. This transparency builds trust with their customer/audience and lets the consumer choose whether or not to engage with the work (regardless of whether that decision is based on unethical system training or other reasons).
While regulation on generative AI system training may be forthcoming, our view is that through the ethical use and disclosure, society has the ability right now to shape what is an acceptable use of AI through consumer attention and spending.
Example Impact of Generative AI on a Painter:
What was Generative AI trained on?
Content from creators was used to train generative AI through the collection of data into large structured datasets. The data was acquired by ‘scraping’ the internet taking text, numbers, images, and coding with it. Artists’ copyrighted works, website content, articles, books, songs, audio, and more were swept up into the training. Permission was not granted, agreements were not made, no royalties were paid… no licensing, no credit, no asking, no ethical considerations – creators’ work was stolen.
What about copyright?
- Even if a work has not been officially published or registered, a copyright still exists the moment a work is created.
- A registered copyright establishes a public record and provides additional legal protections to the creator of the work.
- What can be copyrighted: novels, poems, articles, blogs, songs, paintings, illustrations, photographs, illustrations, marketing materials, movies, plays, computer software, and much more.
- A “style” cannot be copyrighted. Nor can ideas, facts, methods or concepts.
What’s the problem with Generative AI as it relates to creativity?
- The tech was trained by unauthorized scraping of content / not getting permission
- AI software charges a fee for use, but uses stolen content to generate
- Direct, targeted use of style could harm original creator (confusion, loss of income)
- Output lacks the soulfulness of human creation
- It could eliminate creative jobs
- Lacks accuracy in images and facts / "hallucinates"
- The over reliance on AI could oversaturate content with repetitive imagery and “AI look”
- Overall devaluation of art
- Additional energy needed for computing power
But Generative AI has its benefits for creatives too, right?
- Brainstorming, forming ideas, helping with workflows, inspiration
- Time efficiency, productivity tools, automation
- Accessibility and democratizing creativity
- Analyze large datasets to identify trends and patterns
- Recreate and reimagine historical art styles not readily available
- Enhance visual effects / animation that would normal be time consuming
What’s next?
There is no going backwards on generative AI system training as we can’t undo the scraping, but future regulations may establish more ethical guidelines. With continued education about AI and disclosure of its use, the creator and the consumer can make their own choice about whether to engage with AI.
Moving forward, creators can choose to participate in the use of generative AI with the knowledge about how it was trained. SAI certification lifts the veil on the methods of artwork creation and the transparency established with the SAI Seals allows the consumer to also make a choice.
