Generative AI is a Disaster in the Making

[imagination7413]

3 hours ago, estebanep said:

I think they should start being a little more flexible with the new creators and vendors of AI-based services.

... I think that the new creators ought to read the rules first. The "I agree to these terms" box on signup is a business contract, not a suggestion. 

Fiverr already is flexible, especially since they embraced the advent of AI, even while litigation was in turmoil. They could just as easily said 'no' and completely terminated any and all AI services on the platform.

[meldawn9]

On 2/10/2023 at 5:43 PM, maegan_bertrand said:

 I was shocked to learn this past month that FIverr is going to allow generative AI gigs to be sold on their platform. Currently, there are many legal and ethical complications surrounding this technology.

 For months, artists around the world have spoken in opposition to AI generated content, due to the fact that the datasets used to train AI consist of millions of copyright protected photos and pieces of art. These images were used without license, consent, compensation or credit given to the artists. This issue is perfectly illustrated by someone being able to use a specific artist's or photographer's name in the prompt text in order to generate imitations of that person's work and style.   

 Currently, the US does not give AI generated work copyright protection, and in the UK, the government has rolled back a policy that would have given AI developers near unparalleled exemptions for datamining after the policy was met with significant backlash by both the music industry and visual artists.

Linked below is an article on the rollback of the UK datamining exemption:

https://www.musicbusinessworldwide.com/british-government-scraps-plan-for-ai-text-and-data-mining-copyright-exception/

 

 Fiverr's decision looks even more baffling to me in light of there now being two lawsuits against AI companies, one of them by Getty Images, and the other filed on behalf of visual artists.

Artist Lawsuit Against AI:

https://www.cbsnews.com/news/ai-stable-diffusion-stability-ai-lawsuit-artists-sue-image-generators/

You can view the PDF of the lawsuit here:

https://fingfx.thomsonreuters.com/gfx/legaldocs/myvmogjdxvr/IP%20AI%20COPYRIGHT%20complaint.pdf

Getty Images Lawsuit against Stability AI:

https://www.artnews.com/art-news/news/getty-images-lawsuit-stability-ai-12-million-photos-copied-stabile-diffusion-1234656475/

You can view the PDF of the Getty Images Lawsuit here:

https://cdn.vox-cdn.com/uploads/chorus_asset/file/24412807/getty_images_vs_stability_AI_delaware.pdf

 

 In recent weeks, there has also been growing concern around AI's ability to generate deepfake visual and audio of real people, including celebrities and streamers. These deepfakes have been used to spread misinformation, to harass, and to create non-consensual p*o*r*nography. While deepfake technology has existed for a few years, what makes this new iteration potentially more dangerous and terrifying is the speed at which deepfake material can now be generated and spread, and how it will progressively become harder to distinguish from real photos or video/audio as the available models improve.

Below is a link to a news story detailing the Atrioc deepfake porn scandal:

https://www.insider.com/atrioc-caught-qtcinderella-ai-picture-twitch-deepfake-controversy-streamer-trauma-2023-2

 

Now we know why Fiverr is recommending that we buy business insurance. Just imagine you deliver AI-generated art or text to your client. the client loves it and uses it. A few months later, they are sued by Disney/Mattel/Marvel/New Yorker because images/text from their company are being used by your client. Your client sues you. Now you can see how you need this insurance. 

[uk1000]

13 hours ago, meldawn9 said:

Just imagine you deliver AI-generated art or text to your client. the client loves it and uses it. A few months later, they are sued by Disney/Mattel/Marvel/New Yorker because images/text from their company are being used by your client.

Maybe for the AI image generators they could be trained a lot on content the sellers manually create and maybe royalty free images (eg. from pixabay maybe - if that's allowed) so much that the software would be unlikely to output something based on something owned by someone who might sue. Maybe creating natural looking landscapes (images and ai videos) where they are less likely to create something that someone could so over.