杏吧原创

Concerns over copyright and artists鈥� livelihoods have caused photography agencies to remove images created by artificial intelligence models from their databases.

The worries come from the fact that the AIs hoover up vast amounts of human-generated art to train themselves and use this database of knowledge to generate photorealistic images related to almost any text prompt. This has led to ethical debate and highlighted the legislative grey area about who owns the copyright of an image generated by an AI that bases its output on prior work.

Large stock photography agencies including and Shutterstock have announced in recent weeks that they will remove images created by AI models like DALL-E, Stable Diffusion and Midjourney because the legal situation on copyright is unclear.

A Getty spokesperson told New 杏吧原创 that 鈥渢here are open questions and real concerns with respect to the copyright of outputs from these models, as well as unaddressed rights issues with respect to the imagery, the image metadata and those individuals contained within the imagery鈥�.

Although the creators of some AI models, , have said that copyright of output will belong to the user who enters the prompt into the AI, smaller photo agencies are feeling the pressure to follow the large companies in banning the content. Dittmar Frohmann, the owner of stock agency , says his company is deliberating whether to follow Getty and Shutterstock, but is likely to take the same approach.聽

鈥淚鈥檇 rather follow them because we鈥檙e a tiny company,鈥� says Frohmann. 鈥淚鈥檇 rather believe their legal departments than the guy that I have to pay and who looks into it for like an hour.鈥�

at the University of Surrey, UK,聽 says that while in theory you can legally copyright images generated by an AI in the UK, and the same should be possible in the US, it hasn鈥檛 actually been tested in the UK and the US Copyright Office has 鈥渁 longstanding policy of not allowing copyright on AI-generated art鈥�.

Abbott has , with no success, and claims that the US government鈥檚 stance stifles innovation. 鈥淎I is resulting in the generation and dissemination of countless works with social and market value, and this development should be encouraged,鈥� he says. 鈥淪imply removing AI-generated images entirely from a platform is not a real solution to any of the challenges associated with this phenomenon.鈥�

at law firm Decoded.Legal says the UK鈥檚 Copyright, Designs and Patents Act 1988 in the country, which would probably cover AI-generated work, but there are complications. Not only could the output image be covered by copyright, but the input prompt also could, if sufficiently detailed. There are already that allow people to buy and sell detailed text prompts, and anyone able to prove that their text prompt is being used by someone else could have a copyright claim. Not only this, says Brown, but the creators of the AI systems themselves may have their own terms and conditions that could 鈥渇urther muddy the waters鈥�.

鈥淭he work that you create, in my view could be protected by copyright,鈥� he says. 鈥淏ut that isn鈥檛 settled. So you could be in for some expensive litigation.鈥�

Some academics argue that nobody should own the copyright on AI-generated images. at King鈥檚 College London says that society currently seems to be valuing the ability of AI to create images above the creativity of human artists whose work was 鈥渟tolen鈥� to train it.

鈥淩ight now, the public are generally on the side of image-generating AI,鈥� he says. 鈥淭hey see it as exciting and novel. I suspect that won鈥檛 last forever, and eventually we鈥檒l see people shift their view towards it being derivative. Stable Diffusion isn鈥檛 defining its own styles, trends, methods or anything like that. We are marvelling at a system that can replicate human artists, but we鈥檙e yet to see an AI that can contribute something back to human culture.鈥�

Frohmann says Photocase is likely to seek explicit permission from the people submitting images to their agency before allowing AI companies to use their archive for training their AIs, because these artists may well object to the idea of their work being used to develop an AI that could one day replace them. References to AI training is already being for larger companies like Getty.

Despite his stance, Frohmann says that ideologically he sees AI as a tool akin to editing software or to a physical camera that is being used by someone to create work, and that copyright should belong to the person prompting the AI. He also currently sees no threat to the livelihood of artists and photographers because of the quality of the output.聽

鈥淚f you look at the sum of it, it鈥檚 a whole lot of crap,鈥� he says. 鈥淚t鈥檚 the median of all the tastes. The images, if you look at them, they look good 鈥� but totally boring and predictable, and soulless.鈥�