Not every piece of art you don't like was made by AI

[Umbran]

AI is becoming so integrated into technology that if you take an all-or-nothing stance, at some point it seems all digital art itself has to be called into question. Is there not legitimate use of AI tools by artists that is acceptable?

This is a place where we have to be careful. Current ethical issues are with "generative AI", which is something specific. There are other tools called "AI" that don't have the ethical issue.

 

[Ryujin]

This is a place where we have to be careful. Current ethical issues are with "generative AI", which is something specific. There are other tools called "AI" that don't have the ethical issue.

Indeed. In the editor I use there are such things as "AI Denoise" and "AI Resize." Do they use the work of others in order to perform the tasks? Highly unlikely. More likely that they latched onto a catchy expression for sales purposes. I'm not so convinced about MS Paint's "Remove Background" feature though. You have to know what a foreground is, before you can remove a background.

 

[Umbran]

Indeed. In the editor I use there are such things as "AI Denoise" and "AI Resize." Do they use the work of others in order to perform the tasks? Highly unlikely.

Sorry, but the things by that name I just looked up and read about are "trained on millions of images".

They might be a little different, ethically, because Denoising is about smoothing among existing pixels, and Resizing is about how to cleanly add and remove pixels as you change size. The AI has learned about how to preserve textures and shading during manipulation by training on other art, rather than learning how to create entire new works based on other art.

There is still a matter of taking art they didn't have rights to do do it, but at least they aren't in the space of using that to then replace the artists they cheated.

 

[Ryujin]

Sorry, but the things by that name I just looked up and read about are "trained on millions of images".

They might be a little different, ethically, because Denoising is about smoothing among existing pixels, and Resizing is about how to cleanly add and remove pixels as you change size. The AI has learned about how to preserve textures and shading during manipulation by training on other art, rather than learning how to create entire new works based on other art.

There is still a matter of taking art they didn't have rights to do do it, but at least they aren't in the space of using that to then replace the artists they cheated.

To me, ethically, it's not really different. I just looked up the latest versions of the software I use (I haven't upgraded in a while) and they're rather obfuscatory as to how it actually functions. Not something I like. In the version I use those tools are rather rudimentary and nothing that can't be obtained using conventional methods which is what I'll keep doing, going forwards. I'll also be careful about checking for the same sort of thing that Adobe slipped by.

For those interested, it's Corel Paint Shop Pro 2022. It appears that Corel Graphics Suite 2024 actually incorporates a plug-in called Vision FX, which sounds like full-blown generative AI. Might need to find a new app.

 

[Reynard]

Generative AI is getting more ethical by the day as company after company signs on to the platforms. Remember, kids, you don't have to give OpenAI permission, because you already gave DeviantArt permission and THEY are going to or have given OpenAI permission and - oops -- look at that, they just updated their terms of service.

This is not an intractable problem, nor are corporations like the New York Times taking ethical stands. This is about legal protections, not ethics, and it always has been.

 

[WayneLigon]

Kind of like how people scream 'Shopped!!' at every picture of a sunset, or call practical effects 'bad CGI'. (The CGI comments particularly get to me - you see CGI in virtually every TV show that has outside scenes in it, to remove extraneous or distracting visuals, or anything that might cost them money to show).

 

[briggart]

Sorry, but the things by that name I just looked up and read about are "trained on millions of images".

They might be a little different, ethically, because Denoising is about smoothing among existing pixels, and Resizing is about how to cleanly add and remove pixels as you change size. The AI has learned about how to preserve textures and shading during manipulation by training on other art, rather than learning how to create entire new works based on other art.

There is still a matter of taking art they didn't have rights to do do it, but at least they aren't in the space of using that to then replace the artists they cheated.

I’m not convinced they didn’t have rights to those images.

“Millions of images” is not really a lot in photography as compared with, say, painting. I can easily come back from a weekend trip with several hundred shots. Even with digital tools, I doubt an artist can produce a comparable number of images in a similar time frame.

I was listening to an interview to a photographer who collaborated to an AI training project, and mentioned that he provided them with a catalog of ~ half million images he had taken over 10-15 years. And that ‘s just a single case.

I think a lot of photographers/studios/press agency have a backlog of shots that didn’t make the cut, but could still be useful for training something like a denoising algorithm, and could be convinced to provide them for a reasonable fee. After all, unused shots are basically a loss, and the end product is not something that competes directly with their work, but actually allows them to get better results.

 

[Ruin Explorer]

I’m not convinced they didn’t have rights to those images.

“Millions of images” is not really a lot in photography as compared with, say, painting. I can easily come back from a weekend trip with several hundred shots. Even with digital tools, I doubt an artist can produce a comparable number of images in a similar time frame.

I was listening to an interview to a photographer who collaborated to an AI training project, and mentioned that he provided them with a catalog of ~ half million images he had taken over 10-15 years. And that ‘s just a single case.

I think a lot of photographers/studios/press agency have a backlog of shots that didn’t make the cut, but could still be useful for training something like a denoising algorithm, and could be convinced to provide them for a reasonable fee. After all, unused shots are basically a loss, and the end product is not something that competes directly with their work, but actually allows them to get better results.

Indeed - this specific case could be valid and one of the examples of AI being genuinely useful - but it would be great if companies could engage in a bit more clarity on this! I think the reason they don't is that they want to keep their options open for less ethical but potentially "killer" software apps. If you have "ethical" AI de-noise but some other AI feature you have isn't "ethical" in the same way, you merely call attention to that fact by pointing out that the former is.

 

[briggart]

Indeed - this specific case could be valid and one of the examples of AI being genuinely useful - but it would be great if companies could engage in a bit more clarity on this! I think the reason they don't is that they want to keep their options open for less ethical but potentially "killer" software apps. If you have "ethical" AI de-noise but some other AI feature you have isn't "ethical" in the same way, you merely call attention to that fact by pointing out that the former is.

Agreed. More in general, I would like more transparency from all sorts of companies, not just AI. I was merely pointing out that "training with million of images" is likely more commercially affordable in the photography space than in the arts space, so that alone should not raise an alarm bell.

 

[Umbran]

I’m not convinced they didn’t have rights to those images.

If they have rights to those images, these days they should provide the proof and use it as a selling point!