AI is stealing writers’ words and jobs…

[Bohandas]

Training an AI on an image is not rebroadcasying the image or reprinting the image in any meaningful sense. A given training image has about the same effect on the AI as it would on a human artist who happened to see it once.

Not that any of you are going to care, as most of these artists and those of you who support them have made it very clear that the only thing you rrally care about is money, NOT artistic integrity. It's even right there in the title of this very thread. Artistic intrgrity is nothing but an excuse to protect your profits

 

[Bohandas]

Training an AI on an image is not rebroadcasying the image or reprinting the image in any meaningful sense. A given training image has about the same effect on the AI as it would on a human artist who happened to see it once.

Or for that matter, considerably less effect than ut woukd if the human artist in question was Andy Warhol

Why is it ok for Andy Warhol to make a name - and money - for himself on outright trademark infringement but if a computer program borrows a tenth of a pixel suddenly it's piracy?

EDIT:
Another thing that infringes significantly less than AI training is fanfiction.

 

[Sacrosanct]

Or for that matter, considerably less effect than ut woukd if the human artist in question was Andy Warhol

Why is it ok for Andy Warhol to make a name - and money - for himself on outright trademark infringement but if a computer program borrows a tenth of a pixel suddenly it's piracy?

EDIT:
Another thing that infringes significantly less than AI training is fanfiction.

You know Andy Warhol estate was sued successfully for using Prince's image, right?* Might want to find another example before getting on your high horse to pass judgement on why artists and their supporters don't like AI with false assumptions.

*In fact, bringing up Andy Warhol (and the SJC ruling against him) is a huge argument against AI, not for it like you're trying to make it.

 

[Justice and Rule]

Training an AI on an image is not rebroadcasying the image or reprinting the image in any meaningful sense. A given training image has about the same effect on the AI as it would on a human artist who happened to see it once.

But that's not really the case. Again, the thing isn't drawing off itself, but scanning dozens of pictures and continually using them to make their new art. It's not like looking at something once, but a sort of continual tracing.

Not that any of you are going to care, as most of these artists and those of you who support them have made it very clear that the only thing you rrally care about is money, NOT artistic integrity. It's even right there in the title of this very thread. Artistic intrgrity is nothing but an excuse to protect your profits

I mean, it's not like people making AI care about anything other than profits. If they did, then they'd be paying these people or looking for more ethical uses. Instead they are trying to find the best bottom line for themselves. When it comes down to it, I suppose I just fall on the side of the creators instead of the people who are using their artwork for profit without giving them credit or recompense.

Or for that matter, considerably less effect than ut woukd if the human artist in question was Andy Warhol

Why is it ok for Andy Warhol to make a name - and money - for himself on outright trademark infringement but if a computer program borrows a tenth of a pixel suddenly it's piracy?

Warhol got into a ton of trouble. At the end, some attempted to enforce their copyright, while others (like Campbell's) did not.

EDIT:

Another thing that infringes significantly less than AI training is fanfiction.

In practice, not really? Fanfiction is free and generally not for profit. When it becomes profitable, it typically gets changed so that it no longer uses recognizable characters. 50 Shades of Grey is the obvious and immediate example of this.

 

[lynnfredricks]

Indeed. Having regulations is about as "proactive" as we can be.

That's pretty much my point when you were asking for something more. Our present federal laws could really use some work as they for the most part, represent where we were in the late 90s.

 

[Sacrosanct]

Training an AI on an image is not rebroadcasying the image or reprinting the image in any meaningful sense. A given training image has about the same effect on the AI as it would on a human artist who happened to see it once.

This is flat out false. As you can see by the image below. It's pretty much a direct copy. And AI trainers copy the original work without permission to their databases, which is copyright infringement. Also of note, derivitive work is still under protection of the original copyright holder unless it falls under Fair Use (which most AI does not) or is radically different from the original (AI doesn't really do that), and most importantly, "The new, derivative work also can’t have an economic impact on the original copyright holder."---Pike & Lustig LLP

 

[lynnfredricks]

Training an AI on an image is not rebroadcasying the image or reprinting the image in any meaningful sense. A given training image has about the same effect on the AI as it would on a human artist who happened to see it once.

Not that any of you are going to care, as most of these artists and those of you who support them have made it very clear that the only thing you rrally care about is money, NOT artistic integrity. It's even right there in the title of this very thread. Artistic intrgrity is nothing but an excuse to protect your profits

An AI can 'consume' an image down to the individual pixel, including information that can be a challenge to see and compare with the naked eye, as well as any meta data associated with the file. While the 'magic' of AI doesn't really happen at the file level, it is already significant.

The legality of the 'consuming' itself I think (as a layman, not a lawyer) will be a problem given that Google consumes and caches information subject to copyright and the courts seem to be fine with it.

What happens during the training process ultimately comes down to how it is programmed.

A human artist cannot take file data and compare it with millions of other images on the internet and then, run analysis on those images based on the data it already knows. A human artist doesn't take that information and run any number of additional analyses on those.

A human artist cannot easily pattern match across various images created by the same or similar artists to extrapolate various processes that artist uses in their work.

A human artist cannot analyze text, audio, video and other sources to derive additional meta data that could be used to extrapolate various processes that artist uses in their work.

A human artist cannot run all of the analyses above and then extrapolate what new processes could build on the work of the artist.

 

[Bohandas]

But that's not really the case. Again, the thing isn't drawing off itself, but scanning dozens of pictures and continually using them to make their new art. It's not like looking at something once, but a sort of continual tracing.

As opposed to? Do you believe that people just instinctively know what things look like? I suppose you could make an argument that's the case for some elements of the human face and maybe a small handful of other things that consistently played major roles in human life over an evolutionary timescale. But for anything else, and particularly for anything invented in the last several thousand years, you can only draw it because you've seen it before or had it described to you in detail or worked out what it should look like based on similar things that you have seen

EDIT:
And speaking of describing in detail and working out based on similar things, that also works the other way around. If I found a good enough artist and was good enough at describing, and provided they came from the right cultural background to be familiar with all the individual details, I'm sure that artist could produce something passavly similar to those two images that Sacrosanct posted earlier in the page without ever having seen either one of them. ALSO those two images are visibly in different styles; one is a either a photo or photo-like CG and the other is anime. ALSO I suspect that there may have been cheating in the production of that second image; I suspect that that second image is actually the first image ran through an AI filter rather than a true AI reproduction of the first image

 

[Ryujin]

An AI can 'consume' an image down to the individual pixel, including information that can be a challenge to see and compare with the naked eye, as well as any meta data associated with the file. While the 'magic' of AI doesn't really happen at the file level, it is already significant.

The legality of the 'consuming' itself I think (as a layman, not a lawyer) will be a problem given that Google consumes and caches information subject to copyright and the courts seem to be fine with it.

What happens during the training process ultimately comes down to how it is programmed.

A human artist cannot take file data and compare it with millions of other images on the internet and then, run analysis on those images based on the data it already knows. A human artist doesn't take that information and run any number of additional analyses on those.

A human artist cannot easily pattern match across various images created by the same or similar artists to extrapolate various processes that artist uses in their work.

A human artist cannot analyze text, audio, video and other sources to derive additional meta data that could be used to extrapolate various processes that artist uses in their work.

A human artist cannot run all of the analyses above and then extrapolate what new processes could build on the work of the artist.

Hell, I have a hard enough time recognizing images that I took myself, with my own cameras, based on the tens of thousands that I've shot over the years. I have to rely on a trained eye to spot the "signature" of my style and, even then, I mistake others' work for my own on occasion in the relatively small community of photographers, in which I work. Trying to pick out elements of artistic expression, in text or digitally produced image, would be a nightmare if not using a computer to do so.

 

[Justice and Rule]

As opposed to? Do you believe that people just instinctively know what things look like? I suppose you could make an argument that's the case for some elements of the human face and maybe a small handful of other things that consistently played major roles in human life over an evolutionary timescale. But for anything else, and particularly for anything invented in the last several thousand years, you can only draw it because you've seen it before or had it described to you in detail

Humans can make art without reference, AI currently cannot. If it wants to use art (and it has to use that art to make anything) it should be done with permission and possible recompense to those who made it, otherwise it's just piracy. But while humans can do things and be transformative, AI really can't because, at the heart, it's just doing an elaborate trace-job.