Publishers demand to be paid for AI using their work

AppleInsider

ChatGPT, Bard, and all AI tools are trained on existing work, and quote that work without the originators being paid or cited in most cases -- and publishers say this can't continue.

After years of Machine Learning being trained on existing published works, AI tools like ChatGPT and Bard have suddenly become mainstream. Even in their claimed beta or early versions, AI tools are being integrated into Microsoft 365, Adobe Firefly, and more.

According to the Wall Street Journal, publishers have been investigating how their content has been used in training AI tools, and examining legal options. Citing sources familiar with the meetings, the newspaper says that publishing trade group News Media Alliance wants members to be paid.

"We have valuable content that's being used constantly to generate revenue for others off the backs of investments that we make," Danielle Coffey, of News Media Alliance told the publication, "that requires real human work, and that has to be compensated."

News Corp's CEO, Robert Thomson, said at a recent investor conference that the company has "started discussions with a certain party who shall remain nameless."

"Clearly, they are using proprietary content -- there should be, obviously, some compensation for that," he continued.

AI tools search the internet for the basis of their responses to user requests, and the Wall Street Journal points out there is a "fair use" legal provision that might cover some of this.

"We've done a lot with fair use," said Sam Altman, CEO of OpenAI, the company that created ChatGPT. "[However, we're] willing to pay a lot for very high-quality data in certain domains [such as science]."

When Microsoft's implementation of AI returns a response, it includes links to the sources it used. Google's Bard, however, makes at most a text reference to a source, rather than links or citations.

Google Assistant vice president Sissie Hsiao told the Wall Street Journal that the company "is deeply committed in supporting a healthy and vibrant content ecosystem." When AI tools are integrated into search, Google "will be welcoming conversations with stakeholders."

Microsoft has already been making payments to publishers for use of material on MSN, but reportedly those do not cover AI. Microsoft has not commented.

Google has similar deals with firms including News Corp, for a service called Google News Showcase, but again the payments do not include AI use.

Separately, when Adobe announced its Firefly AI tool, the company said that it was planning to have payment systems for artists before the app comes out of beta.

Read on AppleInsider

dwaric

Human artists also learn/train from other artists work. Should they also compensate all the artists whose work they have seen?
Just a thought.

applezulu

Dwaric said:

Human artists also learn/train from other artists work. Should they also compensate all the artists whose work they have seen?
Just a thought.

In copyright infringement cases involving humans, the court determines if the alleged infringer was exposed to the prior work and then essentially assesses if the new work is creatively novel, or if it is simply derivative. The same process applies to AI. 

If an AI system catalogs prior works and then simply regurgitates catalog content in a cut-and-paste collage, then it’s infringement. At least based on what I’ve seen of current AI, that’s all it’s capable of. There is no actual learning followed by a creative spark in those algorithms. It’s like looking at a painting by an artist whose influences are way too obvious, or listening to a band playing songs with arrangements and vocal affectations that are indistinguishable from anything else on this week’s charts. 

Here, the onus is on the AI developers to prove how what they’re up to is not just derivative regurgitation that is nothing more than the sum of its parts. 

foregoneconclusion

Dwaric said:

Human artists also learn/train from other artists work. Should they also compensate all the artists whose work they have seen?
Just a thought.

The "learning/training" lingo is just marketing. AI programs don't function without access to every letter/pixel of gigantic databases full of writing and art. There's no thinking, just brute force. No access to database = no production of writing or art. It's not a parallel to how humans create despite all of the marketing smoke and mirrors from the tech companies. Fair use laws don't apply because those laws were created for human capabilities, not for computer programs. 

retrogusto

Influences are in most cases clearly evident in human artists’ work in all disciplines and at all levels of competence, if you know what to look for. As Picasso said, “good artists borrow, great artists steal.” Just look at his work, and then look at the work of Georges Braque and Julio Gonzalez, whose work Picasso “stole” in this sense. There’s no shortage of examples in music and other art forms too. At this point, the most sophisticated automated systems can create works that are at least as original as what most humans are coming up with (which is a low bar), but of course there are humans pulling the strings here, designing the algorithms and training them, so you still easily argue that it’s just a more complex and technology-intensive form of human creation. 

foregoneconclusion

AppleInsider said: AI tools search the internet for the basis of their responses to user requests, and the Wall Street Journal points out there is a "fair use" legal provision that might cover some of this.

"We've done a lot with fair use," said Sam Altman, CEO of OpenAI, the company that created ChatGPT. "[However, we're] willing to pay a lot for very high-quality data in certain domains [such as science]."

Fair use can't possibly apply here because the programs aren't capable of producing anything without the database. 100% of what they produce is entirely dependent on work that was created by human beings. The fact that 100% might be made up of 1% of the material from 100 different human beings doesn't matter. Look at Adobe: they've already admitted that "AI" producing art based on their stock library requires compensation to the artists that they license stock art/photos from. 

omasou

foregoneconclusion said:

Dwaric said:

Human artists also learn/train from other artists work. Should they also compensate all the artists whose work they have seen?
Just a thought.

The "learning/training" lingo is just marketing. AI programs don't function without access to every letter/pixel of gigantic databases full of writing and art. There's no thinking, just brute force. No access to database = no production of writing or art. It's not a parallel to how humans create despite all of the marketing smoke and mirrors from the tech companies. Fair use laws don't apply because those laws were created for human capabilities, not for computer programs. 

It seems you don't understand how machine learning models are created/trained. I agree with @foregoneconclusion, if I become an expert in my field through study, training, attending meetings, etc. I don't owe anything to content creators, I've already paid them for the books, training, meetings, etc.

The AI was trained on publicly available data, which the authors provided to further their own self-promotion as an expert, painter, etc. There is no difference between the AI returning results and Google returning results. Except the former is creating new works based on it's knowledge and training.

edited March 2023

chutzpah

omasou said:

foregoneconclusion said:

Dwaric said:

Human artists also learn/train from other artists work. Should they also compensate all the artists whose work they have seen?
Just a thought.

The "learning/training" lingo is just marketing. AI programs don't function without access to every letter/pixel of gigantic databases full of writing and art. There's no thinking, just brute force. No access to database = no production of writing or art. It's not a parallel to how humans create despite all of the marketing smoke and mirrors from the tech companies. Fair use laws don't apply because those laws were created for human capabilities, not for computer programs. 

It seems you don't understand how machine learning models are created/trained. I agree with @foregoneconclusion, if I become an expert in my field through study, training, attending meetings, etc. I don't owe anything to content creators, I've already paid them for the books, training, meetings, etc.

The AI was trained on publicly available data, which the authors provided to further their own notoriety to be known as an expert, painter, etc. There is no difference between the AI returning results and Google returning results. Except the former is creating new works based on it's knowledge and training.

The authors provided those works to people expecting a return, likely from advertising.  Training an AI on that content is not an expected use, and some might say not a reasonable one.

foregoneconclusion

retrogusto said:

Influences are in most cases clearly evident in human artists’ work in all disciplines and at all levels of competence, if you know what to look for. As Picasso said, “good artists borrow, great artists steal.” Just look at his work, and then look at the work of Georges Braque and Julio Gonzalez, whose work Picasso “stole” in this sense. There’s no shortage of examples in music and other art forms too. At this point, the most sophisticated automated systems can create works that are at least as original as what most humans are coming up with (which is a low bar), but of course there are humans pulling the strings here, designing the algorithms and training them, so you still easily argue that it’s just a more complex and technology-intensive form of human creation. 

"AI" programs don't operate by influence. They operate by accessing databases of exact copies of writing/art. There is no originality in "AI" produced material because it is entirely dependent on the database of those exact copies. Turn off the access and the "AI" is non-functional. A human being that operates by influence is not. A human being can produce writing/art in an empty room with no art/books available and no internet access. 

robin huber

The internet and all it’s freebies were fun while it lasted, but no more free lunch. 

omasou

chutzpah said:

omasou said:

foregoneconclusion said:

Dwaric said:

Human artists also learn/train from other artists work. Should they also compensate all the artists whose work they have seen?
Just a thought.

The "learning/training" lingo is just marketing. AI programs don't function without access to every letter/pixel of gigantic databases full of writing and art. There's no thinking, just brute force. No access to database = no production of writing or art. It's not a parallel to how humans create despite all of the marketing smoke and mirrors from the tech companies. Fair use laws don't apply because those laws were created for human capabilities, not for computer programs. 

It seems you don't understand how machine learning models are created/trained. I agree with @foregoneconclusion, if I become an expert in my field through study, training, attending meetings, etc. I don't owe anything to content creators, I've already paid them for the books, training, meetings, etc.

The AI was trained on publicly available data, which the authors provided to further their own notoriety to be known as an expert, painter, etc. There is no difference between the AI returning results and Google returning results. Except the former is creating new works based on it's knowledge and training.

The authors provided those works to people expecting a return, likely from advertising.  Training an AI on that content is not an expected use, and some might say not a reasonable one.

What people expect (hope for) and what people get are not always the same. If it is not gated behind a pay wall then it is in the public domain. If they don't like that then should put it in a walled garden and make less.

IMHO this is the whole problem w/the internet and social media today. In the past on forum's people shared knowledge to help others. Period.

Now everyone publishes content in an attempt to become an influencer and make it big. Spotify, Google, etc. make billions on the content of others via advertising. Is that OK? Not according to the smaller artists.

So how is crawling the web for indexing different than crawling the web for knowledge? Oh, b/c the content creators haven't figured out how to monetize it, YET. Selfish them.

As everyone on Facebook knows or should know. YOU ARE THE PRODUCT.

edited March 2023

9secondkox2

It’s only right. They did the work. If you’re going to offer their work, you need to pay them. 

dewme

This does bring up some interesting scenarios. If the creators of the AI are skirting responsibility for citation and attribution of others' works because, in their mind at least, conventional standards for doing those things do not apply to AI, how are we going to reconcile cases when AI is the direct cause of something that leads to injury or death of a human?

Who will be held responsible for AI gone bad? I'm not talking about Alexa Astro choking you out with a pillow, I'm talking about AI presenting advice, a recipe, our suggesting an action that appears to be based on data  the AI considers to be knowledge but is instead based on faulty logic or reasoning from the source,  faulty logic or inferences in the AI processing itself, or a source that turns out to be nothing more than the demented musings of a psychopath? 

Until we get to the point where we can clearly answer the question: "Who are you going to sue when AI does the wrong thing?" we're stumbling blindly into uncharted territory.

robin huber

AI today is much like a very precocious genius child, filled with facts and information but utterly lacking in judgement, wisdom and discrimination. 

edited March 2023

mattinoz

Dwaric said:

Human artists also learn/train from other artists work. Should they also compensate all the artists whose work they have seen?
Just a thought.

Compensate by way of attribution or acknowledgement of the influence.Yes they should and get called out if they don’t.

sometimes the connects and path are as interesting as the destination 

chasm

omasou said:

If it is not gated behind a pay wall then it is in the public domain. If they don't like that then should put it in a walled garden and make less.

1. This flatly isn’t true. You apparently do not have any grasp of copyright.

2. PLEASE say this foolish statement to an actual copyright holder or lawyer, and film their reaction to post here.

Wear a bib, or your shirt will get dirty from all the spit as they laugh right at you.

irnchriz

The publishers could demand that each piece of work created list the generating seed command then list all sources used then demand payment for each source.  Of course, this could end up being thousands of sources and very expensive.

edited March 2023

gatorguy

9secondkox2 said:

It’s only right. They did the work. If you’re going to offer their work, you need to pay them. 

chasm said:

omasou said:

If it is not gated behind a pay wall then it is in the public domain. If they don't like that then should put it in a walled garden and make less.

1. This flatly isn’t true. You apparently do not have any grasp of copyright.

2. PLEASE say this foolish statement to an actual copyright holder or lawyer, and film their reaction to post here.

Wear a bib, or your shirt will get dirty from all the spit as they laugh right at you.

https://www.thefashionlaw.com/ai-trained-on-copyrighted-works-when-is-it-fair-use/
It's an excellent explanation of the arguments on both sides. 

retrogusto

foregoneconclusion said:

retrogusto said:

Influences are in most cases clearly evident in human artists’ work in all disciplines and at all levels of competence, if you know what to look for. As Picasso said, “good artists borrow, great artists steal.” Just look at his work, and then look at the work of Georges Braque and Julio Gonzalez, whose work Picasso “stole” in this sense. There’s no shortage of examples in music and other art forms too. At this point, the most sophisticated automated systems can create works that are at least as original as what most humans are coming up with (which is a low bar), but of course there are humans pulling the strings here, designing the algorithms and training them, so you still easily argue that it’s just a more complex and technology-intensive form of human creation. 

"AI" programs don't operate by influence. They operate by accessing databases of exact copies of writing/art. There is no originality in "AI" produced material because it is entirely dependent on the database of those exact copies. Turn off the access and the "AI" is non-functional. A human being that operates by influence is not. A human being can produce writing/art in an empty room with no art/books available and no internet access. 

You mean a human can create music having never heard music? Maybe, depending on how you define music. So can a bird or a wind chime. A computer can produce writing/art in an empty room with no art/books available and no internet access, as long as it has previously had exposure, same as a human.

applezulu

retrogusto said:

foregoneconclusion said:

retrogusto said:

Influences are in most cases clearly evident in human artists’ work in all disciplines and at all levels of competence, if you know what to look for. As Picasso said, “good artists borrow, great artists steal.” Just look at his work, and then look at the work of Georges Braque and Julio Gonzalez, whose work Picasso “stole” in this sense. There’s no shortage of examples in music and other art forms too. At this point, the most sophisticated automated systems can create works that are at least as original as what most humans are coming up with (which is a low bar), but of course there are humans pulling the strings here, designing the algorithms and training them, so you still easily argue that it’s just a more complex and technology-intensive form of human creation. 

"AI" programs don't operate by influence. They operate by accessing databases of exact copies of writing/art. There is no originality in "AI" produced material because it is entirely dependent on the database of those exact copies. Turn off the access and the "AI" is non-functional. A human being that operates by influence is not. A human being can produce writing/art in an empty room with no art/books available and no internet access. 

You mean a human can create music having never heard music? Maybe, depending on how you define music. So can a bird or a wind chime. A computer can produce writing/art in an empty room with no art/books available and no internet access, as long as it has previously had exposure, same as a human.

Humans are capable of creating something new. Yes, most will learn from prior artists’ work, then start creating their own, and at first it will be bad, and it will transparently show its influences. Good artists then move beyond that and create something unlike what’s been done by others before. That novelty comes from the quirks of the individual artist’s own body and brain, and also from intentionality based on knowledge of the subject and a specific desire to create their own vision. Thus far, at least, artificial intelligence hasn’t gotten to those last steps and created anything that’s more than the sum of its parts. 

Particularly because it’s possible to examine exactly what material has been cataloged by an AI program, what instructions the programmers have written into the algorithm, and what instructions the end user has given to order up a specific output, it’s possible to examine an AI application’s entire ‘creative’ process from beginning to end.  Until such examination can conclude that a given result was unpredictable, an actual surprise, it’s reasonable to determine that all output of such a program is entirely derivative of the artistic works cataloged at the beginning of that process. 

It’s harder to examine a human’s entire artistic process, but it is in fact that unpredictable result in an artist’s output that makes art art.

omasou

chasm said:

omasou said:

If it is not gated behind a pay wall then it is in the public domain. If they don't like that then should put it in a walled garden and make less.

1. This flatly isn’t true. You apparently do not have any grasp of copyright.

2. PLEASE say this foolish statement to an actual copyright holder or lawyer, and film their reaction to post here.

Wear a bib, or your shirt will get dirty from all the spit as they laugh right at you.

and what exactly is your point? Are you attempting to say that...if read and learn from copyrighted material and when asked a question, that I am not allowed to answer or present on the topic b/c my knowledge is based on copyright material? Think about that.

I'm sure there are a team of lawyers much smart than you or me that have reviewed this every which way and come to the conclusion that the output falls under "fair use".

edited March 2023