Business Musings: AI, Copyright, And Writers
To Get a Free AI Audio Version of this blog, click here.
Here we are—the mess of the mess of the mess. Right now, we’re in one of those technological befuddling moments, where the technology is ahead of the law.
What that means, exactly, is this: We’re not sure what the technology can do, so we don’t know if what it’s doing is legal, in a whole variety of ways.
The law is both a scalpel and a cudgel. If we use the law one way, it becomes a cudgel that smashes behavior and does its best to prevent the behavior from ever occurring again. Look at the laws against homicide in your state. Those laws are not scalpels. Those laws are cudgels, deliberately. As civilized humans, we don’t want other humans to commit murder for any reason. End of story.
(Please don’t write to me about exceptions. I know. I write entire novels about them.)
There are many times, however, that we need the law to be a scalpel. We need it to delicately carve good behavior from bad. We also don’t want it to accidentally smash something good to smithereens.
Just today, Dean and I were walking home in a wind tunnel created by the buildings near ours. The wind was bad anyway, but in that little area, it was extreme, like usual. Dean mentioned that there are entire computer programs that could explain why.
Those programs are often used now to examine how the wind works around bridges and tall buildings in relation to other tall buildings. In the past, those calculations were done by engineers and often by hand. One mathematical error and even brand-new bridges and buildings collapse.
If you don’t believe me, look at the footage of the Tacoma-Narrows Bridge before its famous collapse in November of 1940. Technology, math, and engineering of the time simply didn’t allow for prevention in that case. Like so many engineering problems of the mid-20th century, the problems were discovered after the emergency and were added as cautionary tales for designers going forward.
Now, though, tech allows us to prevent all kinds of wide-ranging disasters because of computer modeling.
In some ways, generative artificial intelligence in art, audio, and writing is nothing more than computer modeling. The artificial intelligence isn’t intelligence at all, at least as we know it. It’s an algorithm trained to respond in a particular way to a variety of inputs.
The inputs make the AI program reactive, not creative. My post last week titled “AI And Mediocre Work” dealt with a lot of this, but a comment by Matt Weber capsulized it with a quote from Oliver Sacks, in his book, An Anthropologist on Mars:
Creativity, as usually understood, entails not only a “what,” a talent, but a “who” — strong personal characteristics, a strong identity, personal sensibility, a personal style, which flow into the talent, interfuse it, give it personal body and form. Creativity in this sense involves the power to originate, to break away from the existing way of looking at things, to move freely in the realm of the imagination, to create and recreate worlds fully in one’s mind — while supervising all this with a critical inner eye.
These generative AI programs are useful for a variety of things, some of them mentioned in the comments on the last post, others mentioned in analysis about the programs that you can find most anywhere. What they are not is creative.
Let’s set that aside, though. We will all end up using these programs for one task or another.
What started this little miniseries of blogs was, in fact, my desire to start using AI audio. It had gotten to a level that I feel comfortable putting not only the blog posts into audio, but some of the nonfiction books as well. If you want to find out what I’m thinking about the various audio opportunities for my own work, please look at this post.
Up until that point, a lot of my readers thought I was opposed to using generative AI. I’m not. I have already used several different programs for minor things, and I’m going to use others for relatively major things.
I’m just as interested in the AI art programs as I am in the AI audio programs. I’ve used some mapping programs to help artists visualize the layout of my various worlds. I’m using the free programs, so the tools are often wrong in a variety of ways. I have to use words and bad maps to get my point across. But that’s okay.
I like some of the art I’m seeing from the various programs, and that art would be good enough to use on, say, short story ebook covers, where we don’t want to spend a lot of money. (If any.)
We’re not doing that yet, though, and there’s a really good reason.
The copyright issues on much of the AI usage are a complete mess and that, in my opinion, makes them dangerous to use in any commercial manner.
I don’t use the word “dangerous” lightly. Copyright issues could mean something as simple as removing the item from sale to hundreds of thousand paid in statutory damages.
The problem is that we don’t know what’s happening yet, and because we don’t know, we have to be really careful.
Some of the copyright issues can be resolved with a contract. The Terms of Service on these sites are contracts that you agree to, either by affirmatively clicking I accept or by using the site or by paying money for the service.
The problem with Terms of Service is that they can change on a whim. In its paper on artificial intelligence and copyright published in February, the Congressional Research Service made the passing comment about OpenAI, the developer of ChatGPT and DALL-E.
As I said, these terms can change drastically. It’s up to the user to check the terms constantly.
Contracts can supercede copyright if done properly, but doing the contracts properly means understanding the law.
And the law is just plain unclear. The article that I quoted above, from the Congressional Research Service, has a good overview of where the law stands right now in the U.S., and provides links.
The law is different in different countries—and is changing so fast that four hours after I finished this post, CNBC put up an article titled “Italy has banned ChatGPT. Here’s What Other Countries Are Doing.” Not all other countries, mind you, but some of the big ones. Please read this.
In a very good overview article, the World Intellectual Property Organization examines the different ways countries have already addressed some aspects of copyright and A.I. It notes that many countries “are not amenable to non-human copyright.” It cites the United States, which we will look at in a moment, and also Australia, saying,
…[in] a recent Australian case (Acohs Pty Ltd v Ucorp Pty Ltd), a court declared that a work generated with the intervention of a computer could not be protected by copyright because it was not produced by a human.
The result is a little less clear in Europe. The Court of Justice of the European Union stated in several cases that
…copyright only applies to original works, and that originality must reflect the “author’s own intellectual creation.” This is usually understood as meaning that an original work must reflect the author’s personality, which clearly means that a human author is necessary for a copyright work to exist.
But other countries, including New Zealand, Hong Kong, Ireland, India, and the United Kingdom, seem to give the copyright in the programmer.
Neither of these instances give the copyright to the person who used the AI program to generate the artwork. In the U.S., for example, that work will have no copyright protection and will be in the public domain. In the United Kingdom, the programmer who created the program the author used to create that cover might actually own the copyright in that art.
The cases get complicated, though, when the thing that the computer program spews out is altered by a human being. If you have a chatbot plot your novel and then you use that plot as the basis for the work, taking bits of that result but not all of it—have you created a new work? A derivative work? Is it something created by a human being or is it something that is strictly speaking computer generated?
(Please don’t try to answer this in the comments. Please.)
So…let’s say you don’t care who owns the copyright of the piece of art you generated through an AI program. Let’s say that if the art’s in the public domain, that’s okay with you.
But these matters are unsettled and as they work their way through the courts, so does a ticking timebomb.
The timebomb is the matter of copyright infringement by generative AI.
The programs were created by scraping the internet, searching for imagery (in the case of the art programs) and writing (in the case of the chatbots). The scrapes are exhaustive. Even if something has copyright protection, it probably got scraped.
For example, last fall, I wrote a Spanish paper that touched, ever so briefly, on Picasso’s Guernica. While researching, I got sidetracked by looking at who owned the copyrights for Picasso’s paintings, and found a complaint by a teacher. She couldn’t get “good” quality slides of any of his work because the Picasso Foundation and various museums vigorously defend the copyright. But poor quality images of all of his work exist all over the web. I can guarantee that they were scraped by AI art programs, along with the work of Rembrandt, which is clearly in the public domain.
Copyright here is a mess, because the scrapings also include artists who are still alive, still producing, and still hoping for commissions. You can generate a piece of art on any one of the sites by asking for an image in the style of your favorite artist.
In most cases, you’ll get a fair (if mediocre) approximation. In many cases, you’ll get an almost exact reproduction of a piece of artwork that is under copyright.
Right now, the programmers are declaring that scraping is fair use. The artists say the scrapings are copyright violations. In the United States, a finding of copyright infringement is a statutory one, meaning that the damages are by statute. And that can mean that the person who infringed can be charged anywhere from $750 to $30,000 per infraction. Meaning that if I got a cover similar to a Picasso and it was an infringement, and sixteen of my friends also got infringing Picasso knock-offs, the copyright owner of that new piece of art (the programmer? The person who requested it?) could be charged nearly a half million dollars in damages.
If that infringement was deemed intentional (and how could it not be, if you asked for something “in the style of”), then the amount per infraction goes up to as much as $150,000. In our make-believe case, that’s 2.4 million dollars. As the online legal site Justia says in its article on copyright infringement,
When a case involves many incidents of infringement, statutory damages can become immense.
The real risk is not having to take your cover down. The real risk is being party to a copyright infringement lawsuit that could bankrupt you and take away your own originally created copyrights.
In examining who would be liable for copyright infringement in the above case in the U.S., the Congressional Research Service says this:
Under current doctrines, both the AI user and the AI company could potentially be liable. For instance, even if a user were directly liable for infringement, the AI company could potentially face liability under the doctrine of “vicarious infringement,” which applies to defendants who have “the right and ability to supervise the infringing activity” and “a direct financial interest in such activities.”
Right now, a large number of cases are wending their way through the courts here in the U.S. I don’t know if the same thing is happening in other countries, but I suspect there are a large number of cases there as well.
It will take a good five years minimum for this to shake out in the U.S., maybe more if Supreme Court gets involved. In fact, I’ve been quoting an article from the Congressional Research Service, which exists “at the behest of and under the direction of Congress” to look closely at various issues that might require a legislative solution. The conclusion of February’s analysis from the CRS is this:
Given how little opportunity the courts and Copyright Office have had to address these issues, Congress may wish to adopt a wait-and-see approach. As the courts gain experience handling cases involving generative AI, they may be able to provide greater guidance and predictability in this area through judicial opinions.
Writers need to adopt a wait-and-see approach as well, particularly if they (misguidedly) include anything from a chatbot in their own writings.
As for using other AI services like art, be exceptionally careful. Do not use the generated art for commercial purposes, which is, essentially, in an enterprise designed to make a profit. If you’re unclear about what that means for your project, then someone else will be too.
When you’re unclear and it could lead to a violation of the law, then don’t do it. Period. If you feel that you could defend your actions in court, let me point something out:
You’d have to go to court to defend that action.
There would be legal fees and other consequences, especially if you’re wrong and you lose.
It’s better to wait until this all shakes out.
So how does it all square with me using AI audio right now? Well, I’m being very careful. I look at the terms of service. I’ve rejected several organizations based on the lack of clarity in their terms of service. The lack of clarity might be about ownership of copyright (and in one case it was) and some of it might be a badly designed TOS. One TOS gave me permission to use the audio file for commercial purposes in the copyright section and then repeatedly stated that the audio file can only be used for non- commercial purposes throughout the rest of the TOS.
Most likely, some minion forgot to clean up the language in the TOS and only changed the copyright section. But do I know that for sure? Or was I looking at a typo in the copyright section where the organization left off the word “non”?
There’s no way to know, and I’m not going to court to find out, so I’m not doing business with them.
I’m also using only the generated AI voices. I’m not using voices in the style of Jennifer Lopez or Denzel Washington. I don’t want to risk using a cloned voice of someone famous for the work I’m doing.
And right now, I’m not letting them clone my voice either. Not that my voice is worth a lot of money, but it might be someday. Who knows?
It’s all unsettled—and I don’t like uncertainty when it comes to my copyrights.
Neither, by the way, do other entities. For example, Neil Clarke at Clarkesworld is very clear about this. When he reopened Clarkesworld for submissions in March, he wrote,
There’s also the matter of rights, specifically a lack of clear ownership of these works. I simply can’t sign a contract without those.
So even if you did do something cool and creative somehow with a chatbot as your basis, you can’t license it to anyone who understands copyright, because there’s no way to know who owns the work you’re trying to license.
As I said in the beginning, this is a mess of a mess of a mess. It won’t be settled for some time.
So don’t be in a hurry. Use the tools that you can safely use for commercial purposes (if you understand Terms of Service). Use the other tools for non-commercial purposes if you use them at all.
I would suggest that when it comes to copyright, you learn to cultivate a very low risk tolerance. You could lose everything if you guess wrong.
And really, is a shortcut worth losing everything?
It isn’t to me. I hope it isn’t to you as well.
Update on April 8, 2023. As I was writing this, apparently, the United States Copyright Office issued new guidelines for works containing the use of AI-generated material. Here’s the link. You need to look at this if you live in the U.S. As I mentioned above, laws are different in different countries. And these guidelines end with this conclusion:
This policy statement sets out the Office’s approach to registration of works containing
material generated by AI technology. The Office continues to monitor new factual and
legal developments involving AI and copyright and may issue additional guidance in the
future related to registration or the other copyright issues implicated by this technology.
Most writers don’t understand copyright, figuring they can always learn it later. I write a lot of blog posts on copyright, and you can search for those on this website. Or you can take the Bite-Size Copyright class that we offer through WMG Workshops. The second quarter’s classes are starting now.
I highly recommend it. You’ll spend a little bit of time each week learning a different aspect of copyright. We are not lawyers, so don’t expect this to be a class giving legal advice. The class does make you aware of the issues that creatives need to know under copyright law, and gives you places to find out more.
If you prefer books, Nolo Press’s book on copyright is invaluable. You’ll get a short overview of copyright and other contractual matters in my book, Closing The Deal On Your Terms.
Do yourself a favor and learn this stuff. It’s best to do so in small doses, which is why we set up the Bite-Size Copyright course.
In the meantime, here’s a reminder:
This weekly blog is reader supported.
If you feel like supporting the blog on an on-going basis, then please head to my Patreon page.
If you liked this post, and want to show your one-time appreciation, the place to do that is PayPal. If you go that route, please include your email address in the notes section, so I can say thank you.
Which I am going to say right now. Thank you!
Click paypal.me/kristinekathrynruschr4e to go to PayPal.
“Business Musings: AI, Copyright, and Writers,” copyright © 2023 by Kristine Kathryn Rusch. Image at the top of the blog copyright © Can Stock Photo / studiostoks.
Several companies are now offering AI-generated art work created using fully-permissioned training data.
They include Shutterstock, Adobe, and Getty.
In addition, Stable Diffusion — the main open source alternative — has an “opt out” list for artists and it says that 80 million pieces of art have already been added to the opt-out list.
I personally don’t think that an opt out is as good as an opt in, and hope that Stable Diffusion and Midjourney will soon switch to an opt-in model.
I wrote about this topic here:
And I’ve also checked out Adobe’s Firefly, which is another fully-licensed, permission training data model, and liked it very much.
Now, these platforms aren’t as good as Midjourney yet — but that’s because Midjourney has had several rounds of updates and retraining, based on user inputs. People were able to vote to see which images they liked and which they didn’t, and the developers updated the tool correspondingly.
I think we’re going to see the same kind of improvements with Adobe and the other fully-licensed platforms over the next year.
Plus, Adobe’s Firefly will probably be integrated into Photoshop (if it hasn’t been arleady) making it even easier for cover designers to integrate into their workflows.
From Midjourney TOS:
“Subject to the above license, You own all Assets You create with the Services, to the extent possible under current law. This excludes upscaling the images of others, which images remain owned by the original Asset creators. Midjourney makes no representations or warranties with respect to the current law that might apply to You. Please consult Your own lawyer if You want more information about the state of current law in Your jurisdiction.”
It specifically states that you own all Assets you create with the services.
Then adds “to the extent possible under current law”.
Any thoughts about this legalese?
As I have said repeatedly, it’s impossible to understand any legal document without reading the entire document. So, no, I have no thoughts on the legalese.
Very interesting (and very worrying!).
How can we know that images that we license for our book covers from reputable royalty-free sites such as iStock haven’t been AI-generated and therefore have these copyright images?
On iStock, some images are tagged ‘digitally generated’. I don’t know what exactly that means.
Should we be asking questions about images we buy from such sites? If so, what would be the questions that differentiate between images that are mere GCI and those that have this copyright issue?
(Not expecting you to know all the answers, Kris, I’m just spouting my questions!)
If anything says digitially generated or has AI listed, I don’t use it right now. Also, I’m trying to purchase stock art that was put on the site before 2021. That works too. And yes, you should be asking questions about the images.
Good questions all, Amy.
Thanks, Kris – that’s a good idea about 2021.
Any tips on searching for pre 2021 images?
Adobe, for example, has no date filter. They claim to not have access to upload dates, yet you can sort results by most recent. That also seems true with other image sites that I’ve checked out. Best idea I’ve been able to come up with is to sort by most recent, jump to the bottom of the results, and work up instead of down. Still doesn’t give you an exact date, though.
Even a reverse image search in Google doesn’t give upload date (that I know of).
I use iStock and they give the upload date on each image’s info page.
Turns out iStock, Pexels, Pixabay, (maybe) Unsplash list upload dates.
Shutterstock and Adobestock do not,
Can also download a watermarked copy and use Tineye to see if/when said image was ever uploaded anyplace.
Interestingly, Shutterstock and Adobestock allow “artists” to post AI images for sale. Their TOS requires these images to be identified as AI. Shutterstock has an actual toggle in their search to filter out AI. Adobestock recommends including “-AI” in your search term. All of this assumes that the uploader actually identifies AI as such.
Question is who would take the fall if a writer unknowingly used an AI image in a cover because the person uploading it failed to disclose it as being AI? At a minimum, you’d have to deal with the headache of taking it down and replacing it.
And hiring someone to do your covers . . . you’d probably never know.
Thanks so much for this post, Kris! I have gotten the same accusations that I’m against AI. I’m not, but I also don’t want to end up nearly destitute like photographer David Slater (Naruto, et al. v. Slater et al, 9th Circuit Court of Appeals AKA the Monkey Selfie Case) even though he ultimately won his copyright.
We could have technically perfect music – music that is EXACTLY as written – but we don’t; we go to concerts. Even with the variability modeled in by more complicated algorithms, we don’t want mechanical music (except in its own small niche). I assume there is out there music produced by computers without lips or real strings, but it is probably not classified as ‘the good stuff’ – more in the category of elevator music without the need to pay a license for.
But if I wanted to have Beethoven’s Ninth playing in the background of my indie movie, I could write a program which would turn the score into sound. Assuming I checked out what rights having a copy of the score gives me.
Best to stick with original material everywhere – which is what copyright was created to encourage.
Then don’t forget to register the copyright to your own stuff.