Episode 26: Who owns AI-generated creations (and why you should care)


Lisa Desjardins: You are listening to Canadian I.P. Voices, a podcast where we talk intellectual property with a range of professionals and stakeholders across Canada and abroad. Whether you are an entrepreneur, artist, inventor, or just curious, you will learn about some of the real problems, and get real solutions for how trademarks, patents, copyrights, industrial designs, and trade secrets work in real life. I'm Lisa Desjardins, and I'm your host. The views and opinions expressed in this podcast are those of the individual podcasters, and do not necessarily reflect the official policy or position of the Canadian Intellectual Property Office.

Artificial intelligence, A.I. has been around for many decades, and recently many companies have opened their A.I. platforms for the public. Most people can now use A.I. to create text, music, images, even whole essays. No doubt it's a very interesting time for A.I. developers to see their efforts come to fruition, but it's also a challenge in the context of laws of ownership of these creations.

In schools, teachers are left wondering who actually wrote the essay, and in copyright terms, who is the author? Countries are forced to decide whether they will grant copyright to A.I. creations, which introduces another problem since most countries are signatories to the Berne Convention, which essentially gives at least human-made original creations automatic copyright protection. How will this new class of creations be treated if the national laws are going to be different? And, the challenges don't stop at granting or not granting copyright to A.I. creations.

If A.I.-created works look like works already copyrighted, can the A.I. be sued? Maybe, just maybe, we can ask an A.I. to solve the legal problems. To learn more, we meet with Guillaume Lavoie Ste-Marie, who is a principal at Smart & Biggar Montreal. Guillaume is a copyright patent and trademark litigator, and senior associate at Smart & Biggar since 2011, and has extensive experience in litigation in complex I.P. matters. Guillaume, I'm really thrilled to have this conversation with you. It is such a hot topic, so welcome to the podcast.

Guillaume Lavoie Ste-Marie: Thank you for having me. This is indeed a very interesting topic in a field of technology that has evolved and will definitely continue to evolve. I'm glad to be able to speak to you and your listeners about it today. Maybe the answers I give you today will change if we have the same discussion next year or the following, because it's going to be exciting to see how the law evolves on this point in the coming years.

Lisa: Very important point. Speaking about the law, this is evolving, and so I guess we should perhaps label this as a non-advisory conversation because it is such a new field.

Guillaume: Exactly. A lot of the answers are going to be maybes [laughs] and it depends. I'll do my best obviously to be as helpful as I can, but this will definitely continue to evolve.

Lisa: Before we start, can you talk a little bit about yourself and what you do at Smart & Biggar?

Guillaume: Yes, of course. As you said, I'm an intellectual property litigation principal at Smart & Biggar Montreal. That's where I've practiced my entire law career. We act mainly in contentious matters in the federal court, federal court of appeal. Recently a lot in copyright matters, online piracy. Also patent matters in various fields of technology and obviously trademarks, and also what I would call soft I.P. aspect. Image rights, confidential information, non-compete clauses, this type of thing. Very interesting field. Like A.I. and like the law in general, it keeps evolving.

Lisa: We were talking about A.I. We know these days A.I. creates sounds and images, and texts. I've tried it myself, it's a lot of fun. From a copyright perspective, who is the creator versus who is the owner over these creations?

Guillaume: Unfortunately, there is not a clear-cut answer to that question, and I'll explain why. I think I can still provide some guidance to help everyone understand the applicable law, and maybe make their own decision in the specific facts of their own case. There is no A.I.-specific scheme in our Copyright Act because A.I. is such a new thing and obviously the Copyright Act is amended from time to time, but understandably, it has difficulty keeping up with technological advances. Maybe to start, it's important to understand that copyright law gives authors a monopoly over what the Copyright Act calls "works."

Examples could be a text, music, images, film. That's all works under the Copyright Act. For a work to be protectable by copyright, it has to be original. Now, in Canada original does not necessarily mean that it has to be groundbreaking or extremely creative. It only means that the author of the text, the music, et cetera, has to exercise a certain level of skill and judgment in the creation of the work. I think there are good indications in the language of the Copyright Act that suggests that the author of a work has to be a human being. The Copyright Act talks about the author in certain circumstances has to be the resident of a certain country, or the author can be an employee.

As it currently stands, I would argue that the author of a work has to be a human being. Presumably, it could not be the A.I. itself. Now the question becomes, can it be the user giving input to the A.I. to create a work and the A.I. then produces that as an output? Music, text, and image, et cetera. In order to answer that question, we have to determine if the user of the A.I. is controlling the A.I. sufficiently in its inputs to exercise the sufficient skill and judgment necessary for copyright protection. The fact that you're using a tool to create a work is not in itself an obstacle.

People use digital audio interfaces all the time to create music, and they use computer software to create images. It's not the use of a tool or a technological tool that is a potential problem. It's the notion of degree of how much contribution the user has on the work. Maybe to give you a practical example, if you use an A.I. to generate images, and you give it a prompt, for example, make me a painting of Montreal in the winter, the output of the A.I. will be a painting, it will probably look pretty good. But in my view, that probably would not be protected by copyright because the input given by the user was merely an idea.

It was just a concept but it was not the exercise of any level of artistic skill and judgment to ask an A.I. to paint a Montreal landscape in the winter. However, if I generate a text myself as a user of an A.I., before I interact with the I generate a text, for example, for a blog post, or maybe I have a piece of music that I recorded in my living room, and I provide this to the A.I. For the text, let's say I asked the A.I. correct any mistakes, or for the music, I asked the A.I., change the pitch of the tune or remove some noise that I don't like.

Various A.I.s can do these things. Then in these contexts, I think the output would be protected because the user has generated something that independently of the A.I. would be protected. What the A.I. adds to it really doesn't necessarily change the nature of the work itself. To sum it up, it's always going to be I think, a matter of degree. First of all, whether copyright exists at all in a word that's generated by the A.I. Then second of all, whether you as a user of the A.I. can assert that you own the copyright in that work.

Lisa: This is an interesting perspective. Typically, authors rely on copyright to earn money on their works, it's theirs, and they can charge for it. For the A.I. works, it's not so clear, obviously. Some A.I. apps are free to use, and some are behind a paywall. Most come with some form of license conditions, and some of them, they click, click, click, and then it's you've got something created. I'm wondering, do I really need to read and agree to the license terms if it's an A.I. thing that doesn't have copyright protection anyway? What am I paying for?

Guillaume: It's a boring answer. It can be boring and it takes some time to read terms and conditions and user license agreements on the websites. I'm sure most of us almost never do it. If it's something that has value to you, as an artist, it could be both the input, what you give the A.I. that has value to you, or it could be what the A.I. generates that you think will have value to you.

If you have that perspective, then I think it becomes very important to not just read but also understand the terms of these licenses and agreements. I think that the 2 most important aspects to be on the lookout for is you want to make sure that you don't give away your rights that may already exist in the input you provide to the A.I. For example, in the music example that I mentioned earlier, if I compose a piece of music, and I perform it and I record it, I already own rights in this. When I provide it to the A.I., I want to make sure my granting and unlimited license to the A.I. itself and to everyone who then uses the A.I. to use that piece of work, maybe I don't want that.

Likewise, once I get the output from the A.I., am I allowed to use it, or does the A.I. license include terms such as I'm not authorized to use it for commercial purposes, for example? Or I am authorized to use it, but it's a non-exclusive license, for example, and anyone else is also authorized to use it. I think people have to be very careful about these types of terms. Maybe one last point, because you mentioned, should I read this even if it's not protected by copyright?

It's possible that the A.I. will generate something that's not protected by copyright for the reasons I've mentioned earlier, but the license terms can still create a contractual obligation for you not to use it for commercial reasons, for example, and that could expose you to liability if you end up breaching that license. Even in cases where you're not sure that it's protected by copyright, you want to make sure that you understand these license terms. Basically, if the situation is important enough for you to care about what you're putting in and getting out of the A.I., then it's important enough for you to read and understand these terms.

Lisa: You mentioned that we should be careful and think about what the value is to me as I'm plugging something into an A.I. Now, let's say I do pay and I understand the license terms, and so on, and I want an A.I. to suggest a really cool brand name, for example, or a slogan, or a design, or something for any product that I've got, is it a good idea to proceed with the trademark that an A.I. can suggest to me?

Guillaume: As a starting point, I think it's important to clarify that trademark law is very different from copyright law. Copyright protects an author's right in something that is, let's use the word "artistic", that the author creates. I create a piece of music or I create a painting, I have the right as the author to exploit it commercially as mine. A trademark has a different purpose. The purpose of a trademark is to protect a business's right to be recognized as the source of a product. Then mirror, I guess, perspective from this is that it ensures that a customer who purchases a product or a service knows where this is coming from.

To give an example, if me as a consumer, I buy a product that is branded ABC, I know that it comes from the company ABC behind the product. That's the purpose of a trademark. This leads to different considerations. Now maybe just as an aside, designs and logos could be both trademarks and protected by copyright. For designs and logos, the copyright aspect, my answer would be the same as what I've already mentioned in previous answers, but for trademarks more generally, the considerations are different. First of all, to be protected as a trademark, a word or a slogan doesn't have to be original in the same sense as for copyright.

You have plenty of trademarks who are even just standard normal words, like Apple, for example, Apple is not an original word. It's not protected for that reason. It's protected because it has become distinctive of the company that is behind these products and services. If the A.I. generates a brand name for you, doesn't have to be original to be protectable, but the fact that it is potentially original also does not mean that it's automatically protectable. There is no right contrary to copyright that is generated from the second or from the moment a trademark is created.

In order to generate rights, you have to go out and use it, and you have to make it become distinctive of your business basically. It's subject to acquiring rights through registrations, for example. The danger is not as much as at the step of creating the brand, the danger is at the step of using it on the market, because you don't know at that point if that trademark that was generated by the A.I. for you is already used by a third party or maybe registered on the trademarks register.

If you go out and use that A.I.-generated brand, and it's someone else already has rights in either that brand or something that's confusingly similar, then you're infringing that other person's right. You're infringing by reason as you using the trademark, not the trademark being created in the first place. You would be liable, not the AI. Intellectual property lawyers and trademark agents have various tools to review the market and the trademarks register, and create a landscape of similar marks to yours to the one you intend to use.

They're equipped to reach a conclusion as to whether you would be free to use that mark or not on the market. It will sound like a self-serving answer but I think ultimately if you generate a new brand name or a new slogan, or a new general trademark either by yourself or through the use of an A.I., it's always a good idea to validate it with a trademarks lawyer or trademarks agent to make sure you will actually be able to use it on the market without liability to you and your company.

Lisa: If we step back to the creations again, I was thinking if I'm an artist, and someone shows up with a, say, a painting or something that they say that an A.I. has created, it looks like mine, who do I sue?

Guillaume: Yes, there's that possibility. Then there's reverse also. What if I use an A.I. and it ends up looking like something someone else already has rights on from a copyright perspective? It's more or less the same question just seen from a different perspective. To understand the answer, obviously the answer is always going to be it depends on specific facts, but as a matter of context. Copyright infringement arises when a work that's created reproduces all or a substantial part of an already existing work. There are 2 requirements for copyright infringement.

The first one is the infringer, in your case, for example, the A.I. has to have access to the original the first work. Then the infringer needs to reproduce a substantial part of that first work. Then you get copyright infringement. Without access to the original work, it's basically independent creation, and that's generally not going to be copyright infringement. One interesting aspect of all of this in the A.I. context is that, by their own nature, A.I.s learn to produce outputs, by ingesting and analyzing an incredibly large number of precedents, whether they're text or images, or music.

The A.I., therefore, may or may not, we never really know, but it's possible that it had access to your original work, and used it in a way to create this new work that the question becomes, is it infringing copyright? If we look at it on a spectrum, and I think the focus here should—I'll leave aside the question of whether it's taking a substantial part or not because that's the subjective portion of the test. I think the critical portion of the test, to answer that question, is going to be access to the work. At the one end of the spectrum, the A.I. did not have access to the original, the first work.

In that case, no matter how close it looks, it's not going to be infringement. At the other end of the spectrum, me as the A.I. user, I could give the A.I. as an input, a word that already exists. To give you an example, I can take a, let's say, a picture of the skyline that already exists that is protected by copyright, I give it to the A.I. and I say, make this picture look like the city is attacked by aliens or monsters. It's going to generate a new picture, but now the A.I. and myself, we've had access to the original work, and if that new work generated by the A.I. is very similar to the original picture, then there could be a liability for copyright infringement.

Then the question becomes who's liable? I think that an A.I. as a piece of hardware or software would not be liable, just because I think the Copyright Act uses the word "person" when speaking of someone who infringes a copyright, and I would argue that an A.I. is not a person. The other potential liability could be the creator of the A.I., but then the creator of the A.I. did not really direct the A.I. to generate the infringing work. Probably didn't give it enough direction to do that, and so probably the creator of the A.I. would have some defense. The user of the A.I. in the example that I gave would be more at risk.

Beyond all of this, we also have to look at what happens after the use of the A.I., because if the output of the A.I. is infringing copyright, and then I publish it or I reproduce it, or I make it available on my website, all of these could trigger additional acts of infringement in which case me as the user of not the A.I. but the user of the output of the A.I., I could become liable. If I have to put it as simply as possible, because I understand it's quite complicated, as the user of an A.I., I have to be more concerned about what I provide the A.I. If what I provide the A.I. is protected by copyright and then I use the output, I'm more at risk.

As the owner of the existing work, as the original creator, and then you see something created by an A.I. in the wild that looks like your work, it becomes more difficult to find out who is responsible because you have to figure out how the work was generated, did the A.I. or the user have access to your original work? It becomes quite messy. That's unfortunately the best answer I can give in our existing copyright scheme. This, obviously like I said at the beginning, will likely evolve in the coming years.

Lisa: I would imagine that. If I was going to use this commercially I'd have to do some errors or emissions assessment by someone to say do you think I can use this?

Guillaume: It can easily become a nightmare for the reasons that you've mentioned. If it's something that's potentially highly valuable, I'm exposing myself to an important liability if I do infringe copyright, you have to do a lot of research and make sure everything is clean before you use it. Certainly.

Lisa: ChatGPT is everywhere now. I've seen posts where people are saying, "Oh, this is going to take over work basically," and other people saying, "No, no, no. Smart people use this for work. This is how you get better." What if someone wants to use A.I. for a legal consultation? Just for fun, I actually asked ChatGPT the following questions, and I quote, "Someone used A.I. to create something that looks like my creation. Who do I sue?"

The answer actually said, and I quote, "If someone uses artificial intelligence to create something that closely resembles your copyright work, you may have a claim for copyright infringement." The answer also tells me what I need to prove, and that I should seek the advice of an attorney, which is obviously an American term. You've seen the text. Was the answer correct?

Guillaume: The answer was correct but it was unhelpful, because when you say you may have a claim for copyright infringement, it doesn't mean much. You also may not have a claim for copyright infringement. As we've covered in previous answer, finding out if the use of an A.I. leads to copyright infringement can get pretty complicated from a legal, almost academic perspective. In some cases there will be an infringement. In others, there will clearly not be any infringement. In others, there will be infringement. There's also the issue of national laws.

Copyright law is relatively uniform but there are differences around the world. Was the answer given by the A.I. U.S.-centric? You've mentioned the use of sun language that suggests that it could be. Would it apply to Canada? Would it apply in Europe? In various Asian countries? It's very, very difficult to arrive at an answer that will generally be correct everywhere. I think ultimately a tool like this is helpful to get a general idea of what the answer may be, and educate the user, but it's not legal advice and it shouldn't be treated as legal advice.

I've tested ChatGPT before the interview with various similar legal questions. Sometimes it's correct, sometimes it's correct but not helpful like in this case. Sometimes it's clearly incorrect. Unfortunately, I get a sense that the reason someone would give these prompts to the A.I. is because they don't know the answer. Like the test that I've made, I don't want to be alarming, but it can be "dangerous." As with all types of information if the person is not going to consult with a lawyer then they should at the very least consult more than one source, and certainly more than just an A.I. chatbot to get the correct answer to their question.

Lisa: People don't necessarily want to approach a lawyer. They feel that this is expensive, I don't think I can afford this. We get a lot of people that say I'll just write a cease and desist letter, I'll send it out. Just for fun, I asked ChatGPT to write me a cease and desist letter, and then it gave me a sample letter with some of the key statements like, "I am writing to demand that you immediately cease and desist from using," and then whatever, to copyrighted work or trademark in any matter. Then it was along with the advice to consult a lawyer before sending anything. You saw that text also. What are your comments on this letter and the overall idea of getting an A.I.-generated sample for some kind of legal action?

Guillaume: The letter contained language that you would normally find in some cease and desist letters, and it was structured generally in that way. I don't necessarily have a problem with the letter as a template. I do have some issues though with someone using this type of tool as the substitute for a specialized lawyer in the area in question. In your example, you mentioned copyrighted work or trademark, but someone could ask an A.I. tool like this to generate a cease and desist letter in any field of the law, and I'm sure there will be an output.

I particularly liked the warning that said, as part of the longer answer, you should always consult with a lawyer before sending a cease and desist letter, as a poorly worded letter may cause more harm than good, because that can often be true. You send a cease and desist letter that's relatively aggressively written like that one, and you never know how the other side is going to react. They could, for example, react by launching an action against you for a declaration that they're not infringing your right. In various jurisdictions, including in Canada, you can do that.

Now you've sent a letter hoping to solve a problem, and you've created another problem that's orders of magnitude potentially larger than the one you tried to fix. Another challenge is, you've sent your cease and desist letter and then they respond to you from the other side, what do you do about this? What do you do about the response? Do you know how to negotiate to obtain what you're trying to seek? Are you going to make admissions to the other side in your discussions that you shouldn't be doing? Sending a cease and desist letter is one part of a larger process, and preparing a cease and desist letter is even a smaller part of all of this.

It's not incorrect what the A.I. generated, but it's also not totally helpful. The last consideration that I think is important to point out, and I would finish with this, is when you are asking the A.I. to generate a cease and desist letter, presumably you would provide it with the specific facts of your case. Let's say, I own these rights, the other side's name is this, they live at this address, they've performed this type of task, and here are my demands, and then the A.I. will generate the letter for you. When you do this with a lawyer, the information that you give your lawyer will be privileged.

The information will remain confidential, and that's what it means. Privileged information means it will never get out of your lawyer's office, whatever you tell them. If you give the same information to an A.I., that information is not privileged. A.I. is not a lawyer. They don't have the same benefits protections under the law, and it's possible that without even knowing, you will tell the A.I. facts that can hurt your case if it ever comes to a point where, for example, you have to litigate your case in court.

That information is not confidential. The other side could force ChatGPT to provide them with whatever prompt you gave. Now this is obviously quite an extreme example, but I think a good rule of thumb with A.I. is just to never give them any confidential information, any private information, any financial information, because you're talking with a machine that's learning by using what you give it, and you never know where the information you give it will end up. That would be my, I guess, closing warning in answer to this question.

Lisa: We have talked a lot about copyrights. Most people know that Canada is a signatory to the Berne Convention, which essentially means that our things are copyright-protected by law automatically. The interpretation of these laws, especially when it comes to A.I., is going to be what, an international or at a national level?

Guillaume: It will be at a national level. It already is. Even if Canada is a treaty country, it's part of various treaties regarding the protection of intellectual property. For these to have effects nationally, Canada has to pass national laws that have their own specific frameworks so that Canada's Copyright Act and the U.S.'s equivalent act don't protect rights in exactly the same way even if you're looking at 2 countries that could be part of the same treaties. I think, probably not in the short term, but in the medium to long term, we'll see copyright acts evolve.

Maybe different types of rights be added, not necessarily under copyright, but there could be a completely new right that pops up after discussions between countries or even within a single country to protect A.I.-generated works. Maybe the conclusion in certain countries will be that it's all protected by copyright for this and that reason, but we want to protect it, and here's the framework. This will often happen on a country-by-country level, but also possible that new treaties are signed specifically for that issue.

Lisa: Guillaume, it's been such an interesting conversation. It is a field that is evolving. Maybe we'll cross paths again, but for now, I thank you so much for sharing your insights and your expertise on this topic. Thank you.

Guillaume: It's been my pleasure. Thank you for having me.


Lisa: You're listening to Canadian I.P. Voices where we talk intellectual property. In this episode, you met with Guillaume Lavoie Ste-Marie, principal at Smart & Biggar Montreal to talk about intellectual property protection for works created by artificial intelligence, and if using A.I. to answer I.P. questions is a good idea. Guillaume highlighted the importance of staying informed when using A.I.

There are many aspects that A.I. users should consider before working with an A.I., from reading the license term to access the A.I., to considering what original or perhaps confidential information that goes into the A.I. to create the works. Guillaume also noted that while in some cases an A.I. can give a general answer to some I.P. questions, there are risks in using the A.I. answers for any kind of action without first at least verifying that the answer from the A.I. machine is correct.