1 00:00:18,680 --> 00:00:24,000 Amy Qi: Hi, I'm Amy Qi from McGill University and with me is Jay Kerr-Wilson, 2 00:00:24,040 --> 00:00:28,200 a partner in Fasken's Ottawa office and head of the firm's copyright practice. 3 00:00:28,600 --> 00:00:32,220 Welcome to the third episode in our series, "Perspectives - 4 00:00:32,220 --> 00:00:39,200 AI and Copyright", exploring the legal challenges posed by adoption of generative artificial intelligence systems. 5 00:00:39,920 --> 00:00:46,880 In today's episode, we're exploring the tricky question of whether or not AI generated content should be protected by 6 00:00:46,880 --> 00:00:50,600 copyright and if so, who should be considered the author. 7 00:00:51,560 --> 00:00:56,280 As generative AI has become more accessible to the public with programs like Dall-E, 8 00:00:56,320 --> 00:01:02,790 ChatGPT and Midjourney, there have been a number of cases where generative AI has shown its disruptive potential in 9 00:01:02,790 --> 00:01:04,550 various industries and markets. 10 00:01:05,430 --> 00:01:12,150 With this rise in AI generated works comes the collateral impacts on copyright law and the discourse surrounding the 11 00:01:12,150 --> 00:01:14,630 copyright ability of this content. 12 00:01:15,230 --> 00:01:19,630 To start just briefly again, what is machine learning and generative AI? 13 00:01:20,110 --> 00:01:24,070 Jay Kerr-Wilson: Sure. So machine learning is a form of artificial intelligence, 14 00:01:24,070 --> 00:01:33,390 or AI, where the computer system in effect can analyze large amounts of data and can, 15 00:01:34,070 --> 00:01:43,550 uh, learn the sort of the rules that apply to that data and then predict based on that data what a particular outcome 16 00:01:43,550 --> 00:01:49,390 may be. So an example is, you know, based on everything you've watched on Netflix, 17 00:01:49,550 --> 00:01:57,230 Netflix algorithm then learns your preferences and it can recommend other titles that it thinks you might enjoy. 18 00:01:57,670 --> 00:02:05,170 Generative AI is the next iteration of this, where the system not only learns and predicts, 19 00:02:05,170 --> 00:02:09,410 but is able to generate new content based on this prediction. 20 00:02:09,690 --> 00:02:14,250 So a large language model reads millions of books. 21 00:02:14,290 --> 00:02:20,410 It then learns how books are constructed and narratives are constructed, 22 00:02:20,770 --> 00:02:27,170 and based on that, in response to a prompt, it can generate a brand new book that's not written by a 23 00:02:27,170 --> 00:02:31,890 human, but that is based on what it understands. 24 00:02:32,290 --> 00:02:35,210 The process for for writing a book is. 25 00:02:35,850 --> 00:02:43,130 Amy Qi: And just to really narrow in on this point, what are the current capabilities of generative AI? 26 00:02:44,370 --> 00:02:48,890 Jay Kerr-Wilson: Well, it seems like every day people are pushing the boundaries of what AI can do. 27 00:02:48,890 --> 00:02:56,810 But certainly AI right now, um, can generate, you know, entire novels. 28 00:02:57,280 --> 00:03:02,760 It can create photorealistic images of things that don't actually exist. 29 00:03:03,240 --> 00:03:06,240 It can create full length videos. 30 00:03:07,000 --> 00:03:13,040 Basically, you know, at this point, anything that can be created by a human and then reproduced 31 00:03:13,040 --> 00:03:21,280 in a digital form and put into a training set, there is an AI system out there that someone has figured out 32 00:03:21,280 --> 00:03:25,360 how to create new forms of that same content. 33 00:03:26,000 --> 00:03:34,520 Amy Qi: Okay, so the US Copyright Office launched an initiative in 2023 to examine various copyright law and policy issues 34 00:03:34,520 --> 00:03:41,480 raised by AI. In 2024, they published the first part of their report on Copyright and Artificial Intelligence 35 00:03:41,480 --> 00:03:46,600 focusing on digital replicas. And this year, the second part of the report addressing the 36 00:03:46,600 --> 00:03:50,320 copyrightability of AI generated material was released. 37 00:03:50,880 --> 00:03:58,900 The report seeks to analyse the "type and level of human contribution sufficient to bring these outputs within the 38 00:03:58,900 --> 00:04:01,500 scope of copyright protection in the United States". 39 00:04:02,180 --> 00:04:06,740 Can you summarise some of the key conclusions and recommendations in the report? 40 00:04:07,100 --> 00:04:15,340 Jay Kerr-Wilson: Yes, absolutely. I should caution that the second report that you referred to was actually released in in draft only. 41 00:04:15,700 --> 00:04:20,300 Um, and the register of copyright, who is the person who's in charge of the US Copyright Office 42 00:04:20,340 --> 00:04:26,020 was fired by the, uh, administration shortly after the draft was released. 43 00:04:26,020 --> 00:04:30,860 So we're not quite sure what the status of that second part of the report is. 44 00:04:30,860 --> 00:04:36,420 But, um, in terms of the content and the conclusions and recommendations, 45 00:04:36,980 --> 00:04:39,420 they felt at this point that questions about. 46 00:04:39,460 --> 00:04:44,500 And so what we're looking at is where an AI system generates new content, 47 00:04:44,500 --> 00:04:47,820 so generates a new book, generates a new video. 48 00:04:48,140 --> 00:04:53,220 What is the copyright protection that should be given to that AI generated content. 49 00:04:53,690 --> 00:04:59,370 And the recommendation, at least in draft of the report from the Copyright Office, 50 00:04:59,410 --> 00:05:08,890 was that these questions right now can be resolved under the existing laws without having to have specific legislative 51 00:05:08,890 --> 00:05:14,770 amendments. And and it goes back to sort of what is the purpose of copyright law. 52 00:05:15,170 --> 00:05:21,970 So copyright is intended to protect the original expression in a work created by a human author. 53 00:05:22,290 --> 00:05:29,690 So even if you know, a person uses AI to assist in the creation of a work, 54 00:05:29,730 --> 00:05:37,810 the purpose of copyright is to protect the human element of that creation and not necessarily the AI generated element 55 00:05:37,810 --> 00:05:46,650 of that of that content. So we've seen where the office has ruled that copyright does not extend to purely generated, 56 00:05:46,690 --> 00:05:53,950 purely AI generated material, or where there's sort of insufficient human contribution to 57 00:05:53,990 --> 00:06:00,830 the expressive elements. And, and where you have sort of humans and AI working side by 58 00:06:00,830 --> 00:06:07,070 side, that will be a case by case determination of how much creative input did the human have? 59 00:06:07,070 --> 00:06:12,030 What was the role of AI in deciding whether or not the output or the product, 60 00:06:12,030 --> 00:06:16,950 or the creative solution should be protected by copyright or not. 61 00:06:17,550 --> 00:06:21,310 Amy Qi: As outlined in the Copyright Office's report. 62 00:06:21,350 --> 00:06:26,710 Authorship is a foundational guiding principle for determining copyright protection. 63 00:06:27,110 --> 00:06:32,590 What does this legal concept of authorship entail and how does AI fit into it? 64 00:06:33,310 --> 00:06:39,630 Jay Kerr-Wilson: So authorship is sort of the foundation of all creative endeavour under the Copyright Act. 65 00:06:39,870 --> 00:06:44,150 So author is the term that that not only the US copyright system, 66 00:06:44,150 --> 00:06:47,710 the Canadian copyright system, assigns to the original creator. 67 00:06:47,830 --> 00:06:51,510 So you can think of an author of a book or an article, but, 68 00:06:51,580 --> 00:06:56,380 you know, a sculptor is the author of the sculptures that she produces. 69 00:06:56,740 --> 00:07:01,340 Uh, the photographer is the author of the photographs that they produce. 70 00:07:01,460 --> 00:07:06,980 Author is not explicitly defined under the US legislation. 71 00:07:07,260 --> 00:07:16,780 Um, but US courts have consistently not recognised copyright in works where there is no human author. 72 00:07:16,780 --> 00:07:24,420 So there has at this point been a sort of consistent line of cases that says there must be a human author identified, 73 00:07:24,700 --> 00:07:27,980 uh, in order for copyright to be extended. 74 00:07:28,420 --> 00:07:35,780 And just to quote the Supreme Court, the said the author of a copyright work is the person who 75 00:07:35,820 --> 00:07:40,620 translates an idea into a fixed, tangible expression entitled to copyright protection. 76 00:07:40,860 --> 00:07:46,540 Now, this quote sort of predates the modern generative AI systems, 77 00:07:46,540 --> 00:07:51,840 because arguably, we now have generative AI systems that are also able to translate ideas into fixed, 78 00:07:51,840 --> 00:07:58,080 tangible expression. But the idea is that it's a human being who's doing this translation. 79 00:07:58,320 --> 00:08:08,120 And, you know, certainly you can think of AI being used in ways that assist and enhance human expression without 80 00:08:08,120 --> 00:08:10,000 limiting copyright protection. 81 00:08:10,440 --> 00:08:17,120 But it's where the the entire creative process has been taken over by AI, 82 00:08:17,520 --> 00:08:22,760 where then courts and the Copyright Office have found there'd be a lack of protection. 83 00:08:22,960 --> 00:08:27,560 Amy Qi: And why is this identity of the author so central to copyright law? 84 00:08:28,080 --> 00:08:33,760 Jay Kerr-Wilson: You know, as I said, the whole creative process begins with the author. 85 00:08:33,760 --> 00:08:39,800 And and it's not just that the author has an idea, but the author has then expressed that idea in some tangible 86 00:08:39,800 --> 00:08:42,480 form. So copyright does not protect ideas. 87 00:08:42,480 --> 00:08:44,520 It protects the expression of ideas. 88 00:08:45,080 --> 00:08:50,310 Writing the book, producing the painting, making the sculpture. 89 00:08:50,870 --> 00:09:01,670 Um, and a lot of the, uh, structure and the economic activity 90 00:09:01,830 --> 00:09:09,270 that's based on the Copyright Act in Canada starts with and is dependent on an author. 91 00:09:09,630 --> 00:09:16,590 So in Canada, in most cases, the author of copyright is going to be the first owner of copyright. 92 00:09:17,710 --> 00:09:23,230 They may decide to sell that copyright to somebody and or license it, 93 00:09:23,230 --> 00:09:25,310 but they're the first owner of copyright. 94 00:09:26,430 --> 00:09:34,470 So that means that all economic transactions for the distribution and use of copyright content flows from the 95 00:09:34,470 --> 00:09:42,150 author. Also, it's important to realise that copyright is designed as a time limited protection. 96 00:09:42,630 --> 00:09:48,170 So currently in Canada's, many and many other jurisdictions in the world. 97 00:09:48,810 --> 00:09:51,330 The term of copyright runs for the life. 98 00:09:51,530 --> 00:09:54,250 Life of the author. Plus 70 years. 99 00:09:54,490 --> 00:09:57,490 So 70 years after the death of the author. 100 00:09:58,130 --> 00:10:03,850 The term of copyright expires and the work becomes part of the public domain, 101 00:10:03,850 --> 00:10:10,170 which means it's free for anybody to use without having to get permission or consent. 102 00:10:11,130 --> 00:10:18,930 Even in a situation where the author has sold her copyright to somebody else, 103 00:10:18,930 --> 00:10:27,450 who's then going to, umm, distribute it or or produce some economic value from it, 104 00:10:27,490 --> 00:10:33,530 it's still the author's lifespan plus 70 years that determines the term of copyright, 105 00:10:33,690 --> 00:10:39,330 even if the author doesn't own the copyright anymore. 106 00:10:39,730 --> 00:10:47,680 Um, so so the notion of author is very central to both sort of constructing the economic relationships around copyright, 107 00:10:47,840 --> 00:10:52,640 but also understanding when copyright protection ends in a given work. 108 00:10:53,280 --> 00:10:58,520 Amy Qi: How does this concept of authorship translate to Canadian copyright law? 109 00:10:58,560 --> 00:11:03,200 Jay Kerr-Wilson: I think it's fairly safe to assume that under Canadian law, as under US law, 110 00:11:03,440 --> 00:11:06,000 an author has to be a natural human being. 111 00:11:06,680 --> 00:11:10,200 So AI cannot be the author of a work. 112 00:11:10,600 --> 00:11:18,280 And there's some very practical reasons why this is from a policy point of view. 113 00:11:18,800 --> 00:11:25,840 Uh, makes sense. Um, because if an AI system can be an author, 114 00:11:26,160 --> 00:11:31,000 this will cause complications and disruptions into this system. 115 00:11:31,000 --> 00:11:38,480 We have built over centuries that that recognise copyright as being a human centric, 116 00:11:38,800 --> 00:11:44,040 uh, regime. So first of all, under our current legal system, 117 00:11:44,080 --> 00:11:48,030 a computer system can't enter into a binding legal agreement. 118 00:11:48,510 --> 00:11:56,670 So if the... if an AI system is the first owner of copyright in a work that's been generated by that AI system, 119 00:11:56,910 --> 00:12:02,950 there is no way for the copyright in the work to be assigned or to be licensed. 120 00:12:02,990 --> 00:12:10,550 It will, in effect, get locked without some change to the copyright or some other mechanism that says somebody other 121 00:12:10,550 --> 00:12:15,550 than the author in that circumstance would have to be the first owner of copyright. 122 00:12:16,070 --> 00:12:22,390 And second, how do you determine the lifespan of an AI system? 123 00:12:22,430 --> 00:12:30,510 So if the if an AI system is the author of a work and the term of protection is life of the author plus 70 years, 124 00:12:30,870 --> 00:12:40,190 how do we decide when the life of an AI system has ended in the 70 year clock starts running? 125 00:12:40,190 --> 00:12:47,770 So if the AI I system that generated the work is updated to a new version. 126 00:12:47,770 --> 00:12:54,410 Does that mean the old version has died and works created by that old version will then fall into the public domain in 70 127 00:12:54,410 --> 00:13:01,290 years? So, you know, these are sort of philosophical angels dancing on the head of a pin kind of questions. 128 00:13:01,290 --> 00:13:04,810 But they're central to how copyright works in Canada. 129 00:13:05,290 --> 00:13:11,170 And so we need to have some sort of resolution if we're going to move away from the concept of a human being, 130 00:13:11,170 --> 00:13:14,610 being the first owner of copyright in works. 131 00:13:14,770 --> 00:13:24,490 And conversely, if AI cannot be the legal author of and cannot be the owner of copyright, 132 00:13:24,810 --> 00:13:29,650 then arguably the work that's produced by the AI doesn't have an author, 133 00:13:29,650 --> 00:13:36,690 doesn't have an owner, and isn't protected by copyright, which also has consequences for sort of the economic value 134 00:13:36,930 --> 00:13:44,480 of AI generated work. If anyone can freely infringe those works without any legal repercussions. 135 00:13:44,840 --> 00:13:51,720 But, you know, I don't think we're necessarily stuck because I think that Canadian courts and Canadian courts have always 136 00:13:51,720 --> 00:13:58,880 taken sort of a middle ground between a European version of copyright that's very author centric and an American version 137 00:13:58,920 --> 00:14:02,120 of copyright that tends to be very sort of economic value based. 138 00:14:02,360 --> 00:14:08,840 And Canada has kind of found its own middle ground on deciding rules of copyright. 139 00:14:09,840 --> 00:14:19,680 So I think that Canadian courts could find, even if a work is largely the product of AI or generated by 140 00:14:19,720 --> 00:14:29,240 AI, the person who gave the instructions and took the steps that resulted in the creation of the work will be the 141 00:14:29,240 --> 00:14:39,040 author. So they're the one that made the creative choices that were then reflected by the AI processes to 142 00:14:39,080 --> 00:14:46,500 produce this work. Viewing it that way then we just slot AI into another technological tool, 143 00:14:46,500 --> 00:14:52,780 very sophisticated and very powerful, but, you know, we've been using computers to assist in the 144 00:14:52,820 --> 00:14:56,460 animation process for decades. 145 00:14:56,460 --> 00:15:02,820 We've used technology in a lot of different ways to assist the human creative process, 146 00:15:03,060 --> 00:15:07,780 but always recognise the creativity as a result of human choices and decisions. 147 00:15:08,220 --> 00:15:13,300 So I think the approach a Canadian court would take is, at the heart of creation, 148 00:15:13,300 --> 00:15:17,140 was there a human being making the necessary creative choices, 149 00:15:17,140 --> 00:15:21,460 and if so, that human being is then the author and will be the first owner, 150 00:15:21,500 --> 00:15:23,540 not the AI system they used. 151 00:15:24,740 --> 00:15:29,020 Amy Qi: Okay, shifting back to the US, but still related to what we're talking about, 152 00:15:29,460 --> 00:15:34,300 the US Copyright Office concluded that given the current available technology, 153 00:15:34,540 --> 00:15:41,530 prompts alone, don't provide sufficient human control to make users of an AI system the authors of the output, 154 00:15:41,570 --> 00:15:47,210 like what you're saying. Um, can you explain some of the analysis that they use to come to this conclusion? 155 00:15:47,610 --> 00:15:54,890 Jay Kerr-Wilson: Sure. And this is kind of an answer to the analysis that I suggested that may come through a Canadian court. 156 00:15:55,010 --> 00:16:04,490 So what they're saying is, you know, if you put a one paragraph prompt into ChatGPT and the 157 00:16:04,490 --> 00:16:14,170 prompt is, please write a 500 page historical fictional novel based on the life of Henry the 158 00:16:14,170 --> 00:16:19,370 Eighth, and it produces 500 pages of text in response to that prompt. 159 00:16:19,410 --> 00:16:26,250 I think with the Copyright Office is saying is, you know, you giving it that one suggestion does not 160 00:16:26,250 --> 00:16:31,410 translate into you being the author of the 500 pages. 161 00:16:31,490 --> 00:16:38,870 And I think there's a good argument to that is, you know, how much creativity and how many choices did the 162 00:16:38,910 --> 00:16:45,790 human author or the human make, versus how many choices and how much creativity was left up 163 00:16:45,790 --> 00:16:49,030 to the AI? And it's going to be different in every case. 164 00:16:49,590 --> 00:16:56,110 And I think the office is also probably rightly, looking ahead to the continued development and evolution of 165 00:16:56,110 --> 00:17:05,510 the technology, where we may see less human intervention and less of sort of the prompt 166 00:17:05,510 --> 00:17:15,470 approach to AI, where AI expression becomes more automated and 167 00:17:15,670 --> 00:17:20,870 less controlled by the system of sort of prompts. 168 00:17:21,230 --> 00:17:26,190 So, so, you know, I think that was philosophically what their concern was about simply saying, 169 00:17:26,190 --> 00:17:33,510 'Well, whoever puts the prompt in is the author' is that may not really reflect what's actually happened in the 170 00:17:33,510 --> 00:17:34,710 circumstances. 171 00:17:35,790 --> 00:17:41,780 Amy Qi: The report also includes a specific section addressing expressive inputs. 172 00:17:42,060 --> 00:17:48,180 What are expressive inputs and how do they differ from the traditional text prompts that we were just talking about? 173 00:17:48,780 --> 00:17:58,380 Jay Kerr-Wilson: I could say use my example. Give me a 500 word or 500 page novel based on the life of Henry the Eighth. 174 00:17:58,660 --> 00:18:06,140 Or I could instead give the AI five books about Henry the Eighth, 175 00:18:06,220 --> 00:18:08,660 that somebody else has written i haven't written them. 176 00:18:09,260 --> 00:18:16,260 And say using these books, then generate a play about the life of Henry the Eighth. 177 00:18:16,500 --> 00:18:25,140 So in this case, the prompts are sort of texts that somebody else has written, 178 00:18:25,140 --> 00:18:33,180 and it may be one text, it may be a number of texts, and I'm simply telling the AI system what to do with those 179 00:18:33,180 --> 00:18:43,080 texts. But I haven't really contributed much in the way of the the creative expression. 180 00:18:43,640 --> 00:18:47,800 And the same is now true with using video or images. 181 00:18:47,800 --> 00:18:57,520 So you could, you know, give a AI system a passage from a book and ask it to create 182 00:18:57,520 --> 00:19:00,200 animation based on that passage. 183 00:19:00,560 --> 00:19:09,640 So then the, the, the challenge becomes, in that case, I'm not even the author of the input. 184 00:19:09,920 --> 00:19:14,280 Arguably, if I give the AI a prompt, I'm the author of the prompt and we can decide, 185 00:19:14,280 --> 00:19:20,000 okay, well, was that prompt enough of a contribution that I'm also author of the output? 186 00:19:20,280 --> 00:19:28,760 the challenge the report is identifying is if I'm using other people's works and 187 00:19:29,640 --> 00:19:38,030 having the AI generate new work based on those other works, I can't claim any other any authorship in those expressive 188 00:19:38,030 --> 00:19:41,910 inputs because they were already authored by somebody else. 189 00:19:43,310 --> 00:19:48,470 Amy Qi: Okay. And beyond the copyrightability of AI generated works alone, 190 00:19:48,510 --> 00:19:57,790 there's an additional layer of discourse for modifying AI generated works where the content generated by AI is an 191 00:19:57,790 --> 00:20:03,630 intermediary step. For example, someone can generate an image using AI and then alter it 192 00:20:03,630 --> 00:20:07,870 with their own original art by painting over it or adding text. 193 00:20:08,230 --> 00:20:11,310 So how does copyright work in these cases? 194 00:20:11,750 --> 00:20:21,190 Jay Kerr-Wilson: So again, it's the heart of all of these questions is who has contributed the creativity to the end 195 00:20:21,190 --> 00:20:26,030 result. So if you take a bunch of random AI generated images, 196 00:20:26,030 --> 00:20:31,630 but then you impose your own human creativity and you manipulate those images and come up with something 197 00:20:31,630 --> 00:20:35,500 completely new. So it's not just copies of what AI has done, 198 00:20:35,500 --> 00:20:44,140 but there's a new element of human expression in the work you have created based on the AI one. 199 00:20:44,180 --> 00:20:48,620 I think the office, the Copyright office, is recognising that, 200 00:20:48,660 --> 00:20:57,700 you know, you are going to be the author of the copyright in this new manipulated work, 201 00:20:57,900 --> 00:21:00,100 even if you weren't authors of the input. 202 00:21:00,140 --> 00:21:02,620 And there's non AI analogues to this. 203 00:21:02,620 --> 00:21:09,180 So you know, we're very familiar that people make artwork all the time by clipping pictures, 204 00:21:09,220 --> 00:21:11,860 pictures out of magazines or clipping texts. 205 00:21:12,180 --> 00:21:19,900 And they rearrange these pre-existing clipped photos or texts to make some new artwork. 206 00:21:20,420 --> 00:21:23,740 And you know, they're going to be the author of the new artwork. 207 00:21:24,100 --> 00:21:28,860 It's a separate question of whether they've infringed copyright in the source material, but certainly they're the 208 00:21:28,860 --> 00:21:32,080 author of this new transformed thing. 209 00:21:32,200 --> 00:21:40,640 So I think what the office is saying is it's not the use of AI that's going to disqualify something from protection. 210 00:21:41,320 --> 00:21:49,600 It's going to be whether the end result is primarily the product of AI creativity, 211 00:21:49,600 --> 00:21:55,120 in which case it won't be protected, or human creativity, even if it's human creativity that 212 00:21:55,120 --> 00:21:57,280 manipulates AI generated content. 213 00:21:57,400 --> 00:21:59,520 And then I think that there would be protection. 214 00:22:00,360 --> 00:22:05,000 Amy Qi: And then just circling back, I know we've touched on this briefly already, 215 00:22:05,000 --> 00:22:09,360 but could similar analysis and legal frameworks be used in Canada? 216 00:22:09,760 --> 00:22:15,320 Jay Kerr-Wilson: Yeah. So it may not always come to the same result, but I think that the the analysis would be very similar and 217 00:22:15,320 --> 00:22:21,160 it will be case by case. So there's not going to be one rule that applies to everything. 218 00:22:21,160 --> 00:22:29,280 So the, the rule that the or the test that the Supreme Court has developed in deciding whether or not something is 219 00:22:29,280 --> 00:22:34,550 original. So so in order for a work to be protected by copyright, 220 00:22:34,550 --> 00:22:40,670 it has to be original. And the court has said in order for a work to be original, 221 00:22:41,150 --> 00:22:46,150 there has to be some element of creative decision making. 222 00:22:46,150 --> 00:22:51,710 There has to be some creative choice that's made when the work is produced. 223 00:22:52,110 --> 00:22:58,430 So, you know, you can see imagine a spectrum where I say, hey, 224 00:22:58,470 --> 00:23:03,030 ChatGPT, write me a new poem, and I don't give it a subject matter, 225 00:23:03,030 --> 00:23:08,310 and I don't give it a style, and it will produce a poem in response to that prompt. 226 00:23:08,670 --> 00:23:15,070 Um, and you can see a Canadian court saying, you know, you're not the author of that poem, 227 00:23:16,070 --> 00:23:23,350 and that poem is not protected by copyright, based on the analysis that the US Copyright Office has said 228 00:23:23,350 --> 00:23:27,590 is there simply wasn't enough, there was no creative decision making. 229 00:23:28,030 --> 00:23:34,290 However, if I say to ChatGPT, please compose a haiku. 230 00:23:35,250 --> 00:23:41,450 You know about the Swiss Alps in the springtime. 231 00:23:41,730 --> 00:23:45,730 And it does. Then, arguably, I've given it enough creative choices. 232 00:23:45,730 --> 00:23:52,690 I've chosen a style. I've chosen a subject matter that maybe I am the author of that resulting haiku, 233 00:23:53,050 --> 00:23:55,530 and there should be some sort of protection. 234 00:23:55,810 --> 00:24:04,450 So I think Canadian courts would likely, under our test, maybe protect more than the very strict test 235 00:24:04,450 --> 00:24:06,970 that the US Copyright Office has expressed. 236 00:24:06,970 --> 00:24:10,130 But I think the analysis would be very much very similar. 237 00:24:10,410 --> 00:24:16,810 Amy Qi: Okay. And speaking of case by case, there are a few specific cases where these different 238 00:24:16,810 --> 00:24:21,570 scenarios have come into play and informed the analysis that we've just discussed. 239 00:24:22,010 --> 00:24:27,880 To illustrate these legal frameworks in practice, I think it's helpful to briefly review some of them. 240 00:24:28,120 --> 00:24:34,720 The first is from December of 2023, where upon review, the US Copyright Office refused to 241 00:24:34,720 --> 00:24:37,600 register a work titled tourist. 242 00:24:38,240 --> 00:24:40,080 What was this decision about? 243 00:24:40,160 --> 00:24:45,480 Jay Kerr-Wilson: So this was one where, um, so the subject matter was artwork. 244 00:24:45,480 --> 00:24:51,000 So it was an AI generated that was supposed to resemble a painting, 245 00:24:51,000 --> 00:24:52,640 but it wasn't physically painted. 246 00:24:52,640 --> 00:24:55,080 It was always a digital image generated by AI. 247 00:24:55,800 --> 00:25:02,120 So in this case, the the artist took a photograph, put it into the AI system, 248 00:25:03,280 --> 00:25:13,240 and then asked the AI system to, in effect, apply Van Gogh's Starry Night as a style. 249 00:25:13,760 --> 00:25:18,800 And then the output was a version of the photograph that had been transformed, 250 00:25:20,040 --> 00:25:26,280 being inspired by Van Gogh's Starry Night, and in that case the Copyright Office decided this could not 251 00:25:26,340 --> 00:25:35,100 be protected by copyright, because it was the AI system that was responsible for determining how to 252 00:25:36,260 --> 00:25:43,140 merge the source of the source material and the style, and deciding what are the elements of Van Gogh's Starry 253 00:25:43,180 --> 00:25:46,540 Night that should be incorporated, and how should those be incorporated? 254 00:25:46,980 --> 00:25:56,140 So they found that there was simply no way to distinguish the elements of AI creativity from human 255 00:25:56,140 --> 00:25:59,060 authorship in the final output. 256 00:25:59,060 --> 00:26:05,100 So under their test, that requires sort of really identifiable human authorship, 257 00:26:05,100 --> 00:26:06,540 there could be no protection. 258 00:26:07,380 --> 00:26:13,140 Amy Qi: That's really interesting. Um, there's also another one where there well, 259 00:26:13,180 --> 00:26:17,620 multiple other ones where there's been different attempts to copyright graphic novels, 260 00:26:17,900 --> 00:26:23,700 which utilise a combination of artwork generated by AI and human authored text. 261 00:26:24,060 --> 00:26:28,490 One of them is this graphic novel called 'Zarya and the Dawn'. 262 00:26:29,050 --> 00:26:32,370 How was the legal analysis applied for this? 263 00:26:32,730 --> 00:26:39,530 Jay Kerr-Wilson: So very consistently? Um, so in this case, the Copyright Office determined that the images had been 264 00:26:39,530 --> 00:26:45,250 generated by AI without sufficient human authorship. 265 00:26:45,250 --> 00:26:50,450 So the images that had been generated by AI were not protected by copyright. 266 00:26:50,770 --> 00:26:56,450 But the text of the graphic novel had been written by a human. 267 00:26:57,010 --> 00:27:02,810 Um, so in that case, the the Copyright Office, in effect unscrambled the omelette and said, 268 00:27:02,850 --> 00:27:06,330 okay, so the images aren't protected by copyright, but the text is. 269 00:27:06,850 --> 00:27:12,650 Um, and so it sort of applied the test separately to the two elements of a graphic novel. 270 00:27:13,410 --> 00:27:14,690 And that was the conclusion. 271 00:27:15,410 --> 00:27:21,090 Amy Qi: All right. And what are the impacts of these decisions and where do we go from here? 272 00:27:21,570 --> 00:27:30,750 Jay Kerr-Wilson: So I think we're into a really interesting phase where we are going to continually challenge this idea of 273 00:27:31,550 --> 00:27:36,350 what is human authorship and where is the line between, you know, 274 00:27:36,390 --> 00:27:39,550 AI generated images and human images. 275 00:27:39,870 --> 00:27:47,910 Because, you know, we're probably not that far away where people can use neural networks to, 276 00:27:48,630 --> 00:27:58,350 in effect, think of an idea that will then automatically be translated by an AI into some sort of expression. 277 00:27:58,510 --> 00:28:02,950 So if you have a human who has the idea, but the entire expression of the idea is AI, 278 00:28:03,550 --> 00:28:08,550 what does that look like? I think this is going to challenge the whole notion of derivative works, 279 00:28:08,550 --> 00:28:14,910 where we have human generated content that is then adapted. 280 00:28:14,910 --> 00:28:21,430 So we have a book that's adapted into a movie by AI, and then that movie is re-adapted into a book, 281 00:28:21,470 --> 00:28:28,420 like, I think there's so much potential for people to experiment and to play and to push boundaries, 282 00:28:28,460 --> 00:28:36,020 that there's going to be just non-stop challenges to our the notion of of copyright and protecting human expression. 283 00:28:36,060 --> 00:28:43,300 Ultimately, some of these questions are going to be decided in litigation around the question of authorship and 284 00:28:43,300 --> 00:28:49,540 ownership. Uh, some of it's probably going to require some element of legislative amendment. 285 00:28:49,540 --> 00:28:58,940 We may have to define author, uh, and, and define the limits of the use of AI 286 00:28:58,980 --> 00:29:03,340 technology. But, you know, none of these questions are going to be easy. 287 00:29:03,340 --> 00:29:11,900 And the problem is the technology is advancing so quickly relative to the ability of governments to react is, 288 00:29:11,940 --> 00:29:14,100 you know, I don't think we're ever going to catch up. 289 00:29:14,100 --> 00:29:20,580 I think we're into a stage now where the creative process is always going to be, 290 00:29:21,280 --> 00:29:27,920 uh, have tension with the, the legal structures that are meant to create that or to protect that process. 291 00:29:30,800 --> 00:29:33,040 Amy Qi: Wow, what a great answer to end on. 292 00:29:33,200 --> 00:29:40,720 Thank you, Jay, once again for your insight on the copyrightability of AI generated content and the concept of 293 00:29:40,720 --> 00:29:45,720 authorship. As we've seen, there's a lot to navigate, and I think you've really distilled down the Copyright 294 00:29:45,760 --> 00:29:49,320 Office's report and the two cases that we've seen. 295 00:29:49,760 --> 00:29:55,600 So thank you for listening to this episode of perspectives, AI and copyright.