Last year, a group of Canadian media companies launched a lawsuit against Open AI. Here's a bit of the joint statement they released at the time. Open AAI regularly breaches copyright and online terms of use by scraping large swats of content from Canadian media to help develop its products such as Chat GBT. Open AAI is capitalizing and profiting from the use of this content without getting permission or compensating content owners. Now, an Ontario court has decided the case can proceed here. I asked Kevin Cu, an intellectual property lawyer with Gilbert LLP, how this case is shaping up and how worried Open AI should be about the multiple cases that have been filed against it.

Kevin, welcome to the rundown. How are you doing? >> Good. Thanks for having me. >> Yeah, no worries. Now before we get started, I want to mention that uh and to our viewers that TBO is not participating in this lawsuit that we are talking about and you Kevin are not involved in the case. Correct. >> That's correct. I am a lawyer but I don't represent any of the parties um in the lawsuit.

All right. Well, let's talk about the case. Uh the plaintiffs in the copyright case against Open AI in Ontario include the Canadian Press, Tourar, the Glob and Mail, Post Media, and CBC. What are their main complaints? So this is a conglomeration of I think most of Canada's biggest news publishers and their complaints are um pretty broad but largely they focus on um OpenAI's use of various news articles that they've scraped from um online. So as you know you know you can access CBC or the Toronto Star and get some news articles. um some of them are paywalled and you can access those articles and their complaint is that OpenAI has um copied, downloaded, scraped some of these articles and used it to train um their chat GPT model in order to produce the product chat GPT.

All right. OpenAI uses what's known as retrieval augmented generation. Uh and it's quite central to this case for people who don't know including myself. What does that actually mean? Yeah. So, retrieval augmented generation or RAG is the acronym that's often used. RA a um that refers to one part of the sort of output process for these large language models. So, if you've ever used chat GPT and you um you know enter some input and the chat GPT bot says something um it goes through a couple of processes. One process is is sort of taking your input um determining what it should do and then there's a process at the end called uh RA which is used to prevent or limit hallucinations. What it does is um when the output is about to be um coming out from the LLM, the large language model, the rag system usually is designed to um access a fact storage or or access the internet and fact check its own output before giving you an answer so that it can you know limit the amount of hallucination um that the uh the chatbot gives you.

Now we hear a lot about hallucinations with uh with AI technology. Is that something that happens commonly when we're talking about news articles and things that you know you could quickly search up if you went to one of their websites? >> I mean this depends on the prompt. I think most AI models try to use some version of um retrieval augmented generation to sort of limit all of that so they can you know fact check the the articles if someone asks it you know on what date did something happen and the you know the the chat GPT model will probably look up um that information before giving it to you but it doesn't really prevent all hallucination. And I think most models today still suffer some degree of that to to a greater or lesser extent depending on what you're asking it.

So here in Canada we have fair dealing uh which for people who don't know is a law that allows us to use copyrighted material without permission for certain purposes. One of them including news reporting as well. Um could open AI have its copyrighted material use that instead? >> Yeah. So fair dealing is a very nuanced subject. Um, it's often used as a defense, but it's it's called it's also been called a user's right in Canada. So, if you or I were just researching news um for our own educational purposes, then we would be able to, you know, uh buy a newspaper or access it online and we can read it and perhaps um you know, reproduce a bit of it, you know, send it back and forth for mark it up for research. Um and those are sort of allowable fair dealing purposes.

there sort of a scope um of permitted activities that users in Canada are allowed to do with copyrighted works like books and news articles. And this is a central doctrine in Canadian copyright law. Um but it has certain limits to it. There are specified purposes like education um and research where you're definitely allowed to um use the fair deal uh work in a manner that is in accordance with fair dealing subject to some limitations. But it often depends on what you're doing and how much um you're using that work for.

Help us understand what's at stake. A lot of people might be looking at this and be like, "Oh, well, there's these are some Canadian news organizations who are probably seeing uh you know, their advertising shares plummeting and and and not making enough money." So, this is maybe them kind of looking at that opportunity, but help us understand really what's at stake here with those the plaintiff and open AI. >> Yeah. So, I mean, I think there are two aspects to it, at least two aspects. There are many more than that but one is as you've identified it's it's monetary it's commercial um which you know goes back to your last question about fair dealing. One of the factors as to whe whether some use is fair or not under Canadian law um includes how commercial that is.

So is is someone like OpenAI or any other company making a lot of money off of right you know Toronto Star or CBC's news articles and you know they do charge a fee for um using some of their you know pro or plus versions of their products. So they are making some amount of money. Um and so at stake in part is who can profit off of those works. You know obviously for news organizations um the core of their business is you know selling newspapers, selling subscriptions to their website and licensing their content to various outlets. And you know as far as we know and as complained in the lawsuit, OpenAI hasn't paid any licensing fees for the use of those works. So a large aspect of this is you know should open AI pay for the use of those works and how much should they pay for the use of those works because you know some aspect of it is if you're again let's take an example if it's just me and you going to buy a newspaper you know we're paying a pretty low amount of money and we're not really profiting off of it but a large company like Open AI which is you know has millions of users how much should they pay for um the use of articles and they use many millions of articles. Um, so that's, you know, a key aspect of the case.

What happens, uh, let's say this court case goes through and and the business model of Open AI changes where they do have to pay for something. How does that impact their bottom line and their business model? >> Well, you know, I'm not the I'm not in charge of Open AI, uh, thank goodness, but um, you know, I think it it it's pretty fundamental to how they operate. And I think um if you sort of take apart and look at it at a at a more high level view for most of these companies who develop large language models, right, a part of the technological innovation that they they claim to have brought um to the forefront these days is that they've they've used or they've ingested into their AI models a large portion if not all of the internet. You know, they claim you know that's the large in large language model. Um, and if they have to pay for a significant or all of the content that they're consuming, then it really could um impact their bottom line because they, you know, they're they're certainly using a lot of data. They're they're continually scraping um information and, you know, depending on how much they have to pay to access or use that content. Um, it could really impact whether or not those models become financially viable.

We have about 30 seconds left, but I do have one more question for you. The the case we're talking about right now is Canadian media companies suing OpenAI in Ontario. Uh but in New York, there is a similar case happening right now where New York Times is suing Microsoft and Open AI for copy copyright infringement. Is there something that we can learn from this case? Do do the folks who are following this case here in Ontario uh need to keep a close eye on and what's happening with that? Yeah, I mean from an economic perspective, yes, because OpenAI's business model is going to be influenced by lawsuits all around the world, not just in Canada. But to another extent, not really, because you know, Canadian laws are very different than American laws and there are copyright defenses and doctrines that are different between the two countries.

All right. Well, Kevin, we are going to leave it there. Really appreciate the insights and of course something that we will continue to follow. Appreciate it. >> Thank you for having me.