As year-end approaches, I like to look back at the main themes that emerged over the past 12 months affecting copyright, creators and the content industries, drawing from my blog posts throughout the year. Some of the same issues I highlighted last year, such as payment to news content providers by online news aggregators (Google) and social media platforms (Facebook) for use of news content, and the education fair dealing exception that has decimated the educational publishing market in Canada, are still very current—and unresolved. The biggest new issue that emerged in 2022 is generative AI, presenting considerable challenges from both a copyright and ethical perspective while offering new applications to assist in creative writing, composing music and producing art. Authors and publishers also continued to wrestle with the issue of digital licensing and the end-run on copyright mounted by proponents of “Controlled Digital Lending”. Finally, progress was made on one particular copyright-related issue, implementation of copyright term extension in Canada, bringing the duration of copyright protection into line with that of the US, EU, UK and Australia, among others, along with a commitment by New Zealand to do so in future.
Generative AI and Copyright
The big story of 2022 was the emergence of generative AI, especially AI generated art. It was only in July of this year that OpenAI’s DALL-E text to image generator was made available to the general public in a beta version. Then on August 22, Stability AI’s tool, Stable Diffusion, was released to the public, at no cost, with no barriers and, significantly, no content controls. I was a little slow off the mark (after all, it was late summer, and I was at the cottage) but I got around to writing about this phenomenon in October (“AI and Computer-Generated Art: Its Impact on Artists and Copyright”). This technology raises several questions from a copyright perspective, not to mention a range of ethical issues arising from the potential for abuse from misuse of AI generated images.
The copyright related concerns come from both the inputs and the outputs of the technology. On the input side, the models are based on the ingestion of literally millions of examples scraped from the internet by internet-indexing projects such as Laion, a research-based “non-profit” based in Germany. Many of the ingested images are works protected by copyright–designs, paintings, photographs, just about anything available online–whether watermarked and protected or not and all without permission from the rights-holder. While Laion claims to be a non-profit motivated solely by the “democratization” of machine learning, the fact that it sells its data to outfits like Stability AI (a process often referred to as “data laundering”) has raised questions as to its degree of altruism.
Whether this form of ingestion is infringing from a copyright perspective remains to be seen. Its proponents in the US claim it falls within fair use parameters, even though the copying is unlicensed while licensed databases exist, and despite the fact that in some cases the results of the AI generated product act as substitutes (or have the potential to act as substitutes) for the original work, in effect displacing the original. In other countries, the legality of this kind of text and data mining (TDM) is questionable. The UK is currently floating a process to open up TDM and create a broad exception allowing its use for any purpose, while denying rights-holders the ability to opt out or license their works. I wrote about this dangerous proposal (“Britain’s Proposed Approach to Text and Data Mining (TDM) for AI: How Not to do It (A Lesson for Canada and Others)”, highlighting it as a negative example for Canada and others countries contemplating legislating a TDM exemption. It has been vigorously opposed by British cultural entities and associations. While arguments can be made for a limited TDM exception for non-commercial research purposes where licensed alternatives do not exist, the kind of TDM exception proposed by Britain would throw out the baby with the bathwater, sacrificing the interests of rights-holders on the supposed altar of AI “innovation”. Just about the only country that has implemented this kind of broad TDM exception is Singapore—and it is probably fair to say that Singapore has a lot less invested in its cultural industries than the UK, or Canada, or many others. In the face of widespread pushback, the UK government is reportedly reviewing the proposal and may go back to the drawing board. This has not stopped some proponents of wider exceptions and weaker copyright protection from advocating for a new international treaty that would allow for broad TDM exceptions. This will be an issue to follow in 2023.
Quite apart from copyright questions relating to the inputs used by the AI application, there are all sorts of questions related to the outputs, i.e. the content produced by the technology. Are the AI-generated works subject to copyright protection if they are a result of upstream infringement? Are they sufficiently original to qualify? What if there is no meaningful human intervention in their creation? All these issues remain to be worked out.
Finally, quite apart from copyright considerations, there can be serious ethical issues related to AI generated content. Unless carefully moderated, the technology can and is being used by some to create abusive images, some sexual, some violent, or to misappropriate a person’s image or publicity right. There are already many forms of online abuse, and generative AI adds to the possibilities. As I wrote last week, “AI Generated Art: Another “Technical Breakthrough” Calling Out for Responsible Management and Regulatory Oversight”, if industry won’t self-regulate (and it won’t), then governments will have to do it. (“Grappling with Online Safety Legislation: How to Hold the Platforms Accountable”)
Another challenge faced by rights-holders in 2022, in this case authors and publishers, was the challenge to e-book licensing mounted in several US states, along with the issue of “Controlled Digital Lending”. In February I wrote about a new law passed by the Maryland state legislature imposing a compulsory license requiring publishers to license e-books to public libraries at a price to be determined by the government. (“Books, e-Books, Authors, Publishers and Libraries: A Complex Relationship”). Apart from being a violation of the US Constitution, which reserves copyright legislation to the US Congress, the Maryland law was an egregious seizure of property. The Association of American Publishers sought an injunction to halt implementation of the legislation, relief that was ultimately granted. That has not put the issue to bed, however. Copyright opponents, sadly including some in the library sector, are trying to extend the “first sale doctrine” (referred to as “exhaustion” in Canada) to digital copies of works.
Under a first sale doctrine, once a work is purchased it can be used in whatever way the owner wishes without further reference to the rights holder (except for copying). The best example is a hard copy book which can be resold or loaned without reference to the rights-holder, although it cannot be reproduced except in accordance with fair use or fair dealing limitations or under licence. Licensed digital editions of works are different from works that are sold. They are licensed for a specific time or for a specified use under certain conditions. They are not “sold”, and the licensor (publisher) retains certain rights as to the use of the licensed work. In the case of libraries, most publishers make available under licence a predetermined number of digital copies that may be loaned for a set number of times. From a publishers’ perspective, there are good reasons for these limitations; for example, the fact that a digital copy never wears out and thus never has to be replaced, as well as the elimination of the “friction” associated with physical borrowing. This “friction”, the nuisance elements associated with having to collect and return a book, is one of the factors that lead many people to purchase rather than to borrow. An e-book is a different product from a physical book, with different market characteristics. The producer of that work, the publisher, is entitled to determine the terms on which it is made available to the public. And this is where “Controlled Digital Lending” (CDL) enters the picture.
Controlled Digital Lending
CDL is an unproven theory interpreting copyright in such a way that libraries are free to make and lend a digital copy of a book as long as they have legally purchased and retain a physical copy of the same work. It is very popular in the library community because it allows these institutions to bypass the e-book licensing process. On its surface, the theory can seem plausible (only one digital copy can be in circulation for each physical copy, which is retained in storage, and a digital copy has to be “returned” before it can be reloaned.) However, an “inconvenient fact” is that for CDL to take place, an unauthorized digital copy of a work has to be made, a copy which can then compete with the authorized, licensed, digital version of the work in the form of an e-book. Although digitization of works is authorized for libraries, museums and archives under copyright laws in most countries for specific purposes, such as historical preservation, obsolescence of technology, etc, it is not okay to simply do an end run on a licensed product. In fact, in Canada the copyright exception for preservation is explicitly ruled out if an appropriate copy is commercially available. A lot has been written by academics to justify the CDL theory, but it has never been tested at law. This is about to change in 2023 with Hachette v Internet Archive (IA).
Hachette is the lead plaintiff along with several other publishing houses, supported by the American Association of Publishers, in a case against the Internet Archive (IA) regarding Controlled Digital Lending. The origin of the case lies with the decision of the Internet Archive to lift all CDL controls on its online Open Library during the initial phases of the COVID pandemic in early 2020 in order to create a so-called “National Emergency Library”. The Archive, which has as its core business the digitization of documents and books as well as archiving websites and television programs, brands itself as a “non-profit library”, although it is not really a library at all. In effect, in 2020 the IA unilaterally made available thousands of copyrighted works without authorization, discarding even the fig-leaf of Controlled Digital Lending. Not surprisingly the publishers sued for infringement. The IA then wound up its Emergency Library, but the suit has proceeded and is now before the Southern District Court of New York. The Internet Archive has established a branch in Canada. A number of Canadian university research libraries subscribe to the theory of CDL and have made their collections available to the IA for digitization. It will be interesting to see how this issue plays out in 2023 in both the US and Canada.
Copyright in Canada: Lots of Unfinished Business
On the copyright and content front in Canada, there is lots of unfinished business. In my first blog post of this year (“What Lies Ahead for Canada in 2022 from a Copyright and Content Perspective?”) I outlined the considerable “to do” list; Copyright Act amendments; Broadcasting Act amendments to incorporate streaming content (the Online Streaming Act); payment for unlicensed use of news content by internet platforms (the Online News Act) and digital safety legislation (the Online Harms Act). Of this list, only one subset was actually achieved, an amendment to the Copyright Act to extend Canada’s term of copyright protection. No other action has been taken to date to update the Copyright Act despite a commitment in the 2022 budget speech to do so. (“Copyright References in the Budget: Good Intentions Are Welcome but Early Action is Needed”). As a result of this lack of action, a coalition of 25 author, publisher and copyright groups held a “Day of Action” in late November to urge the government to follow through on its commitment to “ensure a sustainable educational publishing industry, including fair remuneration for creators and copyright holders”. (“The “I Value Canadian Stories” Campaign: Time to Get Serious About Copyright Reform”).
As for the Online Streaming Act (Bill C-11), Online News Act (Bill C-18) and the Online Harms legislation (not yet introduced), they are still works in progress. The public discussion paper for the Harms legislation attracted plenty of negative commentary, leaving the government to go back to the drawing board to take account of the input. It should be introduced in 2023 while C-11 and C-18 should complete the Parliamentary process next year. (C-11/Online Streaming has cleared the House of Commons and is currently being reviewed in the Senate, with the Committee process now complete, while C-18/Online News has completed Committee review in the House and passed third reading just prior to the Christmas break. It will now be reviewed in Committee by the Senate). All have controversial aspects and will be watched closely by copyright stakeholders in 2023. It promises to be an interesting year on a number of fronts–AI generated content, Controlled Digital Lending, and copyright updating and related legislation in Canada. Stay tuned.
This article was first published on Hugh Stephens Blog