Mass media attention following Open AI's release of ChatGPT has pushed the subject of artificial intelligence (AI) back into the limelight. Discussion has focused specifically on the use of generative AI, which is a term used to describe AI programs that have the ability to create entirely new content, rather than merely analyze something that already exists. As ChatGPT demonstrates, the content that can be created through generative AI today can be difficult to distinguish from that created by a human, and when given simple instructions, AI programs can produce digital images, videos, audio, text or code in relatively high quality, in a matter of seconds. If you don't believe us, try asking ChatGPT a question and see for yourself.
The reality is that generative AI is well on the way to becoming not just faster and cheaper, but better in some cases than what humans can create by hand. Every industry that requires humans to create original work—from social media to gaming, advertising to architecture, coding to graphic design, product design to law, marketing to sales—could be up for reinvention in varying degrees.
In this three-part series, we explore how key areas of law will influence the development of generative AI and how it is used by businesses. Part 1 focuses on copyright law and the critical questions of ownership and authorship over AI-generated content, Part 2 discusses privacy law considerations under Canada's proposed Artificial Intelligence and Data Act, and Part 3 explores issues of liability for the creation and use of AI-generated content; namely, who is accountable for AI-generated content and when.
Part 1: Generative AI and copyright law
Canadian law has yet to definitively decide how, if and when copyright laws should apply to content generated by AI. In light of the gaps in the current Copyright Act vis-à-vis AI generally, there have been two government publications containing recommendations that speak to some aspects of the AI/copyright interface: the 2019 Report of the Standing Committee on Industry, Science and Technology (often referred to as the INDU Report) was tabled following the statutorily-required five-year review of the Copyright Act1; and in 2021, the Canadian government's department of Innovation, Science and Economic Development (ISED) published a consultation paper which considered whether and how to adapt the Copyright Act in light of current AI capabilities2. The government has yet to table any proposed amendments in light of the recommendations contained in the INDU Report and the ISED paper. Despite the uncertainty, we know enough to anticipate that the primary commercial concerns and legal questions will arise in the context of the "data inputs," which are used to train the AI system, and the "data outputs," being the content generated by the AI.
1. Input issues: With training data comes risk
Generative AI relies on deep learning, a subset of machine learning, to be able to continuously improve its output. Thus, the quality, usefulness and breadth of applications of generative AI depend on the training set from which the AI learns. In order to create this training set, machine learning engineers need to source a large number of reliable and relevant data. For example, if the AI is being programmed to recognize a certain item in an image, it needs a training set comprised of a large number of images with and without that item in order for the AI to learn and be able to accurately recognize it when it is asked. One of the ways in which these large data sets are obtained is through the use of data mining and web scraping, which includes extracting various forms of data from various sources, one of which includes the internet.
The main copyright liability issue that may arise from data inputs would result from the reproduction of data consisting of, or containing, works protected by copyright being used to train an AI program, or in the process of the creation of "new" works by the AI program. Developers of AI programs must be mindful of the data collection practices they adopt to mitigate this risk, but regardless of due diligence efforts, it can be difficult to efficiently collect the mass amounts of data required to train an AI while ensuring that none of the data inputs are protected by copyright.
The INDU Report recommended (in Recommendation 23) that Canada amend the Copyright Act to "facilitate the use of [copyright-protected materials] for the purpose of informational analysis." The United Kingdom's government has indicated that it will be introducing a new exception in their legislation which would allow copyright-protected works to be mined and used as part of a training data set for an AI program; following consultation on the proposal, the UK's Intellectual Property Office published a paper which indicates that the new statutory exception will allow use of text and data mining for any purpose, including commercial, without needing to obtain permissions or pay a fee to copyright holders. This is a marked departure from the existing rules, which only allow text and data mining practices to scrap copyrighted data if it is to be used for research and non-commercial purposes3. The UK guidance explains, however, that rights holders will still have safeguards to protect their content. The main safeguard will be the requirement for lawful access. That is, rights holders can choose the platform where they make their works available for text and data mining, and through such platforms they should be able to charge data collectors for access4.
2. Output issues
Many of the copyright issues related to content outputs generated by AI remain unresolved by Canadian lawmakers – the Copyright Act simply does not address AI in explicit terms. However, the two primary issues relate to the interpretation of "authorship" and "ownership" of AI-generated works.
- Authorship
Who is the author, for copyright purposes, of a work that is entirely generated by artificial intelligence? This is a foundational challenge for AI-generated content because in order for something to be protected by copyright in Canada, it must have been created by an "author." While the Copyright Act does not define the term "author," most commentators agree that, given the wording of the legislation and axiomatic assumptions undergirding the caselaw, an author must be a human being. That obviously poses challenges for works created by an AI program. Whether or not to treat AI programs as "authors" will inevitably become an issue in Canada. In December 2021, the Canadian Intellectual Property Office (CIPO) registered, for the first time in Canadian history, a copyright for a painting co-authored by both a human being and an AI painting app5. While the precedential value of the registration is questionable (as CIPO does not review applications for registration for substantive accuracy or compliance), the registration demonstrates that copyright owners are taking the position that AI-generated works can and should be protected by copyright, and that phenomenon is unlikely to wane.
The issue has been addressed only in a limited number of other jurisdictions. In the UK for example, the author of a "computer-generated" work is "taken to be the person by whom the arrangements necessary for the creation of the work are undertaken.".6 While there is some ambiguity around the meaning of that phrase, it seems to entail that the author would be the programmer or the user of the computer in question. The matter remains unsettled in light of a paucity of caselaw interpreting the provision, though that is unlikely to remain the case for long, given contemporary generative AI capacities. In Canada, the ISED paper canvasses, without recommending, a number of potential approaches for determining authorship of AI-generated works, including adopting those used in the UK, deeming them to be "unauthored" and thus not eligible for copyright protection at all, or affording them a modified and more limited set of rights. The UK approach to "computer-generated" works, for example, entails a shorter term of protection of only 50 years (as compared to the "life plus 70 years" term applicable to human-generated works). The UK approach also denies moral rights protection to computer-generated works – something the Canadian legislation will also need to address.
Thus in Canada, settling issues of authorship remains to be addressed, and it is likely that the degree of human involvement in the generative process itself (such as programming, providing "prompts", etc.) will play a role in the ultimate settlement of the issue. In situations where AI becomes entirely autonomous and creates content of its own volition, policymakers will need to determine how to allocate authorship status, or if it makes more sense to establish a carve-out for AI-generated works so that they are deemed authorless.
b. Ownership
Following the issue of copyright authorship is the issue of copyright ownership. The issue of ownership only arises after, and typically as a function of, authorship, as the default rule is that the author is the first owner of copyright (absent an employment relationship). Any solution to the authorship question will thus inescapably have to contend with the ownership question. For example, if it is ultimately determined that there is no "author" for copyright purposes, that would mean there is also no "owner" for copyright purposes – and if the author is deemed to be the program itself, then the implications for ownership will also need to be addressed. The INDU Report highlighted the significance of this issue, recommending (in Recommendation 14) "amending the Copyright Act or introducing other legislation to provide clarity around the ownership of a computer-generated work."7 Once an answer as to initial ownership has been identified, most issues regarding title can typically be managed by contract; owners can freely assign or license their rights.
3. Practice tips
Until the Copyright Act is amended to address the issues identified above, a primary concern for users of generative AI will be what they can do with the output generated by the AI – in short, how can the content produced by an AI, be it text or an image, be used? Can you use the image created by an AI program on a t-shirt, as the cover of a book, or in a movie? Can you use the text created by an AI program in your brochure or novel?
The answers to those questions will ultimately turn on who the author and owner of the content is and what uses they have authorized. Pending any legislative changes, users of AI-generated content should, at a minimum, examine the terms of use accompanying the AI program to see if they address any of those issues and if they do not, do not proceed with use unless and until the service provider has provided some clarification (and ideally a license or transfer of ownership). Users of AI-generated content will also need to bear in mind that their use of such content might be viewed by the owners of the copyright, in the content that was ingested by the AI, as infringement of their rights in such content and so, in the absence of suitable representations and indemnities from the operator of the AI program, there could be potential exposure to infringement claims.
4. Conclusion
The following was generated by ChatGPT, with minor edits.
In conclusion, the release of Open AI's ChatGPT has brought attention to the use of generative AI in business, specifically the question of copyright law and how to attribute authorship and ownership to AI-generated content. CIPO has not published guidance on how to address these issues, but other jurisdictions such as the UK have done so and provide valuable insight into the various consequences of certain regulations and may serve as a potential benchmark for Canada.
Evidently, the Canadian Copyright Act will need to be amended to address these concerns and uncertainty. Key commercial challenges and legal questions arise in the context of both the input used to train the AI system and the output generated by the AI. As the use of generative AI continues to grow, it will be important for businesses to be aware of the legal considerations surrounding AI-generated content and data inputs to ensure compliance and mitigate risk.
Part two of our series can be accessed here. For part three, click here.
Footnotes
1. https://www.ourcommons.ca/Content/Committee/421/INDU/Reports/RP10537003/indurp16/indurp16-e.pdf .
4. I Ibid at 60
5. https://www.ic.gc.ca/app/opic-cipo/cpyrghts/dtls.do?fileNum=1188619&type=1⟨=eng
6. https://www.legislation.gov.uk/ukpga/1988/48/section/9. "Computer-generated" is defined to mean a work "generated by a computer in circumstances such that there is no human author of the work"
7. https://www.ourcommons.ca/Content/Committee/421/INDU/Reports/RP10537003/indurp16/indurp16-e.pdf
About Dentons
Dentons is the world's first polycentric global law firm. A top 20 firm on the Acritas 2015 Global Elite Brand Index, the Firm is committed to challenging the status quo in delivering consistent and uncompromising quality and value in new and inventive ways. Driven to provide clients a competitive edge, and connected to the communities where its clients want to do business, Dentons knows that understanding local cultures is crucial to successfully completing a deal, resolving a dispute or solving a business challenge. Now the world's largest law firm, Dentons' global team builds agile, tailored solutions to meet the local, national and global needs of private and public clients of any size in more than 125 locations serving 50-plus countries. www.dentons.com
The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances. Specific Questions relating to this article should be addressed directly to the author.
