In Canada, the Federal Government has just announced a $2.4 billion package of measures “to secure Canada’s world-leading AI advantage.” This sum is in addition to the $2 billion of public funds already invested since 2017, when Canada became the first country in the world to launch a national AI strategy. Now, as then, the stated goal is to position Canada at the forefront of the global race to scale up and adopt AI.
Over the same period, however, the government’s messaging around Canada’s AI-copyright policy has been anything but consistent. In the spirit of the typical tragicomedy, this blogpost tells the story (so far) in three Acts and an Epilogue.
Act I
When the Committee charged with conducting Canada’s 2019 Copyright Act Review turned its mind to AI, its primary concern was with “help[ing] Canada’s promising future in artificial intelligence become reality” (Report 2019). Lawyer Maya Madeiros explained to the Committee that copyright law had the potential to become a serious obstacle to AI’s development:
AI learns to think by reading, listening and viewing data, which can include copyrighted works such as images, video, text and other data.… The training process can involve reproductions of the training data…. It is unclear whether the use of copyrighted works for training an AI system is considered copyright infringement if the…owner’s permission is not obtained.… This uncertainty can limit the data that is used by AI innovators to train the AI system. The quality of the dataset will impact the quality of the resulting trained algorithm.
The problem posed by copyright law to the training of AI systems and the quality of their outputs is by now well known. High profile litigation and media-covered controversies have drawn wide-spread attention to the essential role that pre-existing works play in the process of “teaching” an AI to “think.” What is interesting to note, however, is the ease with which the 2019 Committee could see that AI had an inconvenient copyright problem—one that should be statutorily solved. Its final report simply recommended that “the Government of Canada introduce legislation to amend the Copyright Act to facilitate the use of a work or other subject-matter for the purpose of informational analysis”.
Also striking in retrospect is that there were apparently “very few witnesses who argued against an exception for informational analysis”. Only one witness is quoted as having argued that such uses should be licensed by collective societies.
Act II
The Committee’s recommendation to add an exception for “informational analysis” was not acted upon. Instead, two years later, the Canadian Government launched a public consultation that specifically sought input on the matter of copyright and AI. Submissions received on the training of AI systems were subsequently summarized in the following terms:
Stakeholders from the technology sector, scholars, and user groups generally argued for an exception making clear that the use of works in text and data mining activities (TDM) does not require additional authorization from rightsholders. Creative industries were of the view that a new exception is not desirable, as it would preclude rights holders from receiving fair compensation for the use of their works in TDM activities.
What seemed to have been a relatively uncontroversial proposition in 2018-19—that copyright ought not to obstruct the development of AI by effectively preventing the lawful and optimal training of AI systems—had, by 2021, become the controversial assertion of a select group of stakeholders and experts (including those with whom I made joint submissions, here and here). And this position was now uniformly opposed by stakeholders in the creative industries, who were suddenly and overwhelmingly concerned with ensuring “fair compensation” for rightsholders.
Act III
The political ground continued to tilt. Two years later, presumably unsatisfied with these responses, Canada’s Government launched yet another public consultation, now specifically on Copyright in the Age of Generative AI. While conceding that the actual copyright issues remained unchanged, Canadians were now invited to review the policy considerations in light of their intervening “experience with generative AI”. It was expressly acknowledged that “some stakeholders have raised concerns about the impacts of AI on creators and creative industries going beyond copyright per se”. The consultation paper explained that the Government’s aim was “to balance two main objectives”: “support[ing] innovation and investment in AI and emerging technologies” on one hand, and, on the other, “support[ing] Canada’s creative industries and preserv[ing] the incentive to create and invest provided by the rights set out in the…Copyright Act, including to be adequately remunerated for the use of their works”.
I would be remiss not to point out that Canada’s Copyright Act contains no such general right for authors—never mind the creative industries—to be “adequately remunerated”. Copyright simply establishes limited exclusive rights to perform specified actions in relation to protected works, which owners may exchange for whatever value the market happens to ascribe to them.
More importantly, however, the rights and interests in the balance, as articulated in the consultation paper, include only industries’ interests in incentives to innovate, create and invest, and rightsholders’ claims to adequate remuneration. Nowhere in this so-called balance was any mention made of the public side of the copyright balance—the public’s interest in the creation and dissemination of works, for example, or users’ rights to make fair and lawful uses of protected works, or the importance of the public domain (in which facts and information—i.e. data—reside). The articulation of the government’s overarching policy approach thus overstated the scope of copyright holders’ entitlements, focused on industries’ economic interests rather than the interests of individuals or the public in general, and entirely disregarded the users’ rights that are central to the copyright balance as repeatedly confirmed by Canada’s Supreme Court (here, here, here, here, here—and one more time for the people in the back!—here).
Canada, often torn between its US neighbors and its European colonial roots, is always an interesting jurisdiction to watch when it comes to copyright policy reform. Cross-border trade relations (and political pragmatism) often demand coherence with the US law and policy, while the historical and jurisprudential through-lines lead back to the UK. The Quebec influence and bilingual requirements in Ottawa, however, also feed policy affinities with France and continental Europe. For now, while the US position on the copyright legalities of TDM remains to be resolved through litigation focused on transformative fair use (which may not overlap neatly with Canada’s more restrictive fair dealing provisions), all signs point to the rising influence of the European approach in Canada.
In Europe, where copyright tends to be less utilitarian in its focus and more concerned with protecting owners and le droit d’auteur, the contestation over copyright and AI has been settled (on paper at least) in Articles 3 and 4 of the Digital Single Market Directive. Article 3 creates an exception for TDM for scientific research and available only to research organizations and cultural heritage institutions, while Article 4 requires member states to provide a generally available exception for “reproductions and extractions of lawfully accessible works…for the purposes of text and data mining”, but allows owners to “opt out” by expressly reserving their rights in an “appropriate manner”. Importantly, by adopting these specific exceptions, the EU legislature has confirmed that TDM and AI training processes are copyright-relevant activities (Senftleben, 2023) And, as is typically the case in European copyright laws, limits and exceptions to the copyright owners’ exclusive rights are specific and narrowly drawn.
According to the so-called Brussels Effect, market forces alone are often sufficient to ensure that EU rules govern the global operations of multilateral corporations and ultimately set norm for the international stage (Bradford, 2000). Certainly, the normative baseline seems to have shifted, in the Canadian copyright policy discourse, to assume, as a starting point, that authors and owners have a copyright entitlement to control the use of their works for AI training, such that the policy problem becomes how to operationalize or enforce that right and, almost immediately, how to monetize it.
Canada’s latest consultation questionnaire did not ask, for example, whether remuneration would be appropriate for rightsholders whose works were used in the training of AI; rather it asked, “what level of remuneration would be appropriate for the use of a given work in TDM activities?” It did not ask whether it should include an exception to explicitly permit TDM activities, rather it asked, “If the Government were to amend the Act to clarify the scope of permissible TDM activities, what should be its scope and safeguards?”
Epilogue: Tragicomedy? Canada’s AI-Copyright Policy Challenge
It remains to be seen what kinds of responses this consultation elicited; but it seems increasingly likely that copyright and its mechanisms of control will soon be called upon to play a larger role in Canada, as they have in Europe, in regulating and restraining the capacity of AI developers to train their models on copyright materials—and that this will be hailed by some interested stakeholders as a victory for Canada’s creative industries and the creators they (purport to) represent. Unfortunately, I expect, it will be anything but.
As I argued in my response and elsewhere, copyright law is neither apposite nor equipped to govern the way that generative AI is developed, trained, or deployed. It is unlikely to benefit Canadian creators in any meaningful way; it is, however, likely to obstruct the kinds of research, training, testing, transparency, and competition essential to the responsible development of AI (Fiil-Flynn, et al, 2022). How this could help “secure Canada’s AI advantage” remains a mystery. In an ironic twist at the conclusion of the third Act of Canada’s AI-Copyright play, with industry lobbyists waiting in the wings, copyright policymakers could be poised to deal an unfortunate blow to Canada’s AI strategy.
________________________
To make sure you do not miss out on regular updates from the Kluwer Copyright Blog, please subscribe here.
