Blasting a Wide TDM Hole in the Structure of Copyright is Not the Answer
The ongoing wrestling match-cum-dance between the creative sector and AI developers over the uncompensated and unauthorized use of copyrighted content for AI training is being played out in different ways in different countries. In the US it is largely a legal play in the courts at the moment, with mixed results for both sides. However, President Trump has made concerning public comments siding with the AI industry, saying it is impractical for AI developers to pay copyright holders for AI training (and besides, China doesn’t do it). Congress is still considering its options. In Australia, the Productivity Commission, never a friend of intellectual property, has just issued an interim report recommending the adoption of a Text and Data Mining (TDM) exception in Australia to boost development of the AI industry locally. The Australian creative sector mobilized quickly and has pushed back hard against this proposal, with the government now saying that it has no plans to amend the Copyright Act. In the UK, where there is a TDM exception but only for non-commercial purposes, the Starmer government quickly adopted a pro-AI strategy, part of which was to propose an expansion of TDM to include commercial purposes, although subject to an opt-out for rights-holders. That ignited a major storm among leading British creatives from Paul McCartney and Elton John on down. Through a unified campaign, British creators were able to gain support in the Upper Chamber (House of Lords) to slow down the legislation. As a result, the TDM issue has now been earmarked for further consultation and study. One thing is certain, the creation of a wide TDM exception is a sure way to stifle a nascent but rapidly developing licensing market for copyrighted content used for AI training.
It seems as if TDM, or more permissive TDM, is testing the boundaries of copyright just about everywhere. So, what about Canada? Canada has no TDM exception in its copyright law and, unlike the US, has clearly defined fair dealing exceptions that do not lend themselves to expansive court interpretation. Like other countries, it is trying to figure out how to not get left behind as the AI race accelerates. Canada initially had a first mover advantage in terms of AI research, given the work of Geoffrey Hinton, Yoshua Bengio and others, but recently it has been falling behind, notably lacking native startups. The cluster effect is not happening, with Canadian innovation going elsewhere for commercialization. To address these challenges, the new Carney government has appointed a dedicated Minister of Artificial Intelligence and Digital Innovation, former journalist Evan Solomon. This is the first time such a position has existed. One of Solomon’s first acts was to accelerate launch of an AI strategy beginning with a new consultation released on October 1 (closing at the end of this month), in the form of a survey to “help define the next chapter of Canada’s AI leadership”. This survey asks many relevant questions regarding AI and how it could be best developed in Canada but manages to mostly steer clear of the thorny question of AI training and copyright. The only question tangentially related to this issue is the following;
“Which infrastructure gaps (compute, data, connectivity) are holding back AI innovation in Canada, and what is stopping Canadian firms from building sovereign infrastructure to address them?”
Clearly this consultation is not going to turn over the TDM rock, at least not directly.
In the past couple of years, the government has issued two consultation papers on AI, one in 2021 and another last year as well as a “What We Heard” report. This report, issued earlier this year, summarizes the “great divide” between AI developers and the content industry. It’s first observation was that “Creators oppose the use of their content in AI without consent and compensation” but then goes on to say that “User groups support clarifications that TDM does not infringe copyright”.
After a couple of other observations about the centrality of human authorship and the need for transparency surrounding the use of copyright-protected works in the training of AI, the paper observed that there is “no consensus about whether existing legal tests and remedies are adequate”. That is the nub of the issue. There is no consensus, and while the courts are struggling with this issue (including in Canada, as I wrote about here and here), what Canadian creators fear is the introduction of a wide TDM exception in the name of maintaining “Canadian competitiveness”.
The launch of the new AI strategy and the evolution of the way in which copyrighted content is described in government consultation documents is indicative of the pressures on the government to shore up Canada’s AI strategy. It is interesting to note the shift in the definition of TDM from 2021 to today.
The definition provided in the 2021 consultation document described TDM as follows;
“The process of conducting TDM may require the making of reproductions of large quantities of works or other copyright subject matter to extract particular data and information from them. This process may be carried out using scientific or text-based data, as well as images, sounds, or other creative works.”
In the most recent consultative document, that definition has evolved;
“Text and data mining (TDM) consists of the reproduction and analysis of large quantities of data and information, including those extracted from copyright-protected content, to identify patterns and make predictions.”
Note the shift from “works” to “data”.[i] It’s a subtle difference but is hugely significant because data and facts are not protectable under copyright whereas the creative elements of original works are. The cultural sector is rightly concerned.
The Coalition for the Diversity of Cultural Expressions (CDCE), a major arts and creatives lobby group, is currently pressing Ottawa on a number of cultural issues, including AI. Among its AI asks are to;
Against these demands is the pressure coming from AI advocates who will argue that if the US loosens restrictions on use of copyrighted content for AI training, Canada will have no recourse but to follow. In other words, as goes the US, so goes Canada (or for that matter, the UK, Australia and others). Thus, what is happening in the US courts, and perhaps in Congress, is of critical importance for the creative sector everywhere including, in particular, Canada.
The issue of AI training on copyrighted content will need to be resolved sooner or later. Licensing solutions are developing quickly and if Canada can wait a bit longer it may be able to adopt licensing as the preferred solution (although the “What We Heard” report noted that “Some (intervenors) argued that licensing is an unnecessary burden because it may not be clear that copyright is engaged or that works used in TDM are being reproduced in the first place.”). There is pressure on the Carney government to take early action since AI industry developments are moving at lightning speed. With the TDM train gaining momentum in Canada and elsewhere, Canadian creators are understandably uneasy about what is likely to happen next.
As the CDCE notes, culture is a major economic and social pillar in Canada. In 2023, it generated $63.2 billion in value added and employed 669,600 people. Throwing all that under the bus in the name of remaining competitive on AI is a flawed choice, a point also made by the creative sectors in the UK, Australia and elsewhere. However, with the AI horse well out of the barn, copyright cannot be seen as an obstacle to innovation, an accusation freely levelled at it by some in the AI industry. Rather, it must be seen as a partner in innovation, which is where licensing comes in.
Blasting a wide TDM hole in the protection and incentive structure that copyright provides the creative sector is not the answer. The creative sector is watching and waiting anxiously.
© Hugh Stephens, 2025. All Rights Reserved
[i] I am indebted to Erin Finlay, partner at Stohn Hay Cafazzo Heim Finlay LLP for drawing these changing definitions to my attention