OTTAWA — Canadian creators and publishers want the government to do something about the unauthorized and generally unreported use of their content to train generative artificial intelligence (AI) systems.
But AI companies argue that using the hardware to train their systems does not violate copyright, and say limiting its use would hamper AI development in Canada.
Both sides make their arguments in newly released submissions to a federal government-led consultation on copyright and AI, which is examining how Canadian copyright laws should respond to the emergence of generative AI like ChatGPT and Claude.
Generative AI can create text, images, videos, and computer code based on a simple request, but to do so, systems must first study large amounts of existing content.
In its submission to the government, Access Copyright argued that most, and potentially all, major language models “currently profit from the unauthorized use and reproduction of copyrighted works.”
This takes place in a “black box,” according to Access Copyright, which represents writers, visual artists and publishers.
“Copyright holders know this is happening, but because of the information asymmetry between them and AI platforms, they cannot determine who is conducting the activity, with whose works, and have no mechanism to prevent it from happening.”
Music Canada, which represents the country’s major record labels, said last year that a fake AI-generated song imitating the voices of Drake and The Weeknd “showed one thing very clearly: the models and systems of ‘AI have already ingested massive amounts of datasets without permission from the data source or rights holders’.
The Writers Guild of Canada has called on the government to start imposing basic disclosure and reporting obligations on AI companies. It says developers have full knowledge of what work is being exploited and how it is being used, while creators have none of that information.
Some organizations have signed licensing agreements with AI companies. But the Canadian Authors Association said rights holders face “immense obstacles” in licensing their content “because they don’t know which works are being used” by which companies.
The association asked Canada to clarify that text and data mining is subject to copyright laws.
There are numerous lawsuits pending in the United States over the use of copyrighted material by generative AI systems, including one filed this week by the world’s largest record labels against two music-generating artificial intelligences.
The Canadian Media Producers Association (CMPA) said court cases illustrate the problem with the lack of transparency, citing one case in which an AI company argued that the rights holder could not pursue an infringement claim unless it could specify the exact work used for training.
“Rights holders will also undoubtedly face similar evidentiary challenges, as many datasets used to train generative AI systems are allegedly destroyed once initial training is complete,” the statement said.
The group said it was an issue that “demands immediate attention” and called on the government to implement transparency requirements.
The AI industry hits back
AI companies argue that the kind of transparency rights holders are demanding is unrealistic.
Microsoft has told the government that training AI systems at scale involves “vast volumes” of data and that companies should not have to keep records of that data or disclose the content used for training.
“It would not be possible to record such information and such a requirement would hinder the development of AI,” the company claimed.
The company states that analyzing works and learning concepts and facts does not constitute “copyright infringement.”
Google said that AI training is already exempt under current copyright law, but that the government should adopt an exemption to make this explicit.
Google stipulated that requiring permission to use content for training purposes would expose competitively sensitive information and “effectively block the development and use of large language models and other types of AI peak”.
The multinational also indicates that AI developers do not have access to precise information on the copyright status.
“In fact, such a single data source does not exist anywhere in the world. So complying with disclosure rules may simply be impossible from the start.”
Canadian AI company Cohere said using content to train AI systems works similarly to how an individual reads books to become more informed.
The company assured that the process did not violate copyright and argued that this should be clear in the law. Otherwise, “Canada’s ambitions to become home to world-class AI companies and ecosystems” could be compromised.
The Canadian Council of Innovators, which represents Canada’s technology sector, said the disclosure requirements would hurt smaller companies, as opposed to their big tech rivals. He warns that this would “seriously hamper the potential of Canadian businesses to grow significantly.”
Anja Karadeglija, The Canadian Press
