One of the characteristics of several devices operating on artificial intelligence (AI) is to feed on large directories of texts and images. Generative AI systems use deep learning models capable of generating content based on works from which the systems have been trained. Such systems perform text and data mining, that is, the reproduction and analysis of large quantities of data and information. This massive extraction allows generative AI systems to determine trends and make predictions, even generating text and images from what they analyze.
Among these masses of information, there are some that are extracted from copyrighted content. These texts, sounds and images that exist in public space are generally the fruit of the work of creators.
In Canada, according to the Copyright Act, companies that develop AI systems must seek explicit permission from rights holders before using their works. Those who produce new content using AI cannot assume that rights holders have consented to the use of their works to power systems that generate other content.
With the development of generative AI, pressures from stakeholders in the technology sector are increasing. It is argued that the use of works in text and data mining should not require permission from rights holders. Hence the demands to change the law to exempt data mining operations from the obligation to seek permission from copyright holders.
Copyright holders have another perspective. In a brief submitted as part of a consultation led by the Canadian government, the Coalition for the Diversity of Cultural Expressions, which represents more than 360,000 creators and nearly 3,000 cultural businesses in the country, calls for resistance to the demands of companies that want to waive the requirements of the law to power their AI systems.
The Coalition opposes the addition of new exceptions to the Copyright Act which would allow protected works to be used to train generative AI systems in particular to carry out text and data mining . In addition, the Coalition calls for transparency obligations to be imposed on companies developing generative AI systems regarding the copyrighted works used to train their systems.
Stéphanie Hénault, director of legal affairs at the National Association of Book Publishers, explained to journalist Étienne Paré that these companies must be forced to disclose the register of works they have used for content creation purposes. “This is the best way to ensure that generative AI will continue to evolve within a legal framework. Additionally, users could know the sources of the systems they use. This would allow us to see if artificial intelligence is biased. »
AI GAFAMs
As for products resulting from mechanical AI generation processes, they do not involve original human expression. For this reason, they must not become “works” protected by copyright. Allowing mega-companies that control AI systems to be able to claim an exclusive right to content generated by such devices opens the door to an increase in the control of large digital multinationals not only over the Internet, but also over the information circulating there.
In the European Parliament, an MP who has studied the questions of competition and monopolization that the unbridled development of AI conceals warns against the risk of reconstituting “GAFAM of artificial intelligence”. She deplores that discussions on the regulation of AI have so far paid too little attention to the threat of concentration of this market in the hands of digital giants, Google, Amazon, Facebook (Meta), Apple or Microsoft.
Public authorities must avoid repeating the mistakes made when setting up social networks. At that time, in the name of “promoting innovation”, States considerably limited the responsibilities of search engines and social networks by relieving them of the obligations to anticipate and limit the risks to which their users. We experience the harmful consequences of these misguided policies every day. We cannot seriously claim to “promote innovation” by granting privileges to Web giants if this is done at the expense of creators.
ADISQ explained in its brief on generative AI that the current Copyright Act is sufficiently clear to provide a framework within which AI can develop while ensuring that the rights of creators are respected. A reminder that innovation can and must be deployed while respecting the rules of the game.