Les informations de ce site Web à été fournie par des sources externes. Le gouvernement du Canada n'assume aucune responsabilité concernant la précision, l'actualité ou la fiabilité des informations fournies par les sources externes. Les utilisateurs qui désirent employer cette information devraient consulter directement la source des informations. Le contenu fournit par les sources externes n'est pas assujetti aux exigences sur les langues officielles et la protection des renseignements personnels.
Submission to the Consultation on a Modern Copyright Framework for Artificial Intelligence and the Internet of Things from the Canadian Independent Music Association
Prepared by Gord Dimitrieff
Submitted September 17, 2021
Contact: Andrew Cash, President
The Canadian Independent Music Association (CIMA) is the national not-for-profit trade association representing the English-language, Canadian-owned sector of the music industry. CIMA's membership consists of over 350 Canadian-owned companies, representing over 6,000 Canadian artists and involved in every aspect of the music, sound recording and music-related industries. They are exclusively small businesses which include record producers, record labels, recording studios, managers, agents, licensors, music video producers and directors, creative content owners, artists and others professionally involved in the sound recording and music video industries.
The recommendations made in this document complement CIMA’s remarks to the Standing Committees on Canadian Heritage and Industry during the 42nd Parliament, as they undertook their reviews on copyright. CIMA’s recommendations to the Committees included the repeal of the $1.25 million exemption in Section 68.1(1)(a) of the Copyright Act, which currently provides an outdated and unnecessary subsidy to broadcasters on the royalties they pay at the expense of creators; the amendment Section 2 of the Copyright Act to allow recorded music in television and film productions to be eligible for Section 19 public performance remuneration; and making the private copying regime technologically neutral to cover audio recording devices such as digital audio recorders, tablets and smartphones. These recommendations reflect the broad consensus within the Canadian music industry on what crucial steps need to be taken in order improve the livelihood of our music creators.
CIMA firmly believes that to create is to innovate, that all innovators are creators, and likewise all creators are innovators. While not synonyms, creation and innovation are such closely linked concepts that one cannot exist without the other. It is a false dichotomy to file researchers and innovators into one category and creators and rights-holders into another. To be constructive, public debate about Artificial Intelligence and the Internet of Things must be about how artistic work and technological development contribute the advancement of society as a whole.
In the context of the music industry, creators must be defined as everyone in the music ecosystem of creating, recording, performing and commercializing music. They are the artists, songwriters, composers and the companies that support them — such as labels, managers and publishers. We urge the Government of Canada to view these consultations through this lens to ensure all who create and commercialize intellectual property are properly supported and protected by Canadian law.
Text and Data Mining (TDM)
TDM-based algorithms are of particular concern to the cultural industries when one considers this technology powers both social media feeds and the recommendation-systems of consumer services. Extensive scientific research has shown that artificial intelligence models based on text and data mining from freely available sources on the Internet overrepresent dominant viewpoints, stereotypical and derogatory associations, and are potentially damaging to marginalized populations.Footnote 1 In 2016, a high-profile example of this was revealed by researchers at Boston University and Microsoft Research who discovered that AI models trained on Google News articles exhibited a disturbing level of inherent sexism, equating “Man is to Computer Programmer” as “Woman is to Homemaker” (Bolukbasi et al 2016). These biases extend into the recommendation-systems of streaming services, creating an unequal playing field for any voices that fall outside the white, male-dominated mainstream (Ferraro et al 2021). As legal scholar Amanda Levendowski notes, “even as our banks and our bosses, our cars and our courts increasingly adopt AI, bias remains a significant and complex problem” (Levendowski 2018).
Throughout history, from the emergence of jazz, rock and roll and hip-hop, independent music companies have been at the forefront of new (and often controversial) ideas, finding commercial audiences for marginalized and unpopular voices. For this tradition to continue in the new world of online services, the independent music industry needs AI-enabled editorial and curatorial systems to be as balanced and thoughtful as their old-media counterparts.
AI systems are ‘trained’ with text and data mining — in other words, they ‘learn’ by reading, viewing, and listening to copies of works created by humans. While some companies developing AI systems have argued for broad and sweeping copyright exceptions to permit the unencumbered mining of these works, broad exceptions to copyright for this purpose would force the innovative work of some creators into the unjust role of subsidizing the innovative work of other creators. If not for the artistic and literary contributions to society, there would be no text and data from which to develop AI systems.
Although commercial entities such as Google and Spotify are able to negotiate licenses with rights-holders for TDM activities, it is unrealistic to suggest that academic institutions and independent researchers have the ability to negotiate comparable licenses on equal, or even similar terms. Without having access to the underlying bodies of data, downstream AI researchers are unable to investigate the existence of inherent biases, nor make comparisons to how AI systems react with alternative bodies of data. As the Boston University and Microsoft Research team noted, the proprietary nature of the Google News dataset makes it “impracticable and even impossible … to reduce the [biased] stereotypes during the training of the word vectors” (Bolukbasi et al 2016).
To ensure universal and equal access to socially representative works for AI research and development while fairly compensating rights-holders, CIMA recommends that the private copying regime implemented by Part VIII of the Copyright Act be made media-neutral and expanded to include private copying for the purposes of text and data mining and AI development.
Authorship of works generated by AI
Although there is no universally accepted definition of ‘Artificial Intelligence,’ there is no debate that at its most fundamental level, AI is (as are all computer programs) a form of applied mathematics, constructed with the binary number system for the purpose of controlling a machine. This is true of all AI regardless of the form it takes, including natural language processing, deep learning, algorithmic methods, etc.
CIMA categorically rejects the notion that authorship can be attributed to a machine, and unreservedly affirms its support of Canadian copyright jurisprudence, which suggests that an author must be a natural person who exercises skill and judgment in creating the work.
Infringement and liability regarding AI
Accidental plagiarism happens frequently, and there have been famous cases. For example:
“Anybody Seen My Baby?” is a song by the English rock band the Rolling Stones, featured on their 1997 album Bridges to Babylon. It was written by Rolling Stones vocalist Mick Jagger and guitarist Keith Richards, and writing credits were added for k.d. lang and Ben Mink due to the similarities the chorus possesses with lang's 1992 hit “Constant Craving”. (See https://en.wikipedia.org/wiki/Anybody_Seen_My_Baby%3F)
To help limit their risk of liability, music publishers rely on their expert knowledge of musicology, carry errors-and-omissions insurance, and (as in the above case) often negotiate co-authorship after the fact. It would be entirely illogical, if not patently absurd, for AI-generated work or AI system owners to benefit from any less liability for infringement than human-generated work or traditional publishers, simply because a machine was employed in the creative process. For copyright to be meaningful, infringers must be liable, regardless of the methods they might employ.
The Internet of Things (IoT)
CIMA supports the development of open and standards for interoperability and the right to repair; however, we respectfully suggest that there might be more appropriate avenues to regulate the problem of irreparable products than the Copyright Act, such as the Canadian Environmental Protection Act, the Federal Electronic Waste Strategy and Transport Canada’s Canada Motor Vehicle Safety Standards. Although CIMA does not endorse the use of technological protection measures (TPMs) per se, we do not fault any rights-holder who decides they are a necessary protection for their business model. One can only imagine what would happen to the hotel industry if it were perfectly legal for non-paying guests to defeat the TPM employed in room keycards. Any exemptions included in the Copyright Act that would allow a TPM to be defeated must be for very specific, limited and narrow purposes.
With respect to the Copyright Act, CIMA is concerned about possible unintended consequences created by the Act’s ephemeral recording exemptions within the context of the Internet of Things. With the growing popularity of so-called “smart speakers” (e.g. Amazon Alexa, Google Assistant, etc.) and Internet-connected in-car audio systems, a broadcaster’s “broadcasting” can now include the broadcaster’s proprietary embedded software systems in these devices. Sections 30.8 and 30.9 of the Act permit broadcast undertakings to make and keep copies of sound recordings “for the purpose of their broadcasting” for up to 30 days without the need to pay additional royalties. In this context, one can easily see how broadcasters could operate quasi on-demand programming powered by the Internet of Things for periods of up to 30 days, or possibly longer if the ephemeral recordings are “renewed” on the broadcaster’s IoT devices at least every 30 days.
CIMA therefore recommends that consumer-oriented IoT devices be specifically excluded from the ephemeral recording exemptions provided by Sections 30.8 and 30.9 of the Copyright Act.
Inherent biases within artificial intelligence systems are a real and significant problem, with serious implications for the daily life of citizens throughout society. Developers and independent researchers need access on equal terms to underlying bodies of artistic and literary work to investigate the causes of these biases and formulate solutions. This is of particular interest to the independent music industry, which requires balanced and thoughtful editorial and curatorial systems for a well functioning market. To ensure developers and researchers have access to socially representative works for AI while fairly compensating rights-holders, CIMA recommends that the private copying regime implemented by Part VIII of the Copyright Act be made media-neutral and expanded to include private copying for the purposes of text and data mining and AI development.
Artificial Intelligence is a form of applied mathematics used to control machines; it is not magic or voodoo. CIMA firmly supports Canadian copyright jurisprudence, which suggests that an author must be a natural person who exercises skill and judgment in creating the work — not a machine.
For copyright to be meaningful, infringers must be liable regardless of the methods they might employ. The use of artificial intelligence, or any other machine, in the creative process should not provide a shield from liability.
CIMA supports the development of open standards for interoperability and the right to repair; however, any exemptions included in the Copyright Act that would allow a TPM to be defeated must be for very specific, limited and narrow purposes. CIMA is concerned about possible unintended consequences created by the Act’s ephemeral recording exemptions within the context of the Internet of Things and recommends that consumer-oriented IoT devices be specifically excluded from the ephemeral recording exemptions provided by Sections 30.8 and 30.9 of the Copyright Act.
Bender, Emily M; Gebru, Timnit; McMillan-Major, Angelina & Shmitchell, Shmargaret. “On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?” FAccT '21: Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency, March 2021. Pages 610–623. https://doi.org/10.1145/3442188.3445922
Basta, Christine; Costa-jussa, Marta & Casas, Noe. “Evaluating the Underlying Gender Bias in Contextualized Word Embeddings.” Proceedings of the First Workshop on Gender Bias in Natural Language Processing, 2019. Pages 33–39.
Beltagy Iz; Lo, Kyle & Cohan, Arman. “SciBERT: A Pretrained Language Model for Scientific Text.” In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), 2019. Association for Computational Linguistics, Hong Kong, China, pages 3615–3620. https://doi.org/10.18653/v1/D19-1371
Ferraro, Andres; Serra, Xavier & Bauer, Christine. “Break the Loop: Gender Imbalance in Music Recommenders.” CHIIR '21: Proceedings of the 2021 Conference on Human Information Interaction and Retrieval, March 2021. Pages 249–254. https://doi.org/10.1145/3406522.3446033
Kurita, Keita; Vyas, Nidhi; Pareek, Ayush; Black, Alan W & Tsvetkov , Yulia. “Measuring Bias in Contextualized Word Representations.” In Proceedings of the First Workshop on Gender Bias in Natural Language Processing, 2019. Pages 166–172.
Levendowski, Amanda. “How Copyright Law Can Fix Artificial Intelligence's Implicit Bias Problem.” Wash. L. Rev. 579 (2018). https://digitalcommons.law.uw.edu/wlr/vol93/iss2/2
Tan, Yi Chern & Celis, L Elisa. “Assessing social and intersectional biases in contextualized word representations.” In Advances in Neural InformationProcessing Systems, 2019. Pages 13230–13241.
Zhao, Jieyu; Wang, Tianlu; Yatskar, Mark; Cotterell, Ryan; Ordonez, Vicente & Chang, Kai-Wei. “Gender Bias in Contextualized Word Embeddings.” In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers). Association for Computational Linguistics, Minneapolis, Minnesota, pages 629–634. https://doi.org/10.18653/v1/N19-1064