Deepfakes and the Copyright Connection: Analysing the Adequacy of the Present Machinery
By Akhil Satheesh
Introduction:
As the United States, for the second consecutive year accounts for deepfakes in the National Defense Authorization Act, (NDAA), and calls for the Department of Homeland Security to prepare annual reports[1] on this technology for the next 5 years, the issued presented by it can no longer be swept under the rug.
Deepfakes consists of a pair of competing algorithms[2], a generator that renders artificial content and a discriminator that identifies which aspects are original or fake. This generative adversarial network (GAN) is self-learning, and each time the discriminator algorithm observes content to be fake, this information is supplied to the generator, resulting in the creation of more convincing images/videos with each new iteration.
Deepfakes possesses immense potential for malicious use, with over 90%[3] of deep fake content being false pornographic videos of individuals without their consent and false political propaganda.[4]
Copyright Measures in India:
Under the Copyright Act, 1957, liability for copyright infringement comes under Section 55[5], (civil remedies such as injunction, damages etc) and Section 63[6] (criminal liability punishable with imprisonment between 6 months to 3 years and fine). However, using these provisions as defense depends upon the categorisation of deep fakes as copyright infringement. This is determined by considering the application of the fair dealing doctrine in the deepfake context.
Section 52[7] of the Act enshrines the fair dealing doctrine, wherein, any act included in the exhaustive list under the section is exempt from categorisation as copyright infringement. Since deep fakes function off the repurposing of existing copyrighted materials such as photos, videos etc., such usage is automatically considered to be in violation of copyright protection, as it is not provided for in the aforementioned list. This measure is rather inflexible as it prohibits use of deep fake technologies for any purpose, regardless of legitimate intent or otherwise.
With regard to considering deep fakes as transformative use within under review as per Section 52(1)(a)(ii), the issue is that there exists no judicial foundation for the same, as this exemption has only been granted for literary works (guidebooks containing a substantial amount of copyrighted information from a source book), as was in the Narendera Publishing House[8] case. Till such an explicit categorisation is made, deep fakes will continue to be regarded as copyright infringement.
Deep fakes are also subject to the implications presented by the moral rights bestowed upon the creator, under Section 57[9]. Under Section 57 (1)(b), the author is imbued with the right against mutilation, distortion, modification or any other similar act done to his work, if the same may adversely affect his reputation or honour. Deep fakes primarily function off the modification of copyrighted content which can be brought under the umbrella of distortion or mutilation, and hence considered to be violating these moral rights, depending on the nature of use of the original content. Therefore, despite having no express legislation on the subject, India runs a tight ship with respect to deep fakes.
The US Stance on Deepfakes:
Fair use, under US law, is more flexible and enshrined under S.107[10] of the Copyright Act of 1976, employing a four-factor test[11], the parameters being:
- Purpose and character of use
- Nature of original work
- Amount and Substantiality of content borrowed
- Effect of such use on the potential market
The fair use doctrine, therefore has a wider ambit than fair dealing. Since deep fakes are created for fundamentally different purposes than those envisioned when creating the original work and its nature also being different in most cases, the purpose and character criteria are adequately met. Further, U.S. Courts have repeatedly concluded that even if there is clearly identifiable copyrighted content[12], or a significant portion[13] of the original work is borrowed, if the usage is transformative, it would be “fair use”.
The United States Court of Appeals for the Second Circuit, in the case of Bill Graham Archives case[14], enunciated that transformative use of copyrighted work is excused when it is different from the original expressive purpose of the images. Similarly, regarding market impact, this technology being in its nascent stages, possesses no real market value and consequently poses no threat to the market of the original creation, meeting this requisite as well.
Therefore, US Copyright legislation enables deep fake content creation, considering it as fair use. However, as these laws do not discriminate based on the intent of the creator, it allows the categorisation of deep fakes produced with even mala fide intent, as parodies and even protects as the same.
To systematically combat this issue, several American states are proposing bills to eradicate instances such as the illegal use of digital replicas and other acts pertaining to the depiction of real-life individuals and events[15] and even deep fake pornography, with Virginia leading the charge, amending pre-existing revenge porn laws to account for deep fake content as well, under the “falsely-created material”[16] head.
Copyrightability of Deep Fakes:
As deep fake content freely exists in society, the question arises as to whether the same can be granted copyright protection and if so, who holds these rights. Considering that this content is a product of an autonomously functioning computer program, there arises two possible definitions for the ‘creator’:
- The creator of the deepfake program;
- The user of the already available program, who employed it as a tool[17] to physically render his vision.
Either way, no rights are vested upon the original AI per se, due to the indispensable need for the human element in the creative process. The US Supreme Court, has explicitly held that copyright protection only extends to works that are original, and founded in the creative powers of the mind[18].
The UK copyright law however accounts for works created by Artificial Intelligence, under the ambit of ‘computer-generated work’. Section 178[19] of the Copyright, Designs and Patents Act, defines the same as ‘the work generated by computer in circumstances such that there is no human author of the work;’, which implicitly includes deep fake created content, as the definition above is also applicable to AI generated work. Furthermore, the Act under Section 9(3)[20] also states that in the case of such computer-generated works, the author is the person who established the requisite arrangements for the creation of the work.
In India, in the Camlin case[21], the Delhi High Court elaborated upon the definition of author and stated that authors are natural persons who create works as the product of skill and labour and only they are vested with copyright. Although this is the case, recently India made a bold leap forward, by recognise authorship rights in an AI for an artistic work that was created entirely by the same. RAGHAV, an artificial intelligence painting program, was registered as the co-author of ‘Suryast’[22] along with Ankit Sahni, the co-author and the registered copyright holder by the Copyright Office, a decision rendered in accordance with the AIPPI Adopted Resolution[23], which states that AI generated works only qualify for copyright protection, if the element of human intervention exists in it.
Deep fakes and Privacy Rights:
Subsequent to the Puttuswamy [24] judgement, Indian law has explicitly stated that the right to privacy is an inalienable aspect of the right to life under Article 21 of the Constitution. The judgement gave emphasis to the concept of ‘informational privacy’, the same stating that a person must possess control over the dissemination of personal content and any unauthorised use of such information would be regarded as infringement of such privacy.
Furthermore, the Personal Data Protection Bill, also speaks on personal data under Section 3(28)[25], defining it as data pertaining to a natural person, directly or indirectly identifiable. Section 4 of the Act imposes restrictions stating that such personal data is to be processed only for specific, lawful purposes and as per Section 5, in a reasonable manner, devoid of violation of the data principal’s privacy and only for either explicitly consented purposes or in a manner incidental to the consented purpose. Section 20 of this Act also gives attention to the right to be forgotten, which can be exercised by individuals in certain situations, such as when the consent granted by the data principal was subsequently withdrawn, or the initial purpose of collection of data has been fulfilled, etc.
On the international forum, the prevailing data protection mechanism for the EU, the GDPR, under Article 5(1)[26] emphasises on data accuracy, stating that the personal data must be accurate, necessary and if this data is inaccurate, with regard to the purpose of its collection, the same is to be erased promptly. Therefore, the GDPR effectively states that deepfake content representing falsified events are to be erased without any delay. Furthermore, those individuals that constitute the subject of deepfakes also have the added protection under Article 17[27], the right to erasure. This right empowers the individual to ensure the swift deletion of any personal data, with the aid of the Controller, as long as certain requisites under the Article are met.
Therefore, it can be ascertained that free unconsented use of individual’s data in deep fake content would render violations of their right to privacy.
Conclusion:
Deep fakes are still in its growing phase and already presents several issues in various fields such as copyright law, data protection, right to privacy etc. The identification of deep fakes poses their own set of complications. The Deepfake Detection Challenge organised by Facebook, Microsoft and Amazon could only identify them with an accuracy rate of 65.18%[28].
The United States Government, enacted the Identifying Outputs of Generative Adversarial Networks Act, which imposed a duty upon the National Institute of Standards and Technology and the National Science Foundation to conduct research into this technology and to cooperate with private entities to further develop deepfake identification mechanisms[29].
[1] U.S. Laws Address Deepfakes, SECURITY MANAGEMENT, https://www.asisonline.org/security-management-magazine/latest-news/today-in-security/2021/january/U-S-Laws-Address-Deepfakes/#:~:text=For%20the%20second%20year%20in,of%20people%20that%20appear%20authentic (last visited Dec. 9, 2021, 2:43 p.m.).
[2] deepfake AI (deep fake), WHATIS.COM, https://whatis.techtarget.com/definition/deepfake (last visited Dec. 9, 2021, 7:23 p.m.).
[3] Deepfake porn is ruining women’s lives. Now the law may finally ban it, MIT TECHNOLOGY REVIEW, https://www.technologyreview.com/2021/02/12/1018222/deepfake-revenge-porn-coming-ban/ (last visited Dec. 10, 2021, 12:47 p.m.).
[4] Why the Manoj Tiwari deepfakes should have India deeply worried, THE PRINT, https://theprint.in/tech/why-the-manoj-tiwari-deepfakes-should-have-india-deeply-worried/372389/ (last visited Dec. 10, 2021, 12:21 p.m.).
[5] Section 55, Copyright Act, 1957 (14 of 1957).
[6] Id.
[7] Id.
[8] Chancellor Masters and Scholars of The University of Oxford v/s Narendera Publishing House, 2011 (47) P.T.C. 244 (Del).
[9] Section 57, Copyright Act, 1957 (14 of 1957).
[10] Section 107, Copyright Act of 1976.
[11] Copyright and Fair Use, STANFORD LIBRARIES, https://fairuse.stanford.edu/overview/fair-use/four-factors/#the_transformative_factor_the_purpose_and_character_of_your_use (last visited Dec. 11, 2021, 4:23 p.m.).
[12] Seltzer v. Green Day, Inc., 725 F.3d 1170 (9th Cir. 2013).
[13] Patrick Cariou v. Richard Prince, 714 F.3d 694 (2d Cir. 2013).
[14] Bill Graham Archives v. Dorling Kindersley Ltd., 448 F.3d 605 (2d Cir. 2006).
[15] Disney Comes Out Against New York’s Proposal to Curb Pornographic “Deepfakes”, THE HOLLYWOOD REPORTER, https://www.hollywoodreporter.com/business/business-news/disney-new-yorks-proposal-curb-pornographic-deepfakes-1119170/ (last visited Dec. 11, 2021, 7:19 p.m.).
[16] Virginia bans ‘deepfakes’ and ‘deepnudes’ pornography, BBC NEWS, https://www.bbc.com/news/technology-48839758 (last visited Dec. 11, 2021, 11:44 p.m.).
[17] WIPO Conversation on Intellectual Property (IP) and Artificial Intelligence, ASSOCIATION OF AMERICAN PUBLISHERS, https://www.wipo.int/export/sites/www/about-ip/en/artificial_intelligence/call_for_comments/pdf/org_aap.pdf (last visited Dec. 12, 2021, 3:07 a.m.).
[18] Feist Pubs., Inc. v. Rural Tel. Svc. Co., Inc., 499 U.S. 340 (1991).
[19] Section 178, Copyright, Designs and Patents Act 1988.
[20] Section 9(3), Copyright, Designs and Patents Act 1988.
[21] Camlin Pvt. Ltd. v. National Pencil Industries, AIR 1986 Delhi 444.
[22] India recognises AI as author of a copyrighted work, LEXCAMPUS, https://www.lexcampus.in/india-recognises-ai-as-author-of-a-copyrighted-work/ (last visited Dec. 12, 2021, 4:20 a.m.).
[23] Copyright in artificially generated works, AIPPI WORLD CONGRESS – LONDON ADOPTED RESOLUTION, https://aippi.org/wp-content/uploads/2020/05/Resolution_Copyright_in_artificially_generated_works_English.pdf (last visited Dec. 12, 2021, 6:09 a.m.).
[24] Justice K.S. Puttaswamy v. Union of India, 10 SCC 1 (2017).
[25] Section 3(28), The Personal Data Protection Bill, 2019 (Bill No. 373 of 2019).
[26] Article 5(1), General Data Protection Regulation.
[27] Article 17, General Data Protection Regulation.
[28] Facebook contest reveals deepfake detection is still an ‘unsolved problem’, THE VERGE, https://www.theverge.com/21289164/facebook-deepfake-detection-challenge-unsolved-problem-ai (last visited Dec. 12, 2021, 7:54 p.m.).
[29] U.S. Laws Address Deepfakes, supra note 1.