Introduction
June 2024 saw not only extreme heat in reality, but also extreme heat in reels! Instagram changed its Privacy Policy (‘Policy’) and now uses the information you provide on the platform to train their Artificial Intelligence (‘AI’) model. However, you can restrict your content to be used as training data for the AI model, but it is an opt-out system.[1] This means that the default setting is such that your content will be used for AI training, and if you don’t want this to happen, you will have to change the settings. The implication of the opt-out system is that many users, unaware of the change, will continue to provide content for the training database.
Moreover, the opt-out choice will not be operative on the data that you have already provided.[2] This means that the previously uploaded content could still be used for AI training, even if you go for the opt-out option. According to Instagram’s Policy, the information users provide is used to improve products by customization of feed, personalization of features, giving recommendations and showing ads relevant to the user.[3] This list is not exhaustive and are only some uses explicitly mentioned in the Policy.
When you agree to Instagram’s Terms and Conditions, you actually grant it a “royalty-free, transferable, sub-licensable, worldwide license.”[4] As it is evident from the terms used, this license grants immense rights to Instagram. Another interesting point to note is that even if you are not a user of Instagram, your data could still be used for training if it is available there. The Policy clearly states,
“Even if you don’t use our Products and services or have an account, we may still process information about you to develop and improve AI at Meta.” [5]
This means that not only those individuals who have unknowingly granted license, but also those individuals who have never granted a license will be prey to this changed Policy. This means that Instagram could use the work of a non-user, disregarding the fact that the author, who is also the first owner[1][6] has not granted any license to Instagram.
Is Using User Data for AI Training Covered Under Fair Use Doctrine?
The first thing to note here is that we do not apply fair use doctrine in India, instead we use fair dealing mentioned in Section 52 of the Copyright Act (‘the Act’).[2][7] These two principles are not the same, however, Indian Courts have borrowed principles of fair use from UK and USA, therefore the researcher is taking into consideration the fair use doctrine.[8]
Before addressing the question of fair use, we must understand a little of how AI is trained. For designing an AI model, a training dataset is needed which consists of millions of works, which are often downloaded from the sources.[9] In this case, Instagram intends to use publicly available works, licensed works and user information as training data.[10] It has been contended that since millions of works are used to train the AI, the output of the AI is not substantially similar to the training dataset,[11] therefore, falling under fair use.[12]
However, Jenny Quang observes that in order to train AI models, developers download data from different sources. This downloaded data is nothing but copies of the original work. Although the Act in India[3][13] does not define the term ‘copies’, the US Copyright Law defines ‘copies’ as:
“material objects… in which a work is fixed by any method now known or later developed, and from which the work can be perceived, reproduced, or otherwise communicated, either directly or with the aid of a machine or device.”[14]
She argues that by downloading data, we are fixing since it is no longer transient in nature, so the downloaded data qualifies as ‘copies’. Moreover, Meta’s Policy allows storage of training data for ‘as long as’ they ‘need’ it. Therefore, even if the output of the AI model does not infringe copyright per se, the mere download of copyrighted data to train AI model is an infringement of copyright.[15]
Such download of data for training AI is an infringement only when the works are copyrightable and no license has been obtained. Unfortunately, in the case of Instagram, all users have, knowingly or unknowingly, granted license to Instagram.[16]
Can Instagram use our data to train AI?
Before answering this question, we must peruse Section 18 of the Act.[17] The direct implication of the second proviso is that if a license has been granted for some specific ‘modes and means of exploitation’, the assignee cannot extend it to other ‘modes or means of exploitation’ that were not specifically mentioned during the time of assignment.
All those individuals who had granted license to Instagram before June 2024 did not grant a license for use of their content for AI training. Although the users were informed of this update and were given an opt-out choice, the content which has already been posted prior to this Policy will still be used for training.[18] Simply put, the new Policy is being applied retrospectively, but the safeguard against this Policy does not have any retrospective application.
The researcher would like to clarify why the statement that the AI Policy is retrospectively applicable is made. This is because when you go through the Policy, it provides you an option to “object to your information being used for AI at Meta.” Here, the Policy states:
“You have the right to object to Meta using the information you’ve shared on our Products and services to develop and improve AI at Meta.”[19]
This means that you need to object to the “information you’ve shared” in order to safeguard your information being used for AI training. Now, the “information you’ve shared” means all the information you have ever shared with Instagram, which includes information you shared prior to June 2024. Failure to object will lead to continued usage of your information.
Moreover, this “right to object” is subject to Instagram’s discretion. It is an arbitrary clause which makes the whole “right to object” redundant as the final say rests with Meta itself. It clearly says,
“We may still process information about you to develop and improve AI at Meta, even if you object or don’t use our Products and services.”[20]
The mechanism will also review the request with respect to applicable data protection laws and the request will be moving forward only if it is honoured.[21]
Technically speaking, Instagram does not have license to use works posted prior to June 2024 for the purpose of training its AI model. So, use of such works for AI training is basically an illegitimate extension of ‘modes and means of exploitation’,[22] thereby infringing the copyright of the content creators.
Public good and fair use
The bare reading of Section 52 of the Act[23] is enough to tell us that the use of user content to train AI model is not fair dealing. When we talk about fair use, one main component of it is to ensure public interest and non-commercial use. [24]
Considering public interest is important because conventionally, the copyright holder held more bargaining power than the one claiming fair use.[25] Not allowing fair use in such cases would be prejudicial to public interest as it would stifle education, research, development and further creativity.
However, in this specific case of Instagram, more bargaining power rests with Instagram than the copyright holders.[26] Copyright holders are also the consumers, who have lesser bargaining power.[27] Here, we can observe a swap in the distribution of bargaining powers. Very evidently, the use of user content to train AI models is not in the public interest, if not detrimental to it. This was reflected in various news articles, which highlighted consumer concerns and the unsupportive nature of opt-out option.[28]
The second crucial component for fair use is non-commercial use. Instagram is not a non-profit organization. The major stream of revenue for Instagram is advertisements.[29] Now, Instagram’s Policy states that it uses user information to personalize ads for them. This is basically targeted advertising.[30]
“Firms live and die by their ability to successfully bring innovations to market, which places commercialization as a key capability.”[31]
In line with the above quote, it would be odd if Meta brings the AI feature, which is an innovation, without the intent to commercialize it. Since the data Instagram collects for training its AI models would improve the accuracy of targeted advertisements, thereby increasing its revenue, this way, user content is being commercialized.[32] It gives us all the more reason to rule out Instagram’s conduct from the ambit of fair use.
Conclusion
Utilizing user content by Instagram for AI training is a hot topic and brought multiple questions to our plate. Arguments regarding use of copyrighted material for AI training being fair use pose a threat to public interest and thereby, erode the basic idea behind the application of fair use doctrine.[33] The researcher has shown why Instagram’s conduct to utilize user data for training AI is not protected by the shield of fair use doctrine. The right to object is nothing but a shroud over the reality that users have no option but to either stop using Instagram, or succumb to their AI policy. What Instagram is doing is a sheer infringement of users’ copyrights, since the users have not given license to use their content posted prior to June 2024 for AI training.
[1] Meta, ‘Privacy Policy’ (Meta Privacy Center’, 26 June 2024) <https://privacycenter.instagram.com/policy/> accessed 11 September 2024
[2] Ibid 1
[3] Ibid 1
[4] Ibid 1
[5] Ibid 1
[6] The Copyright Act 1957, s 17
[7] The Copyright Act 1957, s 52
[8] Ayush Sharma, “Indian Perspective of Fair Dealing under Copyright Law: Lex Lata or Lex Ferenda?” (2009) 14 JIPR <Indian Perspective of Fair Dealing under Copyright Law: Lex Lata or Lex Ferenda? (manupatra.in)> accessed 11 September 2024
[9] Jenny Quang, “Does Training AI Violate Copyright Law?” (2021) 36(4) Berkeley Tech. L.J. <https://doi.org/10.15779/Z38XW47X3K> accessed 12 September 2024
[10] Ibid 1
[11] Folsom v. Marsh [1841] 9 F. Cas. 342
[12] Ibid 9
[13] The Copyright Act 1957, s 2(m)
[14] Copyright Law of the United States, s 101
[15] Ibid 9
[16] Ibid 1
[17] The Copyright Act 1957, s 18
[18] Ibid 1
[19] Ibid 1
[20] Ibid 1
[21] Ibid 1
[22] Ibid 17
[23] Supra 2
[24] Wendy J. Gordon & Daniel Bahls, “The Public’s Right to Fair Use: Amending Section 107 to Avoid the “Fared Use” Fallacy” (2007) Utah L.R. 619 <https://hdl.handle.net/2144/22956> accessed 12 September 2024
[25] Nancy S. Kim, “Bargaining Power and Background Law Bargaining Power and Background Law” (2019) 12 (1) Vanderbilt J. Ent. Tech. Law <https://scholarship.law.vanderbilt.edu/jetlaw/vol12/iss1/3> accessed 12 September 2024
[26] Ibid 25
Authored by: Ms. Dev Astha
She is a third year law student at the Indian Institute of Management Rohtak. She has a keen interest in Intellectual Property Law, Competition Law, and Energy Law. She believes that abstract ideas are the foundation of innovation. Having a science background, she has a particular interest in Patents. She strongly believes in the 4Cs (foresees): Competence, Confidence, Curiosity and Conscience.
Leave a Reply