The A.I. Challenge to IP Laws: Examining the ANI v. Open AI Case in Copyright Law
Introduction : The friction between A.I. and Intellectual Property Rights is not exactly new; it has been playing out since around 2018-2019 in Western courts. Those courts have, in various ways, found A.I. to be not eligible as an inventor or as an original maker.
The present issues are like this : In case of Generative A.I., does the use of existing owned and copyrighted data to train these systems count as “fair use” or is it actually a direct breach / infringement of copyright ?
Another question sits right beside it, namely data ownership. i.e., when these AI models spit out content, will that belong to the company which owns the model, or will it belong to the person who typed the prompt. The law seems to be settled for an AI model that cannot own anything on its own - it is not treated as the creator or inventor, as was clarified post the RAGHAV case.
Background to the issue
AI models are trained with already-existing data, so that they can produce outputs according to user prompts. Usually, these systems use crawlers to gather content from across the internet, then feed it into the model. After processing, the information gets converted into data tokens, which are then used to generate the final results corresponding to the prompts. Now, these crawlers can be governed by the model-developing organization, meaning it can set which sites, or which segments of the internet, the model is allowed to crawl to pull data from. The ANI is a large and well known news agency, with multiple reporters and creators, who - after substantial research and consistent work - publish news reports and articles.
OpenAI’s crawlers are also accessing the paid content of the ANI, bypassing the payment mechanism. The ANI has challenged the unauthorised use of its copyright data for the training of these AI models. Furthermore, the AI has generated false content and attributed it to the ANI’s. ANI had requested Open AI to stop its crawlers from scraping data, which it has done while the matter is sub-judice.
Legal Framework In India : Fair Dealing and Jurisdiction
Section 52 of the Copyrights Act, 1957, Fair Dealing provides exceptions to the copyright on certain grounds including the use of copyright data for educational purposes, research etc.
Secondly, the concept of fair dealing is not the same as that of fair use as it is narrower and gives an exception only when the matter comes under the purview of Section 52 of the Copyrights Act, 1957.
Section 62 of the Copyrights Act, 1957 provides for filing of cases by the owners of the copyright for copyright infringements anywhere where they carry on business.
The Questions Raised
Does the storage and use of data by the AI models for the purpose of training their models infringe data.
Whether the generation of output by use of such data amounted to infringement of Copyright.
Whether such data used on a large commercial scale is fair dealing for research or private use under section 52.
Whether Indian courts have jurisdiction as the data used by OpenAI is stored and processed in foreign servers.
Arguments
ANI : OpenAI used and stored ANI’s data without their permission, again circumventing payments to use the paid-for content.
The AI has made up fake data and then falsely attributed that data to the ANI, damaging its reputation and spreading misinformation.
OpenAI®
The jurisdiction of the courts does not matter as the data used and stored is in a foreign country.
The use is fair dealing because the training of AI models is research, and the use of data for research purposes is exempted from copyright infringement, and further there was no material commercial harm.
Provides opt-out functionality, thus the ANI had the ability to opt-out restricting its data to be used for training of the data models.
What the Judgement Entails (Reserved)
If the tokenisation of copyright data is permitted, then the whole copyright would be made available to the AI models, thus allowing for its commercial use, and thereby causing losses to the creator. Then the copyright data can be reproduced by converting the copyright data into numbers. This creates another exception for the use of copyrighted materials.
If the usage by OpenAI is considered to be an infringement, it will block the AI training models for good-quality data sets for its training. This could increase the instances of it hallucinating and providing false information. It will also provide a framework where training models will require paid subscriptions to high-quality data or result in the adoption of revenue-sharing models.
Hallucinations: If this is treated as the responsibility of the AI company, it will invite claims against them in future. If it is not treated as their responsibility, there will be emanation of false use without any liability of the AI training companies.
Whether data tokenisation as a process is technically copyright infringement or not, the view here is divergent among the Amici Curiae as well, while one considers it no different from human learning, therefore should be exempted, while other refers tokenization as reproduction of the data, therefore amounts to infringement.
View of the Amicus Curie
The data in question is generated in India and therefore the issue is squarely within the jurisdiction of Indian courts.
Whether data tokenisation as a process is technically copyright infringement or not, the view here is divergent among the Amici Curiae as well, while one considers it no different from human learning, therefore should be exempted, while other refers tokenization as reproduction of the data, therefore amounts to infringement.
The ownership of data is also questioned for ANI, being a news agency it does not create data, it just reports what is already created or other events, the data that is already known. So, ANI has to prove its data ownership.
Author: Anshul Agarwal in case of any queries please contact/write back to us via email to content@khuranaandkhurana.com or at Khurana & Khurana, Advocates and IP Attorney
References
The Copyrights Act, 1957
The Delhi High Court website
Bar and bench
Live law
Comments