“There should not be any copyright protection over work created used artificial intelligence, provided the work is generated autonomously by an AI algorithm — meaning that you cannot trace a human linkage to the creation of that particular work,” Arul George Scaria, associate professor at NLU Delhi said. “I don’t see any philosophical or economic justification for granting IP rights over that particular work, be it patterns or be it copyright,” he added. However, Aamod Gupte, legal head at ZEEL, contended that works created using AI should not be kept out of the Copyright Act’s purview.

Scaria and Gupte were speaking at MediaNama’s discussion on the proposed amendments to the Copyright Act, for which the Indian government has begun private consultation. The Act was last amended in 2012, when the internet and multimedia landscape was wildly different. Held on October 30 with support from Netflix, Amazon, and Facebook, the discussion looked at whether AI-generated work should have copyright protection, and whether game streamers can claim any kind of copyright of the streamed content, among other things.

Copyright over algorithmically generated work

In the case of the GPT-3 (Generative Pre-trained Transformer 3), there is no clear linkage between what the original creator of the algorithm has intended for it to produce and the actual output generated. And so, there should not be any copyright attributed to it, said Divij Joshi, an independent researcher. In this case, “you are not trying to protect some integral property right of the AI,” and while a case can be made for the economic interest of the AI’s creator, it doesn’t necessarily have to deal with copyright.

“The moral and economic justifications for copyright have been thought about for multiple centuries now and autonomously generated works don’t really fit within that,” Joshi added. Scaria also gave the example of the Monkey Selfie case in the US, where the courts clarified that only human beings can become the authors — “So, I think that should actually give us some light on whether AI can be an author,” Scaria said, and added that even under Indian copyright law “only a human being can become an author of a particular work”.

  • Who owns the right over AI-generated work? In the context of the GPT-3 authored op-ed published by the Guardian, and whether it should have copyright protections, Scaria said that human editors compiled a number of variations produced by an AI algorithm to come up with a better version, and “at least there is some sort of a human linkage there”. However, even a scenario where the AI can autonomously generate the complete report is possible, he cautioned.

“The Guardian thing is actually heavily edited. GPT-3’s actual text is really ridiculous and makes no sense most of the time. It won’t really be considered worthy of creativity or attributing creativity or agency to it [AI]. In the Guardian article the editors would to some extent be able to own copyright over [the piece].” — Divij Joshi

  • What kind of licenses does one need to feed in data to AI systems? Given the kind of algorithmic models that exist today, it’s easy to spot how an infringing piece of work could have potentially been created using AI, Joshi said. “Say someone produces a Monalisa, the only way that can happen right now is if somebody fed the actual Monalisa into a system. But in order to feed that data, you would need a license to use the Monalisa in a way which can develop an algorithmic system based on it, or you have at some level infringed copyright — unless you had some kind of Text and Data Mining (TDM) exemption,” he added (read about TDM exceptions here). He also questioned whether we should have separate copyright regimes for feeding modified training data into machine learning systems:

“One good example of this is, when you look at creative commons licenses on photographs that are used for facial recognition images, that’s a big controversy, primarily because it mixes up issues of data protection, privacy, and copyright,” Joshi added.

However, not everyone seemed to agree that AI-created work should be kept out of the Act’s purview. ZEEL’s Aamod Gupte said that doing that would be a “dangerous thing”. “UK has a view that it’s on the creator, or someone who has funded that AI initiative, to claim ownership over the work created by those algorithms,” he added. Gupte also suggested that data mining should not be kept out of the purview of copyright, and it is “best to have it within that, but with sufficient parameters to control that,” given that data mining also has privacy implications.

Can gamers be considered performers?

Game streaming is when a person plays a game live for an audience, while usually making additional commentary on the game. However, Ikigai Law’s Aman Taneja said that a lot of the streams could be constrained by the license terms of the game publishers, even though the streamers are creating a completely new work as they go along. A debate also ensued on whether gamers can be considered performers under the current Copyright Act, which defines performers as an actor, singer, musician, dancer, acrobat, juggler, conjurer, snake charmer, a person delivering a lecture or any other person who makes a performance. A performance, in turn, means any visual or acoustic presentation made live by one or more performers.

Taneja said that in the course of playing a game, the streamers could potentially be viewed as performers, and in that sense, the recognition of the rights of gamers might be something which is completely new and worthy of giving some considerations.

“The game play footage itself may not be worthy of being called performance because it is not really audio or visual presentation in that sense that is the requirement end of the day [under the Copyright Act]. But certainly when you add a walk through or talking through how a gamer is playing — that combined package can certainly qualify to be a visual presentation which is within the definition of performance.” — Ankit Relan, lead counsel, Facebook

On the other hand, what a player is actually doing is moving the characters — they are not performing any work which is getting recorded, and neither are they the producer according to the definition in the Copyright Act,” said Rahul Ajatshatru of the Ajatshatru Chambers, weighing in on whether game players can be considered performers. Additionally, once a player clicks and installs a game, they are bound by the end user licenser agreement of the gaming company, Ajatsharu added.

Ajatshatru further explained:

“You need to perform the work which is the original underlying work [to be considered a performer], which includes choreography, singing, acting before the camera, or performing a stunt. [In the case of gaming], you are not yourself the performer, who is being captured in the film — you are controlling one character and what ever is happening around him is already programmed by someone else to behave in a certain way. If you are playing Call of Duty, and have a bazooka using which you are hitting a chopper, it is programmed to fall down on impact. So someone else has decided that, what you are doing is you are carrying a gun and walking around and engaging with the environment.”

Relan also contended that while gamers are bound by end user agreements, it should not be the only point to be considered as to why they can’t claim ownership over a game stream. “There is this element of transformative use that is being questioned in lot of game play footage, in lot of walk through videos in a sense where the predominant purpose of the video that is being put out by a gamer is not so much the use of the underlying copy of the game, but to put out a kind of transformed work itself,” he added.

Also read: