Can Copyright Tame Generative AI?

Artificial intelligence has become a household name since late 2022, when OpenAI’s ChatGPT was first released. Prior to that, artificial intelligence had also been utilized in various random applications. In 2016, Daddy’s Car, an AI-generated song emulating the Beatles, was proudly released to fanfare news that its machine producer could make any song using a database of thousands of available songs. Later, in 2018, a painting produced with AI and initially valued at 10,000$ was sold in an auction for 432,500$ in less than seven minutes.

But it was in 2023 that the AI phenomenon took both experts and the public by surprise. No day has passed by without some AI-related news. A prestigious literary prize in Japan was awarded to a novel later revealed to have been co-written with AI. Deepfakes, employing the voice or faces of celebrities have been all over the place. And there seems to be no end to all this. On a larger scale, some blissfully reckon AI echoes the next dot.com boom, the digital Industrial Revolution, a utopia, while others woefully think the new technology might cost them their job and livelihood in a Brave New World.  

It is now widely known how AI operates for producing creative content. Its unbridled prowess feeds off the content available on the internet, for free until very recently. The chatbot is trained by its algorithm to plough the internet to combine features of existing material in the way a user’s prompt has asked for in a matter of seconds.

Many people have tried their hand at using AI to produce songs, music, images, and other works requiring creativity, under the impression that they have suddenly been bestowed upon some marvellous gift which literally heralds The Death of the Author. This thing can write jokes like seasoned satirists, can write novels like a life-battered writer, and can paint like a painter who’s transformed a thousand thoughts into one strike of a brush on the canvas. Then, the bell began to toll. The joy and bewilderment on the part of users was soon translated into the perception of threat on the part of human creators of art.

AI Firms Face Copyright Infringement Lawsuits

Alarmed by the threat of AI, individual creators decided to strike back. Screenwriters in Hollywood staged a successful five-month strike in May 2023 over their royalty payments and the use of artificial intelligence in their industry. Copyright lawsuits against AI companies started to pile up. Meta was taken to court by authors Michael Chabon et al. Stock photo provider Getty Images sued Stability AI, accusing it of misusing 12 million Getty images to train its image generation system.  The New York Times took OpenAI and Microsoft to court, claiming they had used millions of its articles to train their chatbots, which now competed with the newspaper. Visual artists Jingna Zhang, Sarah Andersen, Hope Larson and Jessica Fink went after Google, while authors Brian Keene, Abdi Nazemian and Stewart O’Nan initiated an action against Nvidia. In yet another case, Universal Music Group, ABKCO, Concord Music Group, and other music publishers filed a lawsuit against AI giant Anthropic for “unlawfully” exploiting their copyrighted song lyrics to train AI models. The list of copyright lawsuits against AI firms goes on and on. The verdict on any of these cases would shed light and draw boundaries on what and how things related to AI can be done when it comes to copyrighted material.

AI Firms Fight Back: Arguments Against Copyright Infringement Claims

AI companies have all kinds of arguments against claims of copyright infringement. When the US Copyright Office opened comments on August 30, 2023, about the use of copyrighted content in training AI models, major AI firms rushed to make their case. Here’s some of the arguments:

— Meta contends that ‘the use of copyrighted content to train AI models is a non-consumptive use that does not rigger the rights protected by copyright, and even if it did, it would be a fair use.’ It adds that ‘model training is squarely protected by the fair use doctrine, whose central purpose is to “avoid rigid application of the copyright statute when … it would stifle the very creativity which that law is designed to foster.’ See Meta’s full comment here.

— Microsoft argues that changing the current copyright law would impede innovation and in particular harm the progress of start-ups which may not have sufficient resources to obtain licenses from rights holders. See Microsoft’s full comment here.

— Anthropic urges lawmakers not to amend the current law, holding that appropriate limits to copyright ensure and support creativity, innovation and other values. See Anthropic’ s full comment here.

— StabilityAI refers to how other jurisdictions are treating data mining by AI, contending that for an innovative AI industry, mining of available data, text, and images as input should be allowed and that existing copyright protection should be applied to the output of AI. See StabilityAI’s full comment here.

— Google argues that ‘the act of “knowledge harvesting” like the act of reading a book and learning the facts and ideas within it, would not only be non-infringing, but it would also further the very purpose of copyright law.’ See Google’s full comment here.

AI Copyright Amendment in Europe & US

The first successful regulatory attempt aimed at AI came in March 2024. Lawmakers in the European Parliament adopted the landmark Artificial Intelligence Act that ensures compliance with fundamental rights while boosting innovation. The parts that relate to copyright are as follows:

— Any use of copyrighted material by general-purpose AI models requires the consent of the rightsholder. If the text and data mining is for scientific research, such authorization is not required.

— General-purpose AI providers should put in place a policy to comply with the EU copyright law regardless of the jurisdiction which imposes copyright obligations for the training of their AI model. This ensures fair competition.

— General-purpose AI models using copy-righted or other content for the training of their model are required to make a summary of the content publicly available. This boosts transparency by allowing rights holders to enforce and exercise their rights.

— The AI Office of the EU should ensure the fulfilment of obligations by AI providers “without verifying or proceeding to a work-by-work assessment of the training data in terms of copyright compliance.”

— While verifying compliance with the new AI regulation, simplified ways should be offered to SMEs, including start-ups, not to impose an excessive cost on them or discourage them from developing AI models. This encourages innovation.

Lawmakers in the US soon followed suit. A bill called Generative AI Copyright Disclosure Act was introduced in April 2024 in the House of Representatives. If passed, it would require AI firms to disclose the list of copyrighted material they have used to train their models. The law would apply to both future and past AI models. According to this legislation, AI firms would have to submit to the Copyright Office a list of copyrighted works they have used to train their model within 30 days before releasing it. Any firm failing to comply would face a financial penalty that would be determined based on the company size and its non-compliance history.

Some AI Firms Agree To Pay For Content

As regulation zeros in on AI firms, they have started to shift their focus to competitively ink deals with major media and publishing companies to use their content for training their models. Market leader OpenAI has recently signed multimillion-dollar partnership deals with dozens of companies like The Atlantic, Vox Media, Axel Springer, Le Monde and Prisa Media. Apple is trying to catch up by offering multi-year deals to license archives of news companies, betting its huge global consumer base would give it a fair share of the new market.

These partnerships are crucial for the development of AI technologies as they provide access to a vast amount of high-quality content necessary for training robust models. Such agreements also offer media companies a new revenue stream, addressing concerns about fair compensation for the use of their intellectual property.

Conclusion

There is no doubt that the ground-breaking technology of AI will create astounding changes in every aspect of human life. While the transition is taking place, copyright is trying to bring some method to the madness. In addition to heated discussions about the use of copyrighted content for training AI models, two other matters need to be addressed: AI copyright liability and whether AI-generated material can be copyrighted without human involvement. All eyes will be on the US in the coming months where the Copyright Office has collected more than 10,000 comments on these three issues and is planning to release a report and major rulings on infringement cases are awaited.

The details about the lawsuits mentioned in this article are based on their latest status as of 18.07.2024. New developments might have happened since then.