Fixing AI’s Original Sin: Methods and Approaches

The rise of artificial intelligence (AI) has brought significant advancements, but it has also sparked intense debates over copyright law and the ethical use of content. Recently, The New York Times highlighted concerns that tech giants like OpenAI and Google might be transcribing YouTube videos to leverage the text for AI training. This move has raised questions about whether these companies are infringing on copyrights, as it potentially violates YouTube's terms of service and copyright law. This issue is perceived as controversial and has been referred to by some experts as “AI’s Original Sin.”

OpenAI and Google are reportedly using vast amounts of content, including YouTube videos, to train their AI models. The controversy hinges on whether these tech companies are violating copyright by using content without explicit permission from creators. Meta (formerly Facebook) has acknowledged the challenge, suggesting that its AI models might fall behind if they do not adopt similar practices. This acknowledgment underscores the competitive pressures driving these companies to push the boundaries of copyright law.

In an interview with The New York Times podcast The Daily, reporter Cade Metz and host Michael Barbaro discussed the ramifications, framing copyright infringement as a fundamental ethical issue for AI. The conversation emphasized that copyright disputes are central to the broader conflict over who benefits financially from generative AI technologies. Katherine Lee, A. Feder Cooper, and James Grimmelmann, in their essay “Talkin’ ’Bout AI Generation: Copyright and the Generative-AI Supply Chain,” argue that it is crucial to look beyond legal liability to the political economy of copyrighted content in AI. They suggest that the focus should be on how value created by AI should be distributed fairly among those who contribute to its creation.

Publishers, including The New York Times, argue that AI-generated content competes directly with original works, undermining their business models. They claim that AI-generated summaries of news articles can replace the need to read the original pieces, reducing traffic and revenue for publishers. Consequently, these companies want compensation for their content being used to train AI models. On the other hand, AI developers like OpenAI and Google argue that the sheer volume of data required to train effective models makes it impractical to obtain licenses for every piece of content. They believe that broad access to data is essential for developing sophisticated AI systems. Some experts, like Sy Damle from Andreessen Horowitz, argue that restrictive data access could hinder AI advancement.

Copyright law grants creators exclusive rights to profit from their work, protecting unique expressions rather than facts or ideas. However, AI's use of copyrighted content raises complex issues about fair use. Typical examples of fair use include quotations, criticism, and summaries, which allow limited use of copyrighted material without permission. However, AI-generated summaries might exceed what is traditionally considered fair use, especially if they replace the need to access the original content.

From a practical standpoint, AI developers and content creators must navigate the balance between protecting intellectual property and fostering innovation. AI models trained on high-quality content, such as professionally written articles and books, are more valuable. Hence, developers seek unfettered access to this content, despite potential copyright infringements. Conversely, if the creation of new high-quality content is compromised by unlicensed use, the ecosystem that AI relies on could be jeopardized. Ensuring that content creators are fairly compensated is essential to maintaining a sustainable supply of new material for AI training.

To address these challenges, several approaches could be considered. AI developers could establish clear and fair licensing agreements with content creators, ensuring that they are compensated for the use of their work. This would involve negotiating terms that reflect the value of the content and the benefits it provides to AI systems. Ensuring that AI-generated outputs properly attribute the original sources could help maintain the visibility and recognition that content creators seek.

This approach could involve developing standards for AI systems to cite sources accurately. Clarifying and potentially expanding fair use policies to accommodate the unique needs of AI training could provide a legal framework that balances the interests of all parties. This might involve legislative changes or new guidelines from copyright authorities.

Encouraging collaboration between AI developers and content creators could foster mutually beneficial relationships. For instance, media companies might partner with AI firms to develop new tools and technologies, sharing in the profits generated by AI-enhanced content. AI companies should recognize the importance of sustaining the sources of high-quality content. Investing in the future health of content creation—whether through direct funding, partnerships, or other means—could help ensure a continuous supply of material for AI training.

The debate over AI’s use of copyrighted content highlights the tension between technological advancement and intellectual property rights. While AI developers argue that broad data access is essential for progress, content creators fear that unlicensed use undermines their livelihoods and the quality of future content. To move forward, it is crucial to develop frameworks that balance these competing interests.

Transparent licensing agreements, fair use policies, and collaborative models can help ensure that both AI systems and content creators thrive. By recognizing the value that each party brings to the table, it is possible to create a sustainable ecosystem where innovation and creativity coexist. As AI continues to evolve, the resolution of these copyright disputes will shape the future of content creation and consumption. Ensuring that all stakeholders benefit fairly from the generative AI supply chain will be key to fostering a vibrant and dynamic digital landscape.

The recent protest by longtime Stack Overflow contributors who don’t want the company to use their answers to train OpenAI models highlights a further dimension of the problem. These users contributed their knowledge to Stack Overflow; giving the company perpetual and exclusive rights to their answers. They reserved no economic rights, but they still believe they have moral rights.

They had, and continue to have, the expectation that they will receive recognition for their knowledge. It isn’t the training per se that they care about, it’s that the output may no longer give them the credit they deserve. The Writers Guild strike established the contours of who gets to benefit from derivative works created with AI. Are content creators entitled to be the ones to profit from AI-generated derivatives of their work, or can they be made redundant when their work is used to train their replacements? As the settlement demonstrated, this is not a purely economic or legal question but one of market power.

There are three parts to the problem: what content is ingested as part of the training data in the first place, what outputs are allowed, and who gets to profit from those outputs. Accordingly, AI model developers ought to handle copyrighted content by mandating transparency about the content and source of training datasets—the generative AI supply chain—would go a long way towards encouraging frank discussions between disputing parties. Focusing on examples of inadvertent resemblances to the training data misses the point.

Generally, whether payment is in currency or in recognition, copyright holders seek to withhold data from training because it seems to them that may be the only way to prevent unfair competition from AI outputs or to negotiate a fee for use of their content. As we saw from web search, “reading” that does not produce infringing output, delivers visibility (traffic) to the originator of the content, and preserves recognition and credit is generally tolerated. So AI companies should be working to develop solutions that content developers will see as valuable to them.