Who Owns AI? Reimagining Creative Value and Ownership in AI, Inspired by Amy Whitaker's Analysis

Amy Whitaker’s working paper, “Who Owns AI?”, explores the complex intersection of artificial intelligence (AI) and the creative industries, focusing on who should rightfully own and benefit from the outputs of AI. The paper examines recent lawsuits by authors like Sarah Silverman and John Grisham against AI companies, including OpenAI and Meta, accusing them of copyright infringement. The core of these legal challenges is the claim that AI companies have used creative works without permission as training data for their language models, ultimately deriving value from contributions they did not create. Whitaker uses these cases as a lens to question the broader implications of AI on creative ownership, governance, and economic value.

Whitaker highlights that AI companies like OpenAI rely on vast datasets, often containing copyrighted materials, to train their models. This raises critical questions about whether creators should receive compensation when their work becomes part of these training sets. The lawsuits argue that using entire creative works without explicit consent breaches copyright law, a point that Whitaker uses to illustrate the need for reevaluating traditional understandings of fair use. As the paper explains, fair use in the United States is determined by four key factors, including the purpose of the use, the nature of the original work, the amount of the work used, and the impact of the use on the original market. With AI's ability to process vast amounts of data, determining whether this constitutes transformative use or infringement becomes increasingly complex.

One of Whitaker's key contributions is her reimagined framework for understanding "value" in the context of AI. While AI companies focus on the functionality and utility derived from training data, creators emphasize the inherent value of their original work. Whitaker proposes a "market and value test" that could provide a more nuanced approach to balancing these interests. This shift from a strict market impact analysis to one that recognizes the intrinsic value of creative contributions aims to ensure that creators receive recognition and compensation for their work, even when it forms a small part of a larger AI dataset.

Beyond the legal nuances, Whitaker's paper also delves into broader questions of how society organizes and values creative work in the digital age. Historically, large technology companies like Google, Meta, and OpenAI have leveraged centralized business models where user data becomes a significant asset. Whitaker contrasts this with alternative models, inspired by blockchain and cooperative economics, that envision more equitable frameworks for data and value distribution. She suggests that these models could offer a more collaborative approach to data governance, giving creators greater control over how their contributions are used.

Whitaker uses OpenAI’s own structure as an example of the tensions between public benefit and private profit. Founded as a nonprofit with the mission to develop AI for the betterment of humanity, OpenAI later created a for-profit entity to attract the necessary investment to sustain its work. Whitaker points out that while this model aims to direct excess profits back to the nonprofit, the decisions on how those funds are used remain concentrated among a select group of directors. This raises questions about democratic input and the role of broader societal stakeholders in determining the use of these funds, especially when the underlying value comes from a diverse range of creative contributions.

In the case of Sarah Silverman’s lawsuit, Whitaker explores the specifics of the legal debate. Silverman argues that OpenAI used her entire book without consent, challenging how much of an individual’s work can be used in AI training without violating copyright. The court’s decision to allow Silverman to pursue her claims of direct copyright infringement underscores the uncertainty in interpreting fair use in the context of AI training. Whitaker suggests that while summarizing a work might be considered fair use, using entire texts to train an AI model crosses a legal and ethical line that requires careful scrutiny.

Whitaker's paper suggests that a purely legal response may be insufficient to address the deeper questions of value and ownership in AI. She advocates for new cooperative models that align with the decentralized nature of AI development. Such models would directly involve creators in decision-making processes, allowing them to share in the economic benefits derived from their work. This approach calls for a fundamental rethinking of the relationship between creators and AI companies, emphasizing collaboration over competition and shared ownership over centralized control.

The economic implications of Whitaker’s proposals could be profound. If courts were to side with creators, AI companies might be required to secure licenses for the data they use, similar to traditional media organizations. This could lead to new systems for compensating creators, akin to royalty mechanisms in the music industry. Such systems would recognize the contributions of creative inputs and ensure that those who help build AI training datasets are appropriately rewarded.

However, Whitaker emphasizes that this debate extends beyond economic considerations to questions of how we value and incentivize creative labor. She argues that creators, much like investors, take risks when producing new work, and they should be entitled to a share of the rewards generated by AI models that build upon their contributions. This perspective challenges the notion that AI development should remain dominated by large tech corporations, suggesting instead a vision where the benefits of AI are more widely shared.

This blog post is intended for informational purposes only. It does not constitute legal advice, and readers should consult a qualified legal professional for guidance specific to their situation. The accuracy of this content is based on the original paper and sources available at the time of writing and may be subject to change.