Photo credit: www.publishersweekly.com

Addressing the Challenges of Generative AI in Copyright Law

A comprehensive forum was held in Washington, D.C. on Tuesday, focusing on the implications of generative AI for copyright law and the creative sector. This event, titled “The Story Starts With Us,” was co-organized by the Association of American Publishers and the Copyright Alliance. During the daylong session, several industry veterans shared insights from their AI experiences and how these informed their mission to develop ethically-trained AI models.

Ed Newton-Rex, founder of the U.K.-based nonprofit Fairly Trained, emphasized the need for licensing creative works utilized in training AI systems. He voiced concerns that the prevalent practices not only jeopardize the incomes of creators but also undermine societal welfare.

Newton-Rex reflected on his extensive 14-year career in generative AI, remarking on the breathtaking capabilities of the technology while criticizing its reliance on the unlicensed appropriation of creators’ works. “The vision behind generative AI is remarkable, but appropriating the labor of creators to build these systems is deeply problematic,” he stated.

He pointed out that although AI companies invest significantly in engineering and computational resources—spending up to $1 million per engineer and $1 billion per AI model—many still expect to obtain essential training data without compensation. This data is often derived from the extensive and labor-intensive outputs of artists, musicians, writers, and other creators.

Newton-Rex criticized many AI firms for not licensing the majority of their training data, opting instead for web scraping methods to harvest content. According to a study by the Mozilla Foundation, approximately 64% of large language models developed between 2019 and 2023 were trained on datasets that included copyrighted content.

“Prominent AI companies—ranging from OpenAI to Anthropic and music generator Suno—have adopted this method of training that relies on copyrighted material,” he highlighted. This growing trend is concerning as it poses a direct competitive threat to the very creators whose works are being used.

To illustrate the consequences, he cited anecdotal evidence from filmmakers and artists. Some filmmakers openly declared their intent to incorporate AI-generated music into projects, while artist Kelly McKernan experienced a 33% drop in income after her work was incorporated into training data for the AI image generator Midjourney.

Additional findings from Upwork indicated a decline in job postings for writing roles by 8% following the launch of ChatGPT, with 18% fewer listings for more intricate writing tasks. Graphic design opportunities also fell by 18% in the wake of AI image generation technologies.

Despite AI companies’ arguments favoring fair use, Newton-Rex contended that these assertions cannot justify the widespread exploitation of creative works, which ultimately creates models that compete against the original creators.

He further distinguished between AI training and traditional human learning, asserting, “Artists have always drawn inspiration from each other, which is a fundamental aspect of the creative community. However, generative AI companies indiscriminately scrape content from the internet, often disregarding the wishes of the original creators, resulting in a fierce competitor created at their expense.”

Newton-Rex also raised the point that beneficial uses of AI for society, such as scientific breakthroughs, rely on entirely different kinds of data. “For instance, AlphaFold, which focuses on protein structures, was trained on non-creative data. There haven’t been any notable discoveries from AI models trained on artistic works,” he stated.

Certification of Ethically Trained AI

The motivation behind founding Fairly Trained was to create a certification process for AI models that do not exploit copyrighted works without authorization. Newton-Rex’s initiatives have garnered the attention of several industry leaders, including Fairly Trained advisor Maria Pallante, CEO of the AAP. He also highlighted his leadership role in Stability AI, where he contributed to developing Stable Audio, a platform that utilized only properly licensed music—earning recognition as one of Time magazine’s best inventions of 2023.

He raised alarms about the detrimental effects of illegal training practices, warning that they not only harm creators but also undermine the integrity of the open internet. A study revealed that among the top 14,000 websites used for AI training, the proportion of sites with restrictions due to terms of service increased from three percent to as much as 33% within just a year.

Newton-Rex voiced concern that this trend is closing off the web, adversely affecting both emerging AI models and society at large, including researchers and consumers who benefit from an open internet. Notably, some creators have chosen to cease online publishing out of fear that their work will be appropriated by AI companies, contradicting the very intent of copyright law, which is to incentivize creators.

Public sentiment appears to align with Newton-Rex’s viewpoints against unauthorized AI training. A survey conducted by the AI Policy Institute found that 66% of respondents opposed allowing AI companies to utilize publicly available data without consent, while 74% support compensating rights holders for such use.

To exemplify the unity of the creative community, Newton-Rex presented the Statement on AI Training, which has amassed nearly 50,000 signatures from various creators, including renowned figures like author James Patterson and Radiohead’s Thom Yorke. The statement articulated a clear message: “The unlicensed utilization of creative works for training generative AI constitutes a significant and unjust threat to the livelihoods of those who create such works and must not be condoned.”

In February, he organized a protest album in the U.K., featuring contributions from over 1,000 British musicians. Titled Is This What We Want, the album captures recordings of empty studios and performance venues as a statement against proposed legislation that would release AI companies from paying creators for their contributions. The tracks conveyed the message: “The U.K. government must not legalize music theft to benefit AI companies.”

“These creators are expressing that the exploitation of their work is not only unjust but poses an existential threat to their profession,” Newton-Rex asserted.

In closing his address in Washington, he advocated for a cooperative relationship between the creative and AI industries, suggesting that licensing—while initially a burden for AI companies—will ultimately lead them to achieve equally capable models without alienating creators and content providers.

Source
www.publishersweekly.com