Photo credit: www.publishersweekly.com
As the landscape of content discovery and consumption evolves through generative AI, industry leaders are advocating for licensing frameworks that protect the rights of creators while fostering innovation in technology. At a recent panel titled “Licensing Is a Win-Win: The Exciting AI Partnerships Between Creators and Tech,” which was part of the “The Story Starts With Us” forum cohosted by the Association of American Publishers and the Copyright Alliance, executives from diverse sectors including publishing, entertainment, and AI development emphasized the need for sustainable licensing models.
Geoff Campbell, Senior Vice President of Strategy and Business Development at Condé Nast, stressed the significance of adapting to shifting consumer behaviors. He remarked, “The front door of the internet is changing,” while noting the implications for publishers trying to monetize their content. Campbell referenced data indicating a notable rise of five and a half million in referrals from generative search results, contrasting this with a sharp decline of 64 million in searches for top public publishers.
Condé Nast has reportedly initiated “two public deals” with AI entities such as OpenAI and Amazon, and is actively pursuing additional licensing arrangements with both major players and smaller firms in the AI sector.
The panelists collectively dismissed the idea that current fair use provisions could sufficiently navigate the complexities surrounding AI training. Michael D. Smith, a professor at Carnegie Mellon University specializing in Information Technology and Public Policy, cautioned that lacking appropriate licensing structures could diminish incentives for market players to innovate. He invoked historic parallels to music piracy during the Napster era, asserting, “This is publishing’s Napster moment,” signaling the urgency for well-defined regulatory frameworks in the current climate.
Moderated by Catie Zaller Rowland, General Counsel at Copyright Clearance Center, the discussion revealed an accelerating market for content licensing within the AI sector. Davis, General Manager of Protege Media, acknowledged the understandable trepidation that many rights holders feel about AI’s role, but insisted that businesses like his are working towards fair and viable licensing opportunities. Davis noted that Protege Media has collaborated with over 50 significant content catalogs, underscoring a growing trend among startups receiving substantial funding to develop ethical AI models, such as Moon Valley which has secured over $100 million from leading venture capital firms.
Vered Horesh, Chief of Strategic AI Partnerships at visual generative AI firm Bria.ai, elaborated on their advancements in attribution technology, designed to analyze the impact of authentic assets used in training catalogs on generated synthetic outputs. This system is said to foster a sustainable ecosystem by ensuring all stakeholders receive fair compensation.
The panelists also underscored the benefits of active engagement between content creators and AI companies, extending beyond financial incentives. Campbell noted that such partnerships can lead to collaborative innovation in product and engineering developments. He emphasized that these licensing arrangements should be voluntary to prevent stifling innovation by forcing companies into reluctant agreements.
Horesh highlighted that many of Bria’s content partners are transforming into customers who also act as resellers, aiming to make the technology accessible to wider audiences.
Davis pointed out a considerable legal uncertainty impeding the adoption of AI across creative domains, with a recent disclosure from a senior legal executive at a major studio illustrating a reluctance to integrate AI into production processes due to fears surrounding copyright liabilities.
While the panel expressed optimism regarding voluntary licensing frameworks, Smith raised concerns about the swift and sometimes reckless pursuit of content by companies aiming to dominate the market. He urged lawmakers to establish clear guidelines for all participants in the industry.
Campbell mentioned that Condé Nast is exploring “three different kinds of commercial licenses” for varying use cases, adapting to the diverse needs of licensors. He indicated these would encompass B2B, B2C, and B2B2C frameworks tailored to conditions for accessing and utilizing their content.
In light of discussions surrounding data valuation, Campbell emphasized the importance of a sustainable ecosystem: “If AI companies devalue content significantly, it raises questions about who will create the next generation of content. Our goal as AI developers is to ensure a system where all parties can thrive.”
New firm promises to detect content’s use in AI training
One of the pivotal challenges in mounting effective licensing frameworks is discerning the unauthorized use of copyrighted materials within AI training datasets. In a presentation titled “Tech Innovation to Protect Creators Against AI Abuses,” Louis Hunt, co-founder and CEO of Valent, shared insights about his company’s cutting-edge technology aimed at addressing this issue.
Hunt, who previously held the dual roles of CFO and head of business development at the AI company Liquid AI, expressed that Valent seeks to navigate significant intellectual property challenges intrinsic to AI. He posed critical questions about how to identify which data is most impactful and how to facilitate legal access efficiently.
During his presentation, Hunt detailed two algorithms developed by his team: the first can assess with “up to 98% confidence” whether specific data samples contributed to training an AI system, while another, dubbed the “evident algorithm,” outputs evidence with an exceptionally high confidence level that a model incorporated particular content.
Valent’s technology produced noteworthy outcomes in its analyses, including the identification of 7.54 million URLs and nearly 2.5 million web pages within various AI training datasets for one client. When targeting literary works such as the Harry Potter series, the analysis revealed a striking 31-35% of the text had been “memorized verbatim” by certain AI models. Additionally, the technology flagged unauthorized usage of song lyrics, screenplays, and proprietary software code, showcasing its capability to operate externally without needing direct access to internal model code.
Beyond merely detecting misuse, Hunt explained Valent has developed algorithms capable of assessing how specific datasets could enhance AI model performance, thus providing content owners with leverage during licensing talks. This represents a substantial shift that empowers creators in an AI landscape where proving unauthorized use of their works is often a formidable challenge.
Hunt asserted, “We’ve tackled one of the significant issues troubling model developers,” asserting that Valent’s approach can deliver a superior evaluation of data quality for AI systems compared to existing internal methods used by developers.
Responding to audience inquiries, Hunt elaborated on various technical aspects, clarifying that studies show AI models tend to be cumulative rather than entirely retrained, countering assertions that content could be entirely excluded following rights holder demands. He also revealed that attempts at “unlearning,” which involves erasing previously used content from models, have frequently failed.
Source
www.publishersweekly.com