11 April 2025
Sara Crowley Vigneau, Partnerships Lead, Zendy
As someone who recently transitioned from publishing across to the AI sector, I find that recent conversations about AI’s impact on publishing and the influence of ‘Big Tech’ feel somewhat unbalanced and that the existence and role of smaller, mission-driven companies often goes overlooked.
Most small AI-powered start-ups, like the one I work for, have a small team with a modest budget; the billion-dollar deals being discussed at large AI firms do not apply to us. I work with colleagues who are passionately engaged with sustainability, diversity and equity, and committed to finding a fair and equitable approach to AI among the rapid innovations in our industry.
In this respect, attending publishing events such as SSP, AUPresses2024 and Charleston last year offered some key insights into the main challenges and concerns of publishers facing the rise of AI over the last few years, and enabled us to identify and shape a sustainable and ethical way forward for our own AI developments which I’d like to share in this post, in the hope of inspiring others in a similar situation.
Consent and Ethical Use of Content
As in any partnership, it’s key to ensure that contracts are clear, transparent, and respectful of content ownership. This is as much about ethics and transparency as it is about legal obligations. In my organisation, our approach to AI development has been guided by a clear ethical and legal framework (which we developed as guidelines, or ‘imperatives’), ensuring that we maintain a sustainable and responsible model to guide our development of AI.
A key concern that arose in our discussions with publishers was the use of content to train large language models (LLMs). Whilst many larger well-known AI do this, we decided to go another way, instead leveraging Retrieval-Augmented Generation (RAG) techniques to fetch and process relevant data in response to user queries, a solution that minimizes ethical and legal concerns. Unlike models trained on vast datasets of copyrighted material, RAG fetches and processes only what’s necessary, reducing copyright risks and ensuring transparency. We also updated our contracts to allow publishers more flexibility in how their content is used, offering options for content choice and contract duration, all with a focus on mutual benefit and respect for the intellectual property involved.
Credit and Transparency
Full and accurate referencing has always been a cornerstone of sustainable AI, as it not only upholds academic integrity but also fosters trust in AI-generated content. Throughout the beta testing phases of our LLM, we refined our referencing processes to ensure that all credit is properly attributed. The iterative process of refining our referencing system helped us build greater transparency, ensuring that every piece of content generated could be traced back to its original sources. This enhanced reliability has been crucial for maintaining trust with both publishers and end-users.
Fair Compensation and Revenue Sharing
According to a report by Straits Research, the global market for AI datasets and licensing in academic research and publishing is projected to reach 462 million USD by 2025, with tremendous growth expected over the next decade. Roger C. Schonfeld revealed he and his colleagues are tracking AI deals, some of which report initial payments of 10 million USD and more. However, with no requirement to make these deals public, there is limited information on how much publishers are being paid, what they are giving in exchange for payment and most importantly whether and how authors are being compensated. This raises a broader conversation about how revenue from AI models can ultimately reach authors, who are at the heart of the research ecosystem.
An alternative to this is a revenue-sharing model based on content referencing, which ensures that publishers receive an equal half of the revenue generated. Providing title level reporting, also enables publishers to calculate revenue due on an individual title basis.
Developing a sustainable, equitable model is the responsibility of all stakeholders in publishing, content providers and AI companies alike. AI companies must be transparent about how they use content, while in turn publishers must be transparent about their transactions and how revenue is shared with authors. AI, like any emerging technology, must operate in a way that respects and compensates all stakeholders.
Addressing the AI Access Gap
AI in publishing and research isn’t just about creating smarter tools and licensing content —it’s also about tackling the AI access gap. Working with end-users in low- and middle-income countries (LMICs), we know first-hand that the potential of AI to change lives isn’t available to everyone and threatens to increase the significant disparities which already exist. This access gap only serves to reinforce existing inequalities and threatens to widen the divide between the global North and South.
In addition to these workforce disparities, there is a significant gap in access to quality training in STEM and AI, particularly in underprivileged communities. Furthermore, AI systems often reflect the biases of historical data, which can perpetuate and amplify existing inequalities.
What can publishers, libraries and tech providers do to address this? Many of the end users lack the resources, infrastructure, and expertise to fully benefit from AI advancements. This could be changed by allowing the following:
- Free or Reduced Access: Consider providing free or heavily discounted access to AI tools for users in developing countries, enabling institutions and researchers with limited budgets to start utilizing AI technologies.
- AI Literacy and Training: offer free AI literacy courses as part of a broader initiative to make AI education more accessible. This training contributes to overcoming the technical expertise gap, helping individuals from all backgrounds engage with AI technologies.
- Partnerships for Advocacy and Support: actively seek to partner with organizations that can help source sponsorships to advocate for and support access to AI tools in LMICs. Building infrastructure and developing solutions tailored to these regions can help ensure access to AI is equitable.
Academic libraries notably can play a key role in advocating for and supporting fair access to ethical AI technologies for students and faculty. They can choose their partners with sustainability in mind, exploring subscriptions which promote fair and equitable access to AI tools and training, as well as promoting a collaborative approach to supporting digital literacy and access for all audiences.
There is scope for so much more. Building ethical AI partnerships isn’t just about technology—it’s about reshaping knowledge accessibility for the future. As stakeholders, we must work together to provide the building blocks of a sustainable and equitable AI ecosystem which works to the benefit of all.