It’s a momentous and consequential time in the publishing industry. The promise of generative AI is being realized in new and remarkable ways, offering the potential to significantly transform our workflows, business models, and the products we offer to readers. Yet the technology is still nascent, and the path forward is still being paved. As someone on the front lines of the technology revolution in publishing, I’ve become a firm believer in the promise of generative AI. Its applications are vast and diverse, while its potential to disrupt the industry is simultaneously immense.
In my work, I have seen firsthand how generative AI can be harnessed for many uses. It is not a theoretical tool confined to academic research; it is an operational tool and is here now. By illuminating the promises and pitfalls of generative AI from my perspective, I hope to foster a deeper understanding of this powerful technology and its potential to revolutionize our industry.
For me, generative AI has been instrumental in tasks that previously demanded hours of effort. It’s helped me create engaging and targeted marketing copy in a fraction of the time than before, allowing me to customize and iterate on messaging manually in ways that would be nearly impossible without AI assistance. Further, AI has proven to be remarkably adept at generating book metadata, streamlining a process that can be tedious yet crucial to book discovery and sales.
Other applications of AI I’ve tried have included deciphering long strings of customer service emails, first-cut analysis of content and supply chain vendor contracts, extraction of rights grants and royalty terms in contracts, cleaning up extracted text for creating e-books, identifying competitive titles, and identifying potential DEI issues in manuscripts. For most of these applications, what was once an hours-long task prone to human error can now be completed much more quickly and accurately.
While these applications of AI have been invaluable, they are not without challenges. Harnessing the power of AI in publishing, it turns out, isn’t as simple as plug and play. It requires thought, effort, and an understanding of the technology, its application, and the industry. The prompts to accomplish tasks often require significant iteration, and the results need careful review and editing by humans.
For me, one of the most exciting applications of AI has been extracting contract terms. Generative AI, equipped with a knack for pattern recognition, can sift through dense legalese, identifying and extracting key terms with impressive accuracy. When examining royalty agreements, term duration, and types of rights granted, each element is often buried within a thicket of legal jargon that can be time-consuming to decipher. Generative AI can be trained to identify these specific terms, significantly reducing time spent on contract review to populate royalty or title management systems.
A typical challenge faced by production editors everywhere is extracting text from documents in formats such as PDF or, even worse, from scans of printed pages. The extraction process often results in dirty copy with incorrect character encoding, misplaced line breaks, or missing sections. The standard process often uses third-party vendors to take additional steps to clean up the text and render it suitable for further use. I’ve employed generative AI to replace this entire process. The application can even highlight the corrected elements for a quick review.
Incorporating AI isn’t just about enhancing the operational elements of publishing. It’s equally valuable for data analysis. Using OpenAI’s Code Inspector, I’ve delved deeply into the wealth of market and logistics data publishing operations generate daily. One critical aspect of education publishing, particularly during peak seasons, is the analysis of delivery times. By feeding logistics data into the AI model, I uncovered trends and identified bottlenecks affecting delivery times. The AI model deftly handled large datasets, offering insights that could have taken people days or weeks to arrive at. It was still important to know what to look for and to create the right visualizations to demonstrate the issues, but the basic number crunching took only a few minutes. Watching the tool try various approaches, reach dead ends, and try something else until a suitable result was produced was breathtaking.
Powerful but not infallible
These examples underscore an essential truth about generative AI’s role in publishing: its power is immense, but it is not infallible. AI tools are capable of remarkable feats, but their output needs to be treated with discernment and care.
Take the example of finding competitive titles. This seems like a straightforward way to use generative AI, but it still requires a sound understanding of the industry and its data. In an email exchange with Thad McIlroy, a frequent contributor to Publishers Weekly and a longtime colleague, he noted, “I think we state that AI will be good at finding comps without understanding what that means. The traditional method of finding comps is superficial, almost to the extent of being worthless. What do we want from a comp? It intersects with recommendation engines. We want to identify the top book(s) matching the stylistic/content profile of the manuscript we plan to publish. That’s a tall task... and sidesteps the near-insurmountable challenge of ingesting in-copyright titles into a comp database.”
Thad is absolutely correct. By processing vast data, AI can generate lists of potential comp titles given only a phrase or two as input. In my case, it generated a list of reasonable-sounding comps... that didn’t actually exist! To be fair, the developers behind AI systems, such as OpenAI, the company behind ChatGPT, acknowledge this caveat. They’ve added warnings to AI outputs, noting that generated titles are illustrative of what to look for rather than a definitive list of existing books.
Even with AI’s capability to analyze data and generate insights, the onus is on the user to ask the right questions and to know what to look for in the answers, which underlines the ongoing importance of human involvement in the application of AI. While AI provided the tools, I had to direct its focus and interpret the results.
While this might initially seem like a limitation, it can also be a strength. It reinforces the role of AI as an enabler rather than a replacer of human activity. It helps us become more efficient and informed, enabling us to handle tasks at a scale and speed that wouldn’t be possible otherwise. Yet it doesn’t diminish the value of industry knowledge and human judgment; it highlights the importance of these elements in harnessing AI’s full potential.
Truly scalable enterprise applications strive for predictability, consistency, and accuracy—you don’t want your financial systems to be inventing the data on which your company operates. While generative AI hasn’t yet achieved this level of accuracy, developers continue to work on eliminating the uncertainty associated with the factual and formatting accuracy of the answers returned by the AI. Their goal is to remove much of the routine busy work, thus enabling human creativity and judgment to shine through.
OpenAI continues to release features to assist with this. For example, its developers recently introduced a feature to make the data returned from API calls more systematic and predictable. But there’s still a long way to go.
There are many promising applications of generative AI underway in publishing. For example, PanOpen Education has incorporated AI into its courseware platform. The AI acts as a tutor, assisting students, helping them with misunderstandings, and allowing class time to be used for deeper discussions. As the president of PanOpen, Brian Jacobs, aptly puts it, “Generative AI is helping to realize the long-held dream of person-centric learning, of breaking finally with a factory model of education. In this sense, we see such tools as empowering educators and learners in ways that would be unimaginable without them. And far from supplanting the educator’s creativity, AI can be an extraordinary enabler of it in new forms.”
Similarly, Gutenberg Technology is using AI to enhance the accessibility of content created with its authoring tools. Gutenberg uses AI for accessibility remediation (an issue for all publishers), standards alignment, and test item generation (educational publishers). The president of Gutenberg Technology, Gjergj Demiraj, says, “Our incorporation of AI is about precision and consistency, providing significant benefits to authors and publishers. It helps us ensure that publishers’ content aligns with standards and is accessible to all, without curtailing the creative vision of their authors.”
These examples underline how companies are making headway in marrying AI with human creativity and judgment to provide a more efficient, accurate, and innovative platform. There are many other possible applications of AI in publishing, including title development, sales, marketing, and, of course, operational and financial functions.
As we stand on the cusp of this transformative journey, staying informed and engaged is crucial. Let’s not shy away from the opportunities generative AI offers, but instead lean into the learning curve. Experiment with AI tools, involve them in your projects, and explore their potential. Participate in discussions about the ethical use of AI, its limitations and its promises. Most importantly, consider how we can shape this technology to serve our industry, our readers, and our shared future. The role of AI in publishing is not a question of if but of when and how. It’s up to us to ensure that “how” aligns with our highest aspirations and ideals.
Ken Brooks is the founder of the consulting firm Treadwell Media Group and is a founding partner of Publishing Technology Partners. He has served as chief content officer at Wiley and COO at Macmillan Learning.