Artificial Wave of the Future? Publishing and Artificial Intelligence
By Craig Gipson
The Oxford English Dictionary named “rizz” as its word of the year (meaning “charm” or “charisma”) but 2023 belonged to an acronym: AI. Artificial intelligence, long the stuff of sci fi plots and dystopian novels, entered the mainstream with technology firm OpenAI’s public launch of ChatGPT in late 2022. Publishing, like most industries, will not be immune from AI’s impact. Make no mistake: as technological revolutions go, AI is closer to Gutenberg’s printing press than Bezos’ Kindle. And while AI has existed for many years in less noticeable forms, generative AI’s entry into everyday life and everyday industry requires consideration by publishers large and small.
What is Generative AI?
At a basic level, generative AI is a system of using computers trained on robust datasets to problem-solve by predicting the language, image, or other response appropriate to reply to a human-created prompt. The scope of the technology’s use has few limits: this year witnessed examples of systems able to more accurately predict breast cancer risk than human radiologists [1], adequately translate a 700 year-old work by the poet Rumi from Persian to Urdu to English [2], and read a bedtime story to you in the voice of late actor Jimmy Stewart [3]. Need a travel guide for your free time at the next ECPA event? Ask ChatGPT or Google’s Bard and have a customized itinerary in seconds. But powerful tools like these are not without problems and risks. Their very creation is under examination in a host of lawsuits challenging the use of copyright-protected works to train large language models. The authenticity of their outputs has come under fire after AI-produced book summaries began to proliferate on Amazon and celebrity AI-voice-cloned songs became internet sensations [4]. Their responses sometimes include made-up facts—so-called “hallucinations”—or exhibit insensitivity or bias. Generative AI is here to stay but how and for what purposes it will be used remains the subject of much speculation and prognostication.
How will Generative AI affect publishing?
Generative AI possesses the potential to penetrate virtually every area of publishing. As a tool, it may generate significant efficiencies within publishing organizations. Marketing is often considered the “low hanging fruit” of AI but everything from the author generating ideas for a manuscript to the machine pressing ink on a page could be AI-influenced.
If generative AI brings a proliferation of new content to the market, it may at least also aid discoverability. On the marketing front, organizations like Open Road are already utilizing AI tools to crawl the web for mentions of backlist titles and develop targeted marketing plans [5]. Marketing copy and metadata can be created easily while under oversight of human employees or contractors. Text-based language models are capable of generating or evaluating ideas for authors facing writers’ block. Editors may discover plot holes and inconsistencies in mere minutes. The proofreading of manuscripts may be completed in a fraction of the time previously required.
And beyond traditional editorial, production, and marketing functions, generative AI may make possible entirely new types of products and more efficient means of creating traditional ones. Titles without sales to justify the costs of an audio version may now be read by an AI voice-cloned narrator. Languages without the market size to command a translation may now only require a translator to review an AI-generated version. An author’s entire library of works fed into an AI could create an interactive tool, allowing readers to “discuss” the author’s work with a virtual version of the author.
Despite the myriad possibilities generative AI offers publishers, there are an almost equal number of risks and difficulties. Chief among the publishing industry’s legal issues with generative AI is the application of U.S. copyright law. Currently, any AI-generated text or image is not subject to copyright protection. So while generative AI may be an extremely useful tool, over-reliance on machines in creating content forfeits the ability to control exclusive rights in that content. To be clear, the human-originated portions of a work may still be protectable, but the Copyright Office requires that all AI-originated content be disclaimed.
There is also a downstream copyright concern: can generative AI create a work that may substitute for a human author’s work? If a publisher ingests an author’s work into a publicly available AI tool like ChatGPT, can anyone prompt the AI to create a work in the style of that author on any subject of their choosing? And if they do so, would the resulting work destroy the market for the author’s work or infringe the author’s original copyright? We do not know the answers to these questions yet. Weighing the risks and benefits of the technology will be a balancing act each publisher will have to perform.
What AI-related issues should publishers consider?
Generative AI’s rapid evolution likely means that publishers will be forced to consider its implications or fall behind technologically. From internal policies to contractual issues, there are a range of impacted matters. One pressing question for publishers will be the types of AI tools they deem acceptable for use. Generative AI itself is a different category from more familiar assistive AI technologies (spell check, closed captioning, facial recognition, etc.). But within the umbrella of Generative AI exists an assortment of types.
“Public AI” tools contain models trained on massive amounts of data from the web and other sources that may further train their models on inputs users ingest into the system. These public AI tools can produce unreliable outputs, and concerningly for publishers, may generate outputs similar in content to ingested inputs. Some of the more well-known generative AI tools fall into this category: ChatGPT-4, Google Bard, Dall-E2.
There is also a developing field of “private AI” tools which vary in their terms of use and models but may become useful to publishers. Some of these tools may offer “walled gardens” in which outputs will only derive from a set of specific inputs ingested by the user. Others may allow inputs to further train their language models generally, but will not create outputs similar to content ingested as inputs. For example, the International Society for Technology in Education worked with Google to create a “walled garden,” specialized chatbot for K-12 educators. Two professional development organizations curate data fed into the model to provide assurance to educators that the outputs are trustworthy. Certain custom GPTs are already available but this developing field of specialized AI may have exciting applications in individual industries.
And while there are numerous AI-related issues to address, every publisher can start by evaluating answers to some high-level questions:
- How should our organizations’ authors or independent contractors be able to use generative AI in creating content our organization will publish?
- How should our editorial, marketing, or other staff members be able to use generative AI in performing their tasks and creating products?
- What contracts need to include generative AI provisions to reflect our organization’s AI preferences?
- What types of AI tools are acceptable for our authors or staff to use?
- What policies should be in place for staff members to know how and for what purposes they may use generative AI?
The answers to these questions and their related details will dictate what further action should be taken. The Authors Guild issued several sample AI publishing agreement clauses in 2023 that have impacted discussion, but the technology remains too young for an established industry standard, and agreeing to broad restrictions now could limit innovation later. For example, some publishers may be willing to commit not to license manuscripts for use by public AI engines, but to retain the freedom to ingest manuscripts into private AI platforms to generate cover design comps, concept ancillary products, or assist with the editorial process. Discussing generative AI internally and implementing decisions in contracts and policies at this early stage can help prepare your organization for the future.
Flagler Law Group is working with several ECPA members to guide internal discussions about generative AI use, revise applicable contracts, and draft relevant AI policies. We encourage all ECPA members to work with your copyright counsel to discuss these issues. ECPA will continue to keep membership informed of major AI developments in this quickly evolving technology.
Endnotes:
[1] AI Outperformed Standard Risk Model for Predicting Breast Cancer. Radiological Society of North America. https://www.rsna.org/news/2023/june/ai-for-predicting-breast-cancer. June 6, 2023.
[2] Karen Weise, Cade Metz, Nico Grant and Mike Isaac. Inside the A.I. Arms Race That Changed Silicon Valley Forever. https://www.nytimes.com/2023/12/05/technology/ai-chatgpt-google-meta.html?searchResultPosition=1. December 5, 2023.
[3] Isabella Kwai. Can’t Sleep? Listen to an A.I.-Generated Bedtime Story From Jimmy Stewart. https://www.nytimes.com/2023/12/05/technology/calm-jimmy-stewart-ai.html?searchResultPosition=1. December 5, 2023.
[4] Joe Coscarelli. An A.I. Hit of Fake ‘Drake’ and ‘The Weeknd’ Rattles the Music World. https://www.nytimes.com/2023/04/19/arts/music/ai-drake-the-weeknd-fake.html. April 19, 2023.
[5] Elizabeth A. Harris. Decades Old? No Problem: Publisher Makes a Bet on Aging Books. https://www.nytimes.com/2023/05/24/books/old-books-out-of-print-open-road.html. May 24, 2023.
