By Porter Anderson, Editor-in-Chief | @Porter_Anderson
Chatfield: ‘As Photography Challenged Art’Twice during today’s (September 5) Frankfurt Rights Meeting discussion on artificial intelligence the word mediocrity came up.
In introducing the topic for the seminar The lmpact of AI on the Rights Business: An Outlook, moderator Tom Chatfield said that on a positive note, “I think that AI challenges us to think in new ways about creativity and written work and words just as photography challenged art profoundly.”
Moments later, he captured the galloping ambiguity that surrounds the topic for so many, adding, “At their worst, [AI systems] can be forces of mediocrity or deception or simulation that push us aside or cannibalize our works and ideas while impoverishing our own culture and mental processes.”
That approach-avoidance conundrum, which emblemizes AI for the international book publishing industry, wasn’t dispelled today—nor could it have been—in the first of Frankfurt Rights Meeting’s four online programs. Instead, registrants came away with a thoughtful overview of a debate in which what’s not known is often more important than what is.
This is the reformatting of the venerable Rights Meeting at Frankfurt, this year opening the programming to more attendees—about 120 today, organizers tell us—than might be able to come together for it physically in Frankfurt during one of the most intensely over-booked weeks of any literary-rights professional’s year.
More information on the Frankfurt Rights Meeting series is in our preview here, and registration can be had online here. Recorded sessions are provided to registrants, so registering now will not mean that you’ve missed today’s opening event.
McIntosh: ‘To Make the Discovery Path Easier’
In today’s program, mediocrity was mentioned for its second time by Madeline McIntosh, the former CEO of Penguin Random House USA.
She spoke of a prompt that the author Gabrielle Zevin had used in her work on Tomorrow, and Tomorrow, and Tomorrow (Penguin Random House/Knopf, 2022). When McIntosh dropped the prompt into the aptly if phonetically named Sudowrite—an AI fiction-writing “tool”—instead of something like Zevin’s remarkable opening sentence, she got text “which I couldn’t even be bothered to transcribe … because it was boring me just to try to read past the first sentence. Basically, it would completely put you to sleep.
“And so, to use Tom’s [Chatfield] earlier point,” McIntosh said, “it really does illustrate AI as a kind of force of mediocrity,” at least in terms of that bogeyman of the “book-writing machine” that seems to spook so many people.
“Instead of having publishers in secretive camps and whispering to each other about what’s going on, I really hope that this community will start to come together in sharing ideas, concerns, and possible areas of experimentation with each other.”Madeline McIntosh
McIntosh, however, had come with an issue that lies not in the software but in the people of publishing: AI and its implications for the book business are being discussed in silos, she pointed out, with authors unsure about publishers’ positions; publishers wary of authors’ stances; and many industry players whistling past Silicon Valley as blithely as possible without engaging each other in the debate.
“Instead of having publishers … in secretive camps and whispering to each other about what’s going on,” she said, “I really hope that this community will start to come together in sharing ideas, concerns, and possible areas of experimentation with each other.”
And when it comes to questions of AI and its appropriation of content, McIntosh said, “Part of our interest is thinking about growth. Because if some of the risk associated AI is going to be a kind of nibbling away at some areas of our business, then the best strategy for combating that is to make sure that we are using the technology to try to grow our business, as well. Instead of thinking that AI is going to replace the novelist, let’s think about AI simply as a copilot for human authors.”
The real gift of AI, she noted, could lie in discoverability, which may remind some Publishing Perspectives readers of the intent behind the Shimmr AI start-up being led by London’s Nadim Sadek.
“Anyone who is in trade publishing,” McIntosh said, “knows that the main challenge we have is not in trying to generate more content. It’s been a very long time since that was the first challenge that a publisher would list. Instead our challenge is to try to get our individual books or lists of books discovered in a sea of millions of other pieces of content, whether those are books or streaming movies or what have you.
“So I don’t know exactly who’ll figure this out, but I think we can certainly hope that that there will be entrepreneurs out there who will figure out ways that we can apply AI to speed up our process and make it more effective in terms of finding those consumers who are most likely to respond well to our latest hot novel. And to make the discovery path easier for them.”
Cox: ‘That Human Touch’
Thomas Cox, managing director of Arq Works near Oxford, is familiar to Publishing Perspectives readers for our interview earlier this year, as AI fever was spreading along with consumer interest in ChatGPT.
In today’s session, Cox went over the development of both the major projects in the industry and a second tier–the open-access programs being developed outside of ChatGPT-4, which is backed by Microsoft, and Google’s various products. He talked through the development of Llama 2 (introduced in mid-July by Meta), “which is open-source, so people can take it, they can run it themselves, they can adjust it … a model being developed very quickly by the open-source community.”
Cox also talked about Anthropic, “set up by ex-OpenAI employees.” In it, he said, “you can have a whole book as an input and ask questions about that book, such as what’s the story line, who are the characters, and it has all the information right there to analyze.” At his own company in England, Cox said, “We’re starting to see a huge expansion in the number of AI-based applications. Many of these will claim to be their own AI, but in the background they’re actually making requests to ChatGPT or to Google or one of the other primary providers.”
As regards rights trading, Cox—who for years has worked with the online-trading Frankfurt Rights program—said, “The rights business is a very personal one: that human touch and networking and negotiation are still at the core of the business. Book fairs are key dates on the calendar and that will remain the case. So I think the core of that process will stay the same.
“You have AI-assisted authoring already—whether people are telling you that they’re using AI to help with their writing or not.”Thomas Cox
“However, that doesn’t mean that it’s not going to change. You have AI-assisted authoring already—whether people are telling you that they’re using AI to help with their writing or not.
“We have AI-narrated audiobooks, which is becoming more common. Google and Apple have launched services earlier in the year and there’s plenty of other places that will generate audiobooks as well. AI-assisted translation is already a thing, and they’re only getting better. And there are copyright and contractual concerns around all of that–which will affect the rights of the stream. Because just on a day-to-day basis, the AI tools that we will use are going to be changing, augmented with AI, and that’s happening quite quickly.”
While the concept of AI-generated literary content seems out of reach, Cox gently put onto the table the notion that, “That doesn’t mean that in a few generations’ time, there won’t be new models which are specifically focusing on a particular era or maybe even on a particular genre and have their own persona. Almost like authors.”
And that, he said, could lead to people “becoming more accepting of AI-generated longer-form content.”
Calow: ‘Relevant Rules and Precedents’
One of Duncan Calow’s comments most quickly resonant with publishing professionals watching the program today had to do with the time it takes for regulatory and legal progress. Calow is a partner in the intellectual property and technology department of London’s DLA Piper. He talked of how, “There’s an unavoidable and frustrating time-lag before relevant rules and precedents emerge. And while that’s still true, to some extent with AI, it’s also important to understand the extensive amount of work currently being undertaken around the world to map the relevant regulatory landscape.”
In fact, Calow sees what he calls a “renewed zeal to address the challenges that AI presents” because of “a marked uptick in activity in the wake of recent developments.” He pointed out that regulatory reactions to that uptick swing in various markets and scenarios, from “soft-law, self-regulatory” approaches to “the other extreme personified by Elon Musk and his scary open letter calling for an AI moratorium” and, in policy realms, the “very robust actions of the Italian data protection authorities in implementing [their] blocks on ChatGPT.”
He looks, he says, at the key development lying in the “hardening of regulation in the middle ground” perhaps best represented by the “horizontal regulation” of the European Parliament’s AI Act, which our readers will recall won applause from the Federation of European Publishers’ applause in mid-June. There are risks, of course, Calow pointed out, that premature or over-regulation could become a factor here in retrospect, “very reminiscent,” as he said, of what occurred in regulatory frameworks “of e-commerce 20 years ago.”
“They ‘marry in haste and repent at leisure’: law and technology are always something of a shotgun marriage.”Duncan Calow
As he put it, “They ‘marry in haste and repent at leisure’ and law and technology are always something of a shotgun marriage.”
Relative to rights, he said, “the legal devil is in the local details,” and while copyright is a major focus, of course, other IP interests are in play, as well, including “database rights, trademarks, and personality rights, as well as privacy- and data-protection.”
Comparing inputs and outputs, Calow said, “On the input side, the key question is whether the process of building and training an AI system with content probably scraped off the Web infringes [protected] content,” or whether copyright exceptions “such as transient and temporary copying, the text data-mining provisions in the European Union, or fair use in the United States will apply.” In the interim, there’s an emphasis on licensing, prompting AI operators to look for deals with rights holders.
“On the output side,” Calow said, “there’s a more fundamental question being asked,” namely whether “an AI-generated work is protected at all.” While many territories and markets at this point “seem skeptical” on this point, Calow said, “to the extent the rights do exist, ownership might in practice be varied by contract, anyway,” and not all AI systems will necessarily take the same approach.
Much remains to be seen on where rights outlines will be drawn relative to AI-generated content.
Schoppert: ‘A Little Bit of an Alarm Bell Ringer’
The director of the National University of Singapore Press, Peter Schoppert, provided something of a departure in the Rights Meeting–some commentary from the scholarly and academic side.
His work with scholarly writings, Schoppert said, has given him a clear view into the idea that “what text- and data-mining is about is pulling facts out of expression, and it’s not just science that this has been used for.” For example, he pointed out, social media has been mined for sentiment analysis.
If anything, Schoppert said, he sees a through-line in the digitization of STM publishing and the approach now of AI. Digitization in scholarly circles, of course, “went from creating reference books to creating databases, and some companies realized that [this was] content they could deliver it to professionals along with services and software and processes. So content was unbundled from books and delivered together with workflow processes.”
In this sense, publishers themselves, in the scholarly fields in particular, have been operating in patterns of information extraction and recombined formats and delivery.
“What the model is doing is building a statistical representation of the text. What is ‘style’ but a kind of statistics of how you use language?”Peter Schoppert
“But now,” he said, “generative models are different. They’re doing something different, generating text. It’s a prediction, as Thomas [Cox] showed, but the output is actually text, large amounts of text, images, videos, 3-D models, music, voice. It’s bigger data.”
Schoppert pointed out that Calow had referenced questions about “whether text- and data-mining exceptions actually apply to training a large language model. We’re pulling facts out of text, we’re mining text for nuggets of fact–the large language model wants the whole text. It wants long texts, high-quality texts, as coherent as possible.”
But there’s a key distinction here, Schoppert said. The large language model wants more than the whole text. It also wants the style.
“What the model is doing is building a statistical representation of the text,” he said. “What is ‘style’ but a kind of statistics of how you use language? How long your sentences are, which adjectives you favor, certain patterns and how you use them. So what the models are learning is style, they are learning expression, and expression is what copyright protects.”
Schoppert said he thinks “both on the technology side and on the publishers’ response that a licensing regime can work. And yet, he classified himself as “a little bit of an alarm bell ringer” because that deeper, broader, framework in which AI is created to gather and generate means there are rights beyond copyright on the table, including “privacy, moral rights, and the bias question.”
Nadim Sadek, the founder of Shimmr, mentioned in today’s article, is confirmed as a speaker. In addition, we’ll hear from Anna Soler-Pont of the Pontas Literary and Film Agency and Christoph Bläsi of the Johannes Gutenberg University Mainz, with Arq Works‘ Thomas Cox moderating. That session is scheduled for 12 p.m. on Thursday, October 19, in Room Spektrum of the Messe Frankfurt Congress Center, second level. More on the Publishing Perspectives Forum is here.
More from Publishing Perspectives on artificial intelligence and its debate relative to publishing is here, more on digital publishing is here, more on Frankfurter Buchmesse is here, more on international translation and publishing rights is here, and more on international book fairs is here.