Move 37. Artificial Intelligence in the publishing industry

Artificial Intelligence (AI) burst into our professional and private lives in 2023, on an unprecedented scale and at an unprecedented speed. ChatGPT, the generative AI chat created by OpenAI that kicked off the mass adoption of this technology, reached 100 million users in just two months.

Time to reach 100M users

Months to reach 100 million global monthly active users

Source: UBS/Yahoo Finance

The speed has been such that leaders of technology companies have gone from recommending that children be taught to code in schools to warning them not to bother.

Today, for example, it is already possible to access simultaneous voice translations from your mobile phone, erase a distracted passer-by from the background of a selfie, write with a grammar checker running in the background (or browser extension) and ask a virtual assistant in natural language to suggest where to go on holiday this Easter, within your budget, less than three hours away from the beach, with good weather and less likelihood of your mobile being stolen on the Ramblas.

All of these tasks involve AI, and it is not difficult to identify professions where humans are still employed and paid salaries to perform similar tasks.

On the other hand, the discourse surrounding this technology should not be interpreted as just another example of lobbying, like when the ebook was introduced; it extends far beyond the publishing industry. Its impact, at least at this early stage, would be more akin to the effect of the internet from the 1990s to the present, albeit at an even greater speed.

We will leave the discussion about P(Doom) or Extinction Factor, the probability that AI will cause the extinction of humanity, till last; Elon Musk, co-founder in 2015 of OpenAI and board member until 2018, places this between 20 and 30%.

In the publishing industry, authors have been the quickest to adopt AI in the creation of new works, to the point that last September, Amazon limited the number of books self-published per day under the suspicion of receiving a large number of artificially generated titles.

More recently, it withdrew several titles relating to King Charles III of England’s cancer diagnosis, published hours after his condition was announced, and supposedly created by AI.

Publishers, for their part, have numerous tools at their disposal to integrate AI into their day-to-day work processes, from the translation, editing and proofreading of texts to multiple marketing tasks, to creation of cover and reading reports.

AI has its roots in the past, originating from the same decade in which Elvis released his first album and Tolkien published The Lord of the Rings. However, the radical leap that put it on track to where we are now took place in the early 2010s, the decade of Fifty Shades of Grey and Rosalía’s debut. This transformation was thanks to the work of three researchers: Geoffrey Hinton, Yoshua Bengio and Yann LeCun, winners of the Turing Award, often regarded as the Nobel Prize of computer science.

They laid the foundation for the AI we use today.

Specifically, there are two main types. The first are Large Language Models (LLMs), capable of modelling high-level abstractions that are then applied in understanding, summarising, generating and predicting text, among other tasks.

And the second are Generative Pre-Trained Transformers (GPTs), a class of LLMs specialising in the generation of text or, in the most advanced models, multimodal (text and images) output, drawing from the vast corpus of content used to train them.

And it is in this corpus that all hell broke loose: tech companies, in the best Silicon Valley tradition of "move fast and break things" once again acted first and asked permission later. The systems were trained with pirated books, using all the content the companies could get their hands on, both in the public and private domain, without paying for the right to use it.

This content is known, in cases such as the gigantic catalogues so unoriginally called Books1, Books2 (usded to train ChatGPT) and Books3 (used by Meta), the latter containing more than 190,000 digital books published in the last two decades including titles by Nobel laureates such as Alice Munro, Vargas Llosa or Doris Lessing.

Increasingly (for reasons related to the next paragraph) the corpora used to train these voracious systems are kept secret.

As a results, publishers, authors and guilds as well as newspappers and digital media groups have begun to sue OpenAI along with Microsoft (its biggest investor) and Meta for illegitimate use of its intellectual property. I say started, because experience has repeatly shown that with big tech companies, if you don't make a fuss, you won't get paid, so the way forward is clear.

It would be irresponsibly naïve to expect any of these companies to spontaneously come forward and offer to pay for all the pirated content, present and future, used to train their powerful AI systems.

On the other hand, many publishers are already adjusting their contracts with authors and sales platforms to prepare for future developments, requesting assurances from authors regarding the authenticity of their work and whether it involves AI intervention, which must be declared. Additionally, they are requiring sales platforms to contractually commit to not using their catalogue for AI training without explicit permission (and payment).

For Bookwire, as a leading publishing distribution company, defending the integrity and security of the intellectual property (IP) of our publishing partners is a vital mission.

It forms the basis of a thriving creative ecosystem, where authors, publishers and sales platforms can safely share their ideas, knowing that their work is protected.

We are fully committed to protecting what is most important to the publishing company: the creative works themselves.

And after the ad break, on a final positive note: the extinction of humanity caused by AI.

It’s a positive, obviously, because if the time comes, concerns about royalties will take a back seat for the vast majority of publishers and authors...

I'm not saying that we have to work on this hypothesis, but the speed at which AI has progressed in just a year and a half (see, for example, the generation of video from text in March 2023 and in February 2024) makes it difficult to predict where we will be in a decade, but here are some hints.

Jeffrey Katzenberg, founder of DreamWorks Animation, stated months ago that the animation industry will need less than 10% of its staff in less than three years.

Elon Musk, in his recent lawsuit against OpenAI claims that ChatGPT 4 has already reached AGI (Artificial General Intelligence) level, which is considered to be superior to humans in all aspects.

Although he provides no evidence to support his claim, we witnessed that superiority with another AI eight years ago.

I'm talking about the famous move 37 in the March 2016 best-of-five-games Go tournament showdown between AlphaGo, the AI created by Google's DeepMind, and world champion Lee Sedol.

Brilliantly described by Benjamin Labatut at the end of his latest fascinating novel, MANIAC, move 37 is a glimpse of what is to come - an act carried out by a human creation, an artefact, which will first surpass us and then leave us far behind, just like a calculator surpasses us in calculating capacity, or a car in speed. This time, however, it surpasses us in something inherently human: intelligence.