Artificial intelligence as reviewer, text to speech reader – and author?

28 May 2019

At the beginning of the year, Professor Christoph Bläsi hosted the 14th Mainz Colloquium on te topic of Artificial Intelligence in the Book World – Machines as editors, Machines as Readers? at the Gutenberg Institute for World Literature and Written Media of Johannes Gutenberg University Mainz (JGU). We talked with Professor Bläsi and learned what is currently feasible in the field of Book Studies and where it might be heading in the future.

Will computers one day decide what is worth reading and what is not? "Publishers get sent far more unsolicited manuscripts than they can reasonably handle," says Prof. Bläsi. "That's why they often employ students to undertake a preliminary review of what they have been sent. It is not beyond the bounds of reason that in the future artificial intelligence could actually be used for this purpose instead. It could help with separating the wheat from the chaff."

The start-up company QualiFiction is already offering software that promises even more: "They claim to have developed a system that can predict the percentage chance of a book becoming a bestseller. It analyzes factors like sentence length, subject, or the way the plot ratchets up the suspense." The system was 'trained', as it were, by having aspects like this fed into it that were extracted from a huge library of previously evaluated books, each of which had been assessed with regard to its performance on bestseller lists. The technique known as machine learning was used for training. For example, Frank Schätzing’s successful novel The Swarm was evaluated in a trial run and was rated by the system as almost 90 percent likely to become a bestseller.

Artificial intelligence likes mainstream tastes

At the beginning of 2019, the Department of Book Studies at the Gutenberg Institute for World Literature and Written Media of JGU hosted its 14th Mainz Colloquium. Bläsi organized the conference on the question of Artificial Intelligence in the Book World – Machines as Editors, Machines as Readers?, inviting experts from various disciplines. In addition to representatives of the publishing industry and associated service providers, Professor Stefan Kramer of the JGU Institute of Computer Science was also invited to speak. He gave a brief overview of key aspects such as artificial intelligence (AI) and also deep learning – two topics that have recently been making headlines. Holger Volland, Vice President of the Frankfurt Book Fair, then presented a wide-ranging overview of what artificial intelligence may mean for the publishing industry, from its potential automation – the principal subject of the colloquium – to its ability to create content.

A trained programmer and systems analyst before he studied Mathematics, German, and Linguistics and became a researcher in the field of computational linguistics at the Universities of Heidelberg and Bielefeld, Bläsi himself has been working intensively in the interconnected field of AI and books for some time now. He held managerial positions in which he was responsible for digital products at major publishing houses such as C.H. Beck and F.A. Brockhaus. In 2004, he was appointed Professor of Book Studies at Friedrich-Alexander-Universität Erlangen-Nürnberg, and in 2009 he joined the, then, Institute of Book Studies in Mainz. "When it came to my academic work in the field of Book Studies, my expertise in computer science and computational linguistics – a key area of artificial intelligence nowadays – largely went to waste," he remembers. "I'm glad I can now combine both aspects of my professional life."

The two fundamental issues raised at the colloquium were how artificial intelligence could help make day-to-day work in publishing houses more efficient, and whether it can be deployed to give employees more freedom to carry out those tasks that can’t be delegated to machines. At first glance, an AI-supported initial review phase, as already described, would seem like a good candidate. But this is precisely where the flaws with the approach become evident: “Artificial intelligence favors mainstream books,” notes Bläsi. It picks out what was successful in the past. “And that means innovative material, such as avant-garde literature, always ends up getting discarded.”

New products and new challenges for publishers

But that’s not all: “Artificial intelligence, and in particular the machine learning approach, has a tendency to perpetuate any existing bias that, as such, has been unwittingly incorporated in the training data.” The choices made by the system are exclusively based on the set of values it has been taught. Ideas that are more unconventional, difficult to understand or eccentric are therefore never going to be part of AI’s strong suit. If AI were ever to dominate the entire review process, only clones based on proven patterns of success would ever find their way to market.

Entrusting AI with simply correcting texts is far less problematic. Though more modest in scope, this may very well reduce the workload for publishing house employees. “And that might give those critical of change enough space and time to come to terms with it.”

Artificial intelligence has the potential to not only reduce workload and optimize processes but also to improve products and even create new publishing products and services. Bläsi gives an example: “Non-fiction and other knowledge resources provided by a publisher can be equipped with an intelligent natural-language interface.” This would allow readers to formulate their inquiries as simple questions and enter into a dialogue with the system, providing them with even quicker ways of getting answers. “It might be possible to offer them information relevant to the context of their inquiry that they would have been highly unlikely to find by searching using a keyword.

Bläsi even thinks that AI may allow publishers to open up novel business frontiers. “Perhaps publishers in the future will consider it a core task to help their customers find their way through the flood of digital data. Selecting, preparing, and presenting information as effectively as possible plays a huge role in the age of the Internet.” Bläsi imagines publishers becoming, in effect, curators with the help of large-scale AI applications – curators who are potentially infinitely resilient and can offer personalized solutions. “Unfortunately, to provide AI products of this kind, small and medium-sized companies, in particular, will end up having to rely on existing AI backend products from companies such as Microsoft or Google. This will inevitably make them dependent and leave them no way of withholding valuable data, with no transparency or control over what such companies could do with it.”

Machines compose texts – and what about works of art?

Computers may become assistants to editors in the future, but they have been reading aloud and engaging in dialogue with us (if only in a very rudimentary form) for a long time. Voice assistants are now almost the norm, especially in the mobile devices branch. “In the future, we might be able to ask a computer to tell us in one sentence what a book was about, or to do so in 15 minutes or in half an hour – depending on how much time you have available. In many cases that could lighten your workload, couldn’t it? Of course, to do that well, the AI would have to adapt itself to the questioner, to get a picture of what they already know, for example by having access to what they’ve been reading recently. The AI would then know just what the individual needs to be told. Sounds a little creepy, I’ll admit....”

Artificial intelligence can already independently compose its own texts, if somewhat limited in nature and subject to certain conditions. “It’s already being done for stock market or weather reports, for example. The AI converts data from tables into simple texts that can be read out by a synthesized voice.”

As Holger Volland revealed at the 14th Mainz Colloquium, AI researchers have larger ambitions than such, more or less, functional applications; they want AI to become creative. In one case, a computer was fed with data on hundreds of Rembrandt paintings and was then able to create a new work in the style of Rembrandt using a specialized printer. “That also might be possible when it comes to books, perhaps most readily for fiction, which follows relatively clear rules, for example in the crime and fantasy genres.” In this connection, rejecting material that is hard-going or unconventional mightn’t be such an issue. “We shouldn’t be holding our breath for a second AI-written Ulysses in the not-too-distant future. As scholars in our field of Book Studies, speculating on AI and literature brings us to the frontier of what we might reliably be able to achieve. As intellectuals and as citizens, we are called upon to critically accompany the on-going journey.”

Bläsi is convinced: “In our field of Book Studies, we need to keep our eye on the ball when it comes to artificial intelligence. As academics, we thankfully can’t allow ourselves to be ruled by uncritical, nerdy enthusiasm. Neither, however, should we reject or demonize outright any potential uses of AI in the field of book culture. Only by exploring them in detail will we be able to assess both their potential and their dangers.” As a scholar, that is precisely what Bläsi intends to do. “Right now, we are trying to bring together, at least informally, as many researchers as possible in Mainz at our university who are involved in artificial intelligence, in all its facets. We’ve arranged a date for a get-together. I’m eager to find out what we’ll achieve.”