Feeding Our Writing Machines

MUCH IN THE SAME way as the use of computers permits us to answer questions in new ways, today it also allows us to create texts in new ways. Fuelled by an ever-expanding web of digital texts and driven by a proliferation of language processing techniques of all sorts, a calculating machine was turned into a reading machine. These days the reading machine is evolving into a writing machine. 

As our reliance on technology grows, so too does our capacity to use it as a tool for communication. With the increasing accessibility of language processing software and the vast amount of digital text available, computers are no longer solely used for answering questions, but have become capable of generating texts on their own. Natural Language Generation (NLG) technology has been used to create everything from news articles to advertisements, with the output often being virtually indistinguishable from something written by a human. While the use of writing machines has the potential to revolutionize the way we communicate, it also raises concerns about the future of writing as a profession and the impact it will have on the role of language in society.

Still, it’s difficult to make predictions, the aphorism goes, especially about the future. It’s much wiser to look at the past. Not to set expectations about this particular new technology, but to remind us how history is filled with inventions that people thought would change everything, but turned out to be a minor fad, to acknowledge that we usually spend way too much time worrying about the wrong things when a new invention is on the block and, most importantly, to recognize that in general it might take years to find out what we can essentially do with a new technology. Sure, computer-generated texts are here to stay, but predicting what sector, what medium or which group of people will eventually experience the biggest impact is simply not possible yet.

In addition to the uncertainty surrounding the impact of writing machines, there are also concerns about the potential for misuse. Just as the internet has been used to spread disinformation and propaganda, the ability to generate convincing texts with ease could lead to an increase in fake news and other forms of manipulative content. It could also raise questions about the authenticity of written works, as it becomes increasingly difficult to determine whether a text was written by a human or a machine. These issues will need to be addressed as the technology continues to advance and become more widespread. Despite these challenges, however, the development of writing machines represents a significant step forward in our ability to communicate and process information, and has the potential to bring about a new era of creativity and innovation in the field of writing.

And there is more good news to celebrate. ChatGPT is built on top of a Large Language Model (LLM). Trained on huge volumes of textual data, these LLM’s learn the patterns and relationships between the different elements of a language, allowing it to imitate human text by predicting the next words in a sentence. The vast majority of text processed – for the time being largely written by humans – is taken from the web, but increasingly the output of various mass digitisation efforts are also processed. It’s a previously unthought-of use of centuries of textual heritage. More than any human author, our new writing machines are standing on the shoulders of millions, present and past.

This text was written in February 2023, by both a human and a machine. The human is Steven Claeyssens, curator of digital collections at the KB, the national library of the Netherlands. The machine is ChatGPT, a chatbot developed by OpenAI based on GPT-3.5, a LLM. The human wrote the first, third and fifth paragraph. The chatbot was given prompts to write the second and fourth paragraph (in italics), based on the preceding paragraph(s) ‘as part of an essay’.

Steven Claeyssens is Curator of Digital Collections at the KB, the national library of the Netherlands. He holds an MA in Philology (Ghent), and a PhD in Book and Publishing Studies (Leiden).