AI and I

As a child, I learned to program in BASIC on my father’s Commodore 64: If X is Y then go to Z. It gave me a basic understanding of computer logic. If you made a typo in a 160-line program, it wouldn’t work anymore, and you had to find the mistake yourself (the computer couldn’t do it for you). I got those programs from a little book I borrowed at the Voorburg library—proof, really, of the idea: “Discover what you’re capable of” at the library.

I went on to study history in Leiden. My specialization in Russian history came to a halt when unrest around the 1996 elections prevented me from traveling to Russia to conduct archival research. I decided to spend that unexpected free time on something I simply enjoyed: books. Under Adriaan van der Weel, Berry Dongelmans, and Paul Hoftijzer, I was introduced to the history of the book. I became fascinated by the interplay between ideas, technology, and society. Later, when I began working at the KB National Library, it was partly because those three threads came together so well there.

I also noticed that, although programming languages had become much more advanced, the logic I had learned with BASIC still held true. In the early 2010s, colleagues Theo van Veen, Willem Jan Faber, and Rene van der Ark at the research department under Paul Doorenbosch were experimenting with named entity recognition and machine learning—and with vector databases instead of relational databases. I followed it all with fascination. Just imagine what would be possible if we could apply those techniques to all the texts preserved by the KB—or better yet, to all collections in the Dutch Digital Heritage Network.

I consider myself lucky to live in a time of technological change and the rise of new information technologies. I tried to understand what that meant for my work—particularly everything that happens between an author and a reader. In 2014, I wrote an article about what I called the Creative Communication Cycle, exploring how people communicate through text. I went back to school for a master’s in Managing Information and Sustainable Change (a sort of social business studies for the information society). For my thesis, I ran a scenario-planning process on the future of the general book, together with a wonderful group of people from the book industry. I concluded that the concept of the book is changing under the influence of technological development—and that this, in turn, impacts the future of the book.

In 2015, I gave a presentation on visualization as the next step, using the NGrams developed by Theo and colleagues, at OCLC Research at the request of Titia van der Werf (incidentally also the first time I had the pleasure of meeting Marc van den Berg). I argued that every expansion of available information is followed by a new technology to tame that abundance: lists were followed by catalogues, catalogues were enriched with classifications and thesauri, those were followed by search systems, and in my view, visualization was the next step.

In 2016, Mike Kestemont discovered the real author of the Dutch national anthem Wilhelmus: Petrus Datheen—something that, as someone who sang the Geuzenliedboek in primary school, seemed perfectly logical to me. In 2021, I read You Look Like a Thing and I Love You—which, I hope, gave me the same basic understanding of AI that BASIC had given me about computer logic. It works with probability: so a sheep on the backseat of a car is likely to be classified as a dog, while a white chihuahua in a field might be mistaken for a sheep. The process used to develop this remains partly a black box, as @erikgroeneveld recently explained to me—the concepts are too abstract and contain too many unknown variables. And fundamentally, it lacks meaning: that’s why it thinks “You look like a thing and I love you” is a great pick-up line…

Meanwhile, the KB continued working on AI. In 2020, Jan Willem van Wessel published seven AI principles. And then came ChatGPT, in the winter of 2022. My first question to it was: “What would Treebeard the Ent think about the nitrogen crisis?” (Treebeard, in Tolkien’s classic The Lord of the Rings, is a tree-herder, wise and old). It answered: “He wouldn’t think anything because Treebeard is a fictional character and a tree and therefore can’t think.” That was clearly nonsense, so I left it at that. A few months later I read—probably here—that AI is basically a mansplainer: “It talks with great confidence about things it knows hardly anything about.” That didn’t make it more appealing to engage with.

Until I took part in CENL in Warsaw in 2024, where we discussed the preservation of electronic publications. Several libraries told me they were already dealing with a substantial number of AI-generated publications. Because many countries have a legal deposit requirement, they were obliged to include them. That meant they had to make an effort to process those materials—time that, in my view, should be spent on the core task of national libraries: enabling interaction between human minds across time and space, as I described in this post (in Dutch).

It made me uneasy. For the past three years, the KB National Library’s brand story has been: Built with words – shaped by people. Every day as I walked up the stairs from the National Archives to the KB, where this sentence is displayed, I realized this might no longer be entirely true. It made me uncomfortable. It helped a little to realize that AI is a tool—one that can support every step of creative communication with text.