Luhmann's Zettelkasten

A small language model?

[This article is a work in progress. I decided to write it in public to optimize the publishing workflow under real world conditions.]

I spent the better part of my time at university researching what some scholars used to call “the most advanced artificial intelligence of our present day”: a card index maintained by the German sociologist Niklas Luhmann.

He used this Zettelkasten to collect, organize, and access vast amounts of hand-written notes that he based his entire theory of social systems on. Sometimes, Luhmann half-jokingly remarked that his Zettelkasten is doing all the writing for him.

This specific remark got me thinking. After all, Large Language Models (LLMs) are doing the writing for millions of people.

With ChatGPT, Google Bard and other, so called, artificial intelligence chatbots seeing such widespread adoption, this made me wonder: How does Luhmann's Zettelkasten compare to, and differ from, Large Language Models?

---

Let me start by roughly outlining the working principles of Luhmann's Zettelkasten:

• Individual notes focus on a single idea or concept
• Notes are not organized hierarchically or categorically
• All notes have a unique address that makes linking of ideas and concepts possible
• Through linking, notes find their place based on connections and relevance to other notes
• Instead of maintaining a coherent content structure, the focus is on maintaining connections between ideas
• Notes are revisited and expanded via follow-up notes that go deeper on specific topics
• By adding new notes and linking them to existing notes, a networked structure will emerge
• This structure will eventually start acting as a secondary memory for the user
• This memory follows its own internal logic that is familiar enough to be accessible while being different enough to resurface unexpected ideas and concepts
• The memory is queried via prompts formulated by the user

The two most interesting aspects for this article are the emergent structure and the prompting on the user's behalf.

I won't go into more detail, as it would go beyond the scope of this article. In case you are curious, here is some further information: Introduction to the Zettelkasten Method

Why did Luhmann start maintaining a Zettelkasten?

Luhmann's motivation for starting a systematically organized collection of notes was about the "technological and economic limits of scientific work". At the outset, he was looking to establish a system that made collecting, processing and accessing large amounts of information easier.

For this system to be reliable it needed to have a well established structure. A structure, simple enough to not get in the way of writing but flexible enough to accommodate unforeseen detours, branching thoughts and dead ends in his research.

Luhmann wrote that: "Without writing, one cannot think. At least not in a sophisticated, interconnected manner." ¹ —another hint as to why he choose to take note-taking so seriously. He was looking for a way to become a better, more efficient, more independent knowledge worker.

When did Luhmann's Zettelkasten turn into a "Kommunikationspartner"?

One thing that Luhmann may not have anticipated when he first started his Zettelkasten 1961, is that he would later come to call it a "Kommunikationspartner". When he began his scientific career, he may not have had a clear understanding of communication, and he also may not have known that his Zettelkasten would play such a prominent role in his scientific pursuits.

He pointed out that it needs to be trained with a sufficient amount of information to develop a certain degree of autonomy, eventually becoming a valuable research assistant or a “Junior Partner” of sorts.

Luhmann goes on to describe describe interactions with his Zettelkasten as prompt-based. He would come up with a question first. In response to that prompt, the Zettelkasten would produce a number of relevant note cards. These note cards …

What did it enable him to do?

• It made him productive
• Connecting seemingly unrelated ideas in a meaningful way that would fit with the
• It gave him autonomy -> Pursuit of his goal
• It began to mirror the structure his theory
• Bias

Combined, these observations paint a picture eerily similar to our interactions with LLMs. However, there is one crucial difference between Luhmanns Zettelkasten and Large Language Models: Luhmann painstakingly controlled which data would find its way into his note taking system and linked it himself. Every single note was hand-written by him.

I recently came across an article by Oliver Reichenstein called the End of Writing. While the title might suggest that AI is here to put and end to all writing done by humans, the article instead asks the question: Why do we write? What is writing an end to? Reichenstein's answer: "The end of writing is to understand each other through time and space." [Source] To answer it with Luhmann: Writing is an end to thinking.

Ted Chiang suggests calling it “Applied Statistics”