M.024 The paper that responds
How writing forgot it was dialogue
You are reading Molekyl, finally unfinished ideas on strategy, creativity and technology. Subscribe here to get new posts in your inbox.
Picture a writer at work. Most likely, you envision a solitary figure at a desk, scrambling their mind for ideas, wrestling with words and sentences. Calm on the outside, fighting an epic battle with their ideas on the inside. The image of writing as thinking in solitude.
This image is so deeply ingrained in our culture that we rarely question it. Writing is a solitary act. A battle with a blank page. That’s how it works. That’s how it has always worked.
Except it hasn’t.
Dialogue was thinking
If we put on a historic lens, the image of the solitary writer suddenly appears more as an artefact of modern times than representative of how thoughts and ideas usually have been crafted by humans.
Back in the days, writing wasn’t thinking. Dialogue was.
It was dialogue on the African savannahs when early humans made plans to relocate their tribe or catch a large animal. It was dialogue in the grand cafés of Europe where intellectuals debated ideas back and forth, long before anything reached paper. And it was dialogue in the universities where researchers sent letters back and forth, sometimes for years, bouncing arguments until something clarified.
Socrates is perhaps the clearest example. He took to the streets to work through an idea. He found someone to talk to. Said something, got a response, started a conversation. Dialogue as a thinking tool, used to beat half-baked ideas into shape by bouncing them back and forth between different perspectives.
For Socrates, and for most of human history, the dialogue was the thinking tool. You talked, you debated, you exchanged ideas. The written artefact, whether it was a book, a treatise or a paper, was the crystallised product of this thinking. The evidence that thinking had happened. With low literacy and expensive writing materials, it made perfect sense to postpone (or drop) any crystallising until you had a pretty good idea of what you wanted to say.
Writing is dialogue
When literacy spread, pen and paper became cheaper, and later, the typewriter arrived, things started to change. Suddenly thoughts could be put on paper faster than before, and with that speed it became easier to think and type at the same time. But it didn’t remove the dialogue. It just changed it. The dialogue turned inward, and became more a conversation between your own mind and the paper in front of you.
Still, the typewriter had friction. The ink on paper was permanent, and every keystroke was a small commitment. You could toss a page you weren’t satisfied with and start over, but the fact that you needed to start over on that page meant that you had to think before you struck the key.
The computer with its word processors later removed that friction. Suddenly, every word, sentence and paragraph could be changed instantly. No more sunk costs with putting half-baked thoughts down on paper. We could think on screen in real-time. Seeing our thoughts as they formed, rearrange, delete and rewrite. Endlessly.
But conversing with a word processor is a very different type of dialogue than those practiced by Socrates on the streets of Athens. It’s one where your partner doesn’t push back. Your partner, the digital paper, never challenges your words. It never asks the questions that force you to sharpen your arguments. And it never brings reflections that might push your thinking somewhere unexpected. It just sits there, patiently waiting for you to take the role of both sender and receiver.
Over time, this self-contained dialogue, started by the pen and paper, continued by the typewriter and accelerated by the computer, became our default image of writing. The writer alone at the desk. Thinking in solitude. Wrestling with words in silence.
As that happened, we gradually seemed to forget that external dialogue was ever part of the writing process. The conversations that shaped both thinking and writing for centuries slowly faded into the background.
Until the large language models appeared. And suddenly, the paper could talk back.
The paper that responds
More than anything, large language models are machines you can converse with. Intelligent partners each of us can turn to in the process that precedes putting words on paper. The dialogue part of writing has made its return.
A paper that talks back. A paper that helps us improve our inner dialogues. And, when we want it to, a conversation partner that can bring other perspectives, ideas and a critical eye that challenge our own thinking. The kind of partner that lets us test half-baked ideas, explore our own assumptions, and follow the rabbit of our thoughts to see where it jumps. You say something tentative. It responds, reflects, questions. You elaborate. You follow the thread. Ideas sharpen in ways they wouldn’t have if you had just been staring at a blank page, talking to yourself.
This way of working with large language models is a modern echo of Socrates in the marketplace, and of the old scientists and writers exchanging letters. It’s a quiet rebirth of dialogue as thinking technology. Restoring the conversation’s role in the writing process. A role that got backgrounded by the launch of technology, is now brought back by another.
The fast typewriter fallacy
The issue, however, is that productive dialogues with LLMs don’t happen out of the box. The reason is that LLMs are tuned to be as helpful as possible, and not to be your socratic thinking partner in crime. If you tell ChatGPT you want to write an essay about the role of dialogues in writing, it will reply with — you guessed it — an essay about the role of dialogues in writing.
When LLMs write for us this way, we are jumping directly to the crystallisation of an idea, while skipping the entire process that should make that crystallisation worthwhile. In a sense, it is like asking someone to write a summary of a conversation that never took place. It is therefore unsurprising that the result of “LLM-writing” more often than not feels generic and soulless, and not like crystallised thinking. The kind of text that any sceptic of AI rightfully worry about.
Even if the technology isn’t helping us out of the box, it’s not to blame. We are. All it takes to turn our LLM of choice from an automatic typewriter to a dialogue partner, is to tell it to. But more often than not, we don’t. We treat our new dialogue partners as fast typewriters, and complain that its writing is shallow and generic.
The greatest potential of LLMs is not as fast typewriters, but as conversation partners. That never get tired. That keep an open mind. That can challenge and sharpen our thinking as much as we want them to.
The paper responds now. What we do with that is on each of us.