Thinking with physical nodes

When an idea won’t come together, our founder lays out round labels on a desk. Each label holds one word. Related ones move closer; unrelated ones move apart. As the arrangement shifts, connections emerge that thinking alone wouldn’t have revealed. We call these labels physical nodes.

Where the desk runs out

The method has three limits.

• It can’t grow beyond the size of the desk.
• Clearing the desk erases the arrangement. To pick the thought back up, you have to rebuild it from a photo.
• A label can only be in one place at a time. Using the same word in another context means writing a second label.

A prototype on the iPhone

So we are prototyping an iPhone app, provisionally called MineField. Each group is a horizontal row of nodes, and the groups stack vertically. Arrangements are saved, so each session picks up where the last one left off. A node can belong to several groups at once, which means the same word placed in a different context is still the same node.

The first version leaves out image recognition. Eventually, we want to import a whole arrangement from a photo of the desk.

The larger idea

This prototype is the entry point to an idea about ten years in the making: a way to keep thoughts and experiences outside your head, so you can return to a line of thinking long after you set it aside. For now, we call it knowledge navigation internally. What the technology of the time couldn’t do is becoming possible, thanks to progress in vision models and large language models.

It shares roots with our work on organizational memory. That work is about memory that lets an agent resume a task; this is about memory that lets an individual resume a thought.

Open questions

• How much of the feel of moving things by hand, and remembering them by place, survives on a screen?
• What counts as a single node — a word, a concept, a question?