18 Feb 2024

Infocom games and accessing knowledge

Another brick in the usefulness of the graph model of knowledge: back in the 80s, I played various Infocom games – essentially, interactive fictions that you played with a command prompt. You could type OPEN BAG, for instance, and the game would tell you what was in the bag.

(This genre was mostly moribund until it was revived and perfected by Homestar Runner’s Peasant’s Quest (which is still online), which allowed the command THROW BABY IN LAKE. Don’t worry, the baby comes back out.)

Anyway, the reason I’m bringing this up is that these games were, of necessity at the time, incredibly precise. Say you entered a room and were told there was a newspaper on the table in front of you. Naturally, you READ NEWSPAPER.

YOU DO NOT HAVE A NEWSPAPER, the game replies.

First, you have to GET NEWSPAPER, and only then can you READ NEWSPAPER. (This is less annoying when you get into the flow of what the game expects and assumes, but until you get to that point it’s very close to “This dad follows his kid’s PB&J sandwich instructions very literally” SLYT)

All of this is to say: our memories are closer to an Infocom Games parser than we often assume. Retrieving knowledge – especially when we’re first acquiring it – can be incredibly context-dependent. I have some Anki cards to help me learn about Merkle trees, but they all operate from the direction of the label (“What is a Merkle tree?,” “What are Merkle trees for?,” etc.).

It took me fifteen minutes this morning to remember the name when I started from the point of how the data structures worked instead of the other way around.

So: back to graphs. Ideally, we link concepts (or in this case, labels and the things-labeled) bi-directionally, allowing us to get to the knowledge we need regardless of where we start.