Munching Squares adds a constant factor in the test word to the value in the accumulator, does some XOR bit operations and splits this into X (lower 9 bits) and Y (higher 9 bits) coordinates. – Just start and play with the test word switches (which are read anew for each iteration, so it's interactive).
Snowflake runs on its own without further input (but the stand-alone version has a few display option on the sense switches).
The Minskytron reads 3 sets X and Y factors for 3 oscillators from the test word on start and then runs on its own.
Each X and Y factor is made up of 3 bits, so there are Y1,Y2, Y2, X2, Y3, X3 on the test word from left to right, each 3 bits/switches.
While it's a rather short program, it's actual workings are pretty complex. It's kind of visual synthesizer, but probably more of a network feedback explorer.
This was probably written just when Minsky finished his paper "Steps Toward Artificial Intelligence" – so there's some plausibility to the latter interpretation.
I've an unfinished, rather extensive blog post on the "Minskytron" lying around.
(What may there to to be said about a few lines of code? 😀)
This also includes an extended version (by me), which allows to produce even more patterns and variety.
The unfinished and unlisted blog post is here (which also allows to play with the Minskytron, including the "Minskytron II" extensions for changing the operator inside the oscillators, which are mapped to the sense switches in the extended PDP-1 version):
https://masswerk.at/nowgobang/2023/minskytron
(This also allows to switch to a 24-bit resolution – just as the never-built PDP-2 would have featured. Switching to the higher resolution, patterns that have dissolved into "snow" in 18-bit resolution often reappear!)
Please handle links to that blog post with discretion.