Sidechain
In the game Super Mario World, there exists an exploit used by speedrunners called a credits warp. By performing a certain sequence of actions to spawn sprites at exactly the right pixels, the player can effectively write machine code through the game interface and then cause it to execute (for details see this video). This can be used to trigger the end credits scene, but also to trigger other events, or code in other games, or anything a programmer could do with assembly on a SNES console. The player/coder is nominally playing a game, but encoded somewhere within their actions in game is a message in a different language, with a very different effect.
The Jumbled Part
A few days ago I wrote a disjointed and schizophrenic post about “geometric psychology” (title pending). This essay was a personal experiment, an attempt to reorient my inner perspective when looking at myself and the world I live in. I talked a lot about dimensionality and regressors but implicit behind it all is the extended metaphor of people-as-geometric-models. My hope was that, by trying to relate cognition to mathematics, I might form some novel ideas about both.
But what ended up happening was much worse. The more I wrote, the further I strayed from English syntax. When I went back to try and clean it up, I found that copy editing pulled the text away from what I was trying to say. The easiest way to express my thoughts was also the most unreadable. It wasn’t just a matter of expanding shorthand into full speech; adding more words took away meaning. I did my best, but the end result was still a bit messy. At the time I chalked it up to laziness, and published anyway, since I have no editorial standards and no qualms about disappointing my readers.
A couple days later I received this message. I’ve been thinking about this message and the interaction that followed a lot. Is it difficult to read? I’m not sure — at first I thought so, but I found to my surprise that I had very little trouble responding to it, and we seemed to intuitively read each other’s meaning, even if we struggled with each other’s text. Another similar experience later convinced me — this isn’t simply a lazy shorthand, or, if it is, then it’s not one that only I intuit. There are other people out there who read this and speak it easily. This burned madman didn’t have the same computer science grounding I did, but immersed himself into the language nonetheless and mapped out new concepts I hadn’t yet considered.
Spelled out explicitly like that, this revelation seems more trivial than it did to me at the time. “Oh, you and your friends can talk to each other without using proper English grammar? And you have your own metaphors and references that you use, too? Congratulations, you’ve invented slang.” And maybe this is just a way of turning mathy jargon into slang, repurposing words in the same way that people do all the time online. Even so, I think there are some interesting things to learn from this encounter about the relationship between language and meaning.
It’s one thing to throw extended metaphors around, but this language goes a bit further than that. When Madman talks about dimensionality surge, he’s not using established geometric terms. This part of the language comes from general English applied to this metaphorical frame of reference post hoc. This part I understand because I can put those pieces side by side in my own mind and see the parallels. The contextual meaning of a term like this is not explicit, but it’s not hard to figure out, either. Worse is when we start using multiple meanings of the word “recursion”, or start tearing up grammar. Why do we keep drifting away from proper English?
The Interesting Part
Language is a way to turn thought into a message and back again. The meaning of a word is not in its written shape or auditory sound, it’s in our agreed consensus that the word means something. Language is like sheets of scribbled-on tracing paper layered on top of each other — shift one of the sheets, and you change the image.
The important thing about geopsych is that it’s not about using different words to describe the mind, it’s about using different concepts. The shift in language is incidental — what we really want is to start thinking different thoughts. We don’t want to use English to attack math or English to attack psychology, we want to use math to attack psychology. The actual language, then, is of secondary importance. Grammar is vestigial as long as the appropriate maths are understood on both sides. It would be possible, I believe, to have a conversation on the subject of psychology mostly in mathematical symbols and references to properties, theorems, etc., with few or even no words of general English.
English, written well, is a pleasure to read, but it can also be limiting. Language shapes the way we think, guides our values and attentions, is a major source of all cultural difference. Geopsych thinking and English thinking do not mesh well. To write “well” on geopsych would mean adding a lot of verbiage and actually distracting away from geopsych lines of thought. Better to exclude English words, only using them to refer to mathematic concepts which are the true semantic atom in this mode of communication. Things that would be crucial in regular speech —identifying the subject of a sentence, joining clauses properly, consistent tenses — are superfluous in this context. If you’ve ever said “sure, sure, I know what you mean,” you’ve experienced this disconnect between English and communication.
English is an ill-fitted pipeline for these thoughts, a repurposed tool. What we’re doing is hijacking the normal function of the English language to do something it’s not really designed for, just like the speedrunners in the opening paragraph used video game sprite positions to write assembly code. I call this process a sidechain, after the compression technique used by music producers to embed drum patterns in non-drum instruments (for example, this, this and this). Encoding critical information in a parallel signal is sidechaining. Piggybacking off of another channel of communication, hiding your information in it in a way that is only revealed through meta-analysis.
One parent spells out the words “ice cream” to the other so their excitable six year old doesn’t hear: sidechain. Producers fortify milk with vitamin D so that the milk doubles as a vitamin supplement: sidechain. A spiteful manager thanks every member of the team except one: sidechain. An opera singer shattering glass, a flashlight blinking in Morse code, the coloring in Sin City. There are sidechain-like structures everywhere, some more interesting than others. Spoken language is itself a sidechain in this sense, manipulating auditory signals to encode meaning. Geopsych is a sidechain on top of that, (at least) two layers deep. Any time something becomes more than its nominal self, there is probably some message encoding going on.
I don’t care very much about using things the way they are meant to be used, so let’s exclude those from the sidechain umbrella for now. The credits warp is still a sidechain, but spoken language is designed explicitly to encode meaning in soundwaves, best not to refer to it with our new term. A sidechain should be novel, as well as meaningful. Also, a sidechain should draw from one information source, encode it in a different medium, be received in that other medium and then decoded back to the original meaning. The parent spelling “ice cream” and the parent listening only join the letters into words in their minds: writing them out one by one on a sheet of paper would completely defeat the point.
I wrote about geopsych in English, but I didn’t want to communicate in English. I wanted to communicate in machine learning and maybe a bit of linear algebra. So I embedded all the machine learning talk into English, with little regard for how the English itself looked and a lot of care for how the math looked. It turned out the English itself didn’t matter, so long as the embedding was clear, which, at least for Madman, it was. Disjointed speaking is easier because, rather than translating algorithmic thinking into English thinking and back again, we relied on our mutual understanding of general English to use it as a medium in which we spoke math, just as sound is a medium in which we speak English.
The more I think about it, the less I believe my thoughts are novel. Someone else has doubtless done similar and more rigorous work before me. If you have thoughts on this, or know of someone who did a more thorough job than I, please let me know. Either way, I doubt this will be the last post in this series.
