I often think about this Picasso art pice:

I have some more extended thoughts and examples over in Gerard Thomas and conceptual ideas.

Applying the same deconstruction logic to code yields some of the following:

val database = map[string, array[row]]

fun execute(structured_statement)

This should present as a SQL database.

index = page_rank(internet)

fun search(query)

This is Google’s search product. Now this requires you to know that Google search uses the PageRank algorithm… but doesn’t the Picasso example require you to have seen a bull? Are those two comparable?

I am leaning towards no, as PageRank is an internal thing of Google’s tech (however famous it may be). As a layperson you see the search box, you don’t really delve into the algorithm. Given Google’s popularity, would it then be possible to just do:

fun search_internet(query)

to represent it? Would anyone seriously consider “Bing” or “DuckDuckGo” being the answer here?

In Gerard Thomas’s examples, you need a certain amount of exposure to car culture to “get it”. If you’ve never seen the car he’s referring to, you cannot “get it”. Picasso’s example works similarly. So far we’ve operated under that assumption. The artist simplifies the object’s gestalt, its concept or “entireness” if you will down to a simplified form that can impress upon you the idea of the whole. Notably, some research on that was done by Norbert Wiener, and you can read about it in Chapter 6 of my Thoughts on reading Cybernetics.

We need to go deeper

What if we broke that? What if I tried, through code, to make you envision a concept that you have not (necessarily) encountered before? A tech tool that doesn’t exist and yet you can imagine it existing based on these few lines:

fun search(query)
	result = actual_search(query)
	if result == null
		return imagined_result(query)
	
	return result

“In some ways the above is not far from an LLM”, I can hear you cynically commenting, and yet, to be far, it’s not quite one, is it? It’s just a search engine that always finds a result, whether one exists or not. Is it useful? Probably not. Is it what you the concept you got in your head? Unclear to me!

person = population.random()
soulmate = population.random()
while match_quality(person, soulmate) < 0.9
	manipulate(person)
	manipulate(soulmate)

What do you get from this description? What’s the idea of this “app”?

In my head, it’s a dating app that works by manipulating its users, rather than matching them. Now that description, “a dating app that works by manipulating its users rather than matching them” is 76 characters. The example above is 145, nearly double the length.

Are we showing that code is a poor abstraction choice because it’s structured text, and unstructured text using the full power of language can carry far more signal per character? Probably, though if we’re being honest, a more accurate description would be “A search app that selects random users, pairs them, then manipulates them until they are a good enough match” which is 108 characters to the 145. Nevertheless the fact that unstructured text compresses better is as meaningless here as the fact that Russian, or Spanish may use less characters on average to convey a concept. So what?

On the idea of abstraction

Abstraction or summarization is, in the end, just some variations of reversible lossy compression and the fact that Ideas are uniquely expressed in mediums doesn’t bother us because we are only doing text->text abstracting here, right? The lossiness doesn’t bother us either because all human communication is lossy. Yet compression, as a computer process functions because we know the algorithm to reverse it! Abstraction or summarization in this sense, when targeted to humans has the magic extra quality that… we don’t. Lossy compression is a balance between signal length, signal accuracy, and compute required to process.

This is why Picasso’s and Thomas’s art examples are interesting, and this is what Wiener was studying. The human algorithm that decompresses an image representation into an idea worth thousands of words, sounds, emotions, and memories in your head. This is why the idea is fascinating to me. Imagine, being able to command such information density in your everyday speech!

In one of my favourite comics, “Red Son”, by Mark Millar, the “villain” Lex Luthor has created a note which he manipulates Superman into reading. The note is this:

“Why don’t you put the whole world in a bottle too?”

The note is created as a sort of basilisk, a brain virus or perhaps a zip bomb which when processed by Superman’s brain is made to mentally crush him (it does). It is tailor-made to Superman himself, and using the grand total of all his insecurities, past, and fears to manipulate him.

What if you did that not manipulatively, but cooperatively? What if you could do this to make your friend, or your lover feel good? “Well Alex, that’s just called a good love note” - maybe?

In school I was taught how to summarize. There is some craft to it. Picasso summarizes an idea, and there is clearly some art to it. I would like to learn how to wield the two to create messages that can decompress into something valuable.