Hash Function Saboteurs - Visualizing Narrative Spin
Label-Content Orthogonality and Flattened Maps
Language and Fire
Fire and language might compete as the most important developments in human history. Our mastery of fire and the rest of the material world has continued to grow along with our knowledge of its bounds. While our fundamental understanding of information is catching up to that of fire1, our techne for transmitting ideas doesn't seem to have advanced other than in the practical matter of distribution.
Ideas are slippery high-dimensional things. They are passed around and connected by means of words. Sensory input and other symbols (e.g. math notation) also play into the creation and manipulation of ideas. Minds create containers for ideas and label them. Ideas can wear multiple labels, labels are their own ideas and can be labeled themselves.
Communication is challenging for many reasons, but miscommunication due to different staking of words is a familiar one.
Verbal communication consists of roughly two stages:
Semantics (/pragmatics) for connecting words to objects
Syntax (/morphology) connecting words to each other in meaningful structure
Ideographic communication attempts to make symbols that explain their own connection to reality. These include some character/hieroglyphic languages and some mathematical notation (e.g. "=", "( )").23 Words composed from letters are practically beneficial for reducing memorization load and for allowing interesting recombination (e.g. learn 26 letters instead of 3,000 characters; prefixes don’t require different brush strokes, just add the letters in line).
We could go even further down the line of symbol to object-label realism with the help of Magritte and Hofstadter,4 but I'd like to go back up a level from the substance of the object of a label, to the processes of labeling and the methods by which this labeling is exploited. This is the world of label-object hashing functions.
Label-Object Hashing
Computer science presents the concept of the hash function, which we'll explore mapping onto the problem of labels.
A common example of the use of a hash function is the way that websites store passwords: when you log in, your password (key) gets passed through a hash function producing a hash. The hash function is intentionally designed to make it difficult to work backwards from a hash to the key it was generated from. This means that if the list of login hashes is stolen, the thief isn't much closer to logging in, since they need to enter the correct key.
A central problem in hash function design is avoiding collision - the case where multiple keys map to the same hash.
By now, you might be getting an idea of where we're headed.
The world of communication already suffers from the complexity of tenuous links between ideas and their names. Those seeking to persuade will push the limits of the connections between ideas and their labels by the following method:
Chose labels with the explicit purpose of antonym collision.
Deride opponents for using the (often reasonable!) trivial hashing function (hashing a word to what it means).
or
Deride opponents for signaling meaning through increasingly non-trivial hashing functions.
Demagogues play groups against each other by knowing which phrases are "hash function wedges." Every person runs a different hashing function and similar ideas render differently in each person's mind. This leaves us with the rough map - Key : Hash Function : Hash :: Label : Mind : Object/Idea5
Language Saboteurs through time
Bolsheviks
The Russian Revolution of 1917 against the tsar brought to power the Bolshevik party. The Bolsheviks were forked from the Russian Social Democratic Labor Party (RSDLP) as a stronger independent political party after originating as the hard Leninist wing of the RSDLP. After a marginal victory in elections of the RSDLP, the Leninists began calling themselves the Bolsheviks, which translates to "majority."
Discourse against the hard communists was subsequently weighed down by the literal reading of opposing "the majority." This is the same naming technique as other communist nations calling themselves “People’s” republics and so on. Our modern American political party names seem to have leveled out on this dimension since the names "democrat" and "republican" don't tell you much more than "whig" or "tory" without knowing the particular context.
Peacekeeper
If you are a nation upgrading your nuclear weapons portfolio with an ICBM designed explicitly for a retaliatory strike and the program was previously canceled, what should you call it?
The Peacekeeper missile program6 was successfully restarted in 1982 by Ronald Reagan after the program was terminated by Congress in Jimmy Carter's presidency. The program at that time was called MX. Peacekeeper missiles were installed in former Minuteman missile silos, which I must say sound more positive than MX missile silos.
Present Day
We’ll cover some present day examples in a moment, but I’d first like to present a visual framing of the effect that the language saboteurs are applying to the idea landscape.
Spin Visualized
When naming something to make it difficult to debate against, a saboteur’s goal is to tie an opponent’s point to something that is undebatable. You sabotage the debate by twisting the axes.
Take a simplistic 2D map of the underlying idea domain. It’s obviously difficult to say where any idea exactly goes, but understanding this trick doesn’t require it. Instead, we care about the effect of intentionally conflicted naming.
By naming your view with a moral positive or the negation of a moral negative, you rotate the map to look at it side-on and hide the full picture.7
Wanting war and making missiles are not the ends at odds. Leninism and rule by the majority are not aligned. Narrative spin is about hiding the idea landscape to collide labels. Debate about the content is obfuscated by directly involving a moral statement.
Density
Another pernicious aspect of the spun idea domain is the way it presents a false sense of the typical person to an ideologue. Here, an overwhelmingly good population slightly favoring View -A are recast by spin.
Anyone associating moral goodness and view A now see a distribution peak of middle-of-the-road goodness. Furthermore, the false map indicates that being stronger in view A gets you further from the immoral masses who lean toward -A.8
This false framing is obvious once you’ve separated the two components, but compared to the typical compression of topics to one-bit for-or-against positions, it’s a huge step up in breaking the trick.
Propagation
Logical systems, by their nature, are broken by just one falsehood, so the trick of the saboteurs is to use the emotional charge of an issue to slip in the logical contradiction. The emotionally implied version of the opponent's view then propagates out to break other parts of the underlying idea landscape.
Of course, there are real correlations between views held across issues, but failing to compartmentalize allows the contagion to spread. An imported moral invective can generate caricatured views of opponents along other dimensions.
Modern spin
BLM
What does “Black Lives Matter” mean?
The trivial hashing function (the literal reading) tells you that supporters believe that the lives of black people matter. Alternative contemporary hash functions pointed the “BLM” label to specific goals like "end police brutality." Simultaneously, BLM was also made to point to the idea "defund the police."
Remarkably, the most surprisingly part of the whole affair was not that defunding the police would likely make majority-black neighborhoods less safe (though, trivially, police brutality would be ended if there were no police).
The most surprising part is that the unspoken part of the literal hashing, black lives matter as much as any other lives, gets you to a statement - All Lives Matter - that BLM true-believers hash to the opposite meaning!
How could you be against Black Lives Mattering? It's anti-democratic to oppose the will of the bolsheviks (majority)!
It's almost cartoonishly simplistic, and yet it weighs down every conversation.
Those who named BLM knew what they were doing.
The non-trivial hashing of BLM points to empathy for people who have suffered at the hands of bad police (even though police and brutality are nowhere in the name).
Then the promoters of BLM advocate for positions like defunding the police using the support accrued by those using the non-trivial hash to empathy.
People who oppose these recommendations (object/ideas) get labeled as opposition at the level of the key, as "anti-BLM."
Then the same people who use the non-trivial hash function to understand BLM, use the trivial hash function on anti-BLM. They see this confirming their existing belief about their opponents - whom they believe don’t value black lives.
Some supporters get caught up in the rush, but others feel little smug about their branding, all the while hanging onto their uncharitable prejudice.
Leading up to the 2016 democratic primary, Hillary Clinton was bashed for saying "all lives matter." Bernie Sanders said "Black lives matter. White lives matter. Hispanic lives matter" in an interview and was later shouted down at his own rally. They changed their rhetoric.
What does this do? Groups with other afflictions take a different non-trivial interpretation- that saying BLM is intentionally leaving out whether white, Hispanic, etc. lives matter.
The whole thing is gross- intentionally inflammatory and polarizing (sorts people by interpretation). Now, finally the visual with a new question:
Again, the rotated map hides the real landscape of idea space from view. Take a moment to consider- who is this center person?
At the center of the sabotaged spectrum, the distortion is the most severe, projecting good beliefs and terrible ones as coincident. This is even worse when paired with the tendency to view this poorly defined “middle” as typical, as previously illustrated.
Again, this map doesn’t usually exist concretely in mental space, but the diffuse, emotional version and group bias cloud rationality to make your enemy the “other” or an average simpleton.
Anti-fascist
Now, surely most people aren't falling for the same trick of confusing negation at the key and hash level for an organization who puts the negation in the key as literally as the "Anti-Fascist" Antifa?
I wish I had some solid statistics on social media posting during the summer of 2020. Instead, I will just report that I saw Instagram story reposts with the sophistication to explain that anyone anti-"anti-fascist" is a fascist.910 sigh
“Anti” indicates a 180 degree rotation, but you have to be explicit about the axis. Subsequent “anti”s don’t necessarily rotate around the same axis.
Orthogonalism means not being complacent about which axes you rotate about.
In this case, the difference is between rotating to oppose the object held under the new label instead of just taking the opposite of the label.11
It may be illustrative to imagine both interpretations of an anti-antifreeze activist.
Pride
Pride is perhaps the perfect example of the wedge label. Textbook antonym collision.
Again, people often choose to see what they already believe, and only use labels to be more combative.
Liberals take the antonym of "pride" and get to the word "shame." They are proud of supporting people who were made to feel ashamed of themselves.
Conservatives take the antonym of "pride" and get to the virtue of "humility." They see people boasting, and boasting about something which they believe shouldn’t be.
The combativeness of the term is ramped up to draw attention. So, a general question is- who is looking for attention?
Trump
Yeah, that guy is. Trump mastered picking the statements that most acutely divide people. He also balanced which side was doing the trivial and non-trivial hashing to hide this.
Eric Weinstein brilliantly illuminated this by the analogy of a coin landing on its edge, over and over and over again.
In each case, here, you could also interpret alternate hashing or literal vs. implied as the order of the interpretation. 1st order interpretation is the literal meaning, 2nd order is what you get by contextualizing, 3rd order by contextualizing that, and so on.
This interpretation shows why there is confusion about whether Trump is an idiot or an idiot-savante or a savante-idiot-savante and so on.
The term Rorschach Test also pops up a lot, which is a great analogy in that it emphasizes how the ink on the page is the same. Nonetheless, people automatically recognize something in a way that reveals how they think.
I mention all this to say that the examples tabulated above are again just one way of going from first to second order. There is simply no such thing as a complete parsing. Presenting the mechanism and demonstrating a sample of mappings are ways to explain a pattern without claiming to know internal states of mind.
The Trivial Hashing
Technical names
It's not a trivial task to name something well-enough that the trivial hash function tells you what you need to know. This is where we get to the meta problem of this entire essay needing to be hashed as an idea. The maps become too complicated and recursive to untangle.
Here are just a few examples of well-named technical terms
Pulse width modulation. It could have been called Smoothed Digital Control or Binary Output Weighting, or some other marketing term that isn’t as good.
Metadata. Data about data.
Shock Response Spectrum. The distribution of amplitudes of responses to a shock impulse by mode frequency.
Chicken Noodle Soup. Like it says on the tin.
Working in technical fields, you can get accustomed to the benefits of a concrete space, though science naming can suffer from hieroglyph and saturation problems.12 Acclimating to ubiquitous literal labeling likely elevates susceptibility to deception by mislabeling.
Axis Twisting Arms Race
You are now armed with the familiarity you need to diffuse the traps of mislabeling and axis twisting. The brain is a pattern-matching machine and once you see a new pattern you can spot it other places. However, you don't want to pattern match too strongly since that opposes the compartmentalization necessary to evaluate new or distinct topics in the first place.
Every tool in deception is another round of an evolutionary arms race.13 That is, you can take each concept out another order.
Under damning names like conspiracy, and perhaps some less emotional name, compartmentalization is used as a tool to hide bigger patterns. The arms races of ideas, communication, and deception all continue at many different levels.
Finally, these examples are not perfect, they are chosen datasets that show a way to turn the data and find a very well-fit function. I don't consider this cherry picking, but rather applying the necessary scientific view of attempting to isolate an effect. As previously mentioned, the world is extremely high-dimensional, so there may be other strong explanations. The world is also structured to allow many-step-deep self-reference. The best you can do is to apply the theory to more complicated problems and see what other patterns you can find.
I particularly like the framing of the paired tools to move up and down self-referential systems like orthogonality-mimesis (from my last article), axis hiding, and now compartmentalization-conspiracy.
Theories develop by broader testing, so I'd love to hear about patterns that these tools explain or new tools you discover.14 Fare well, fellow function finders!
Physics and math can play with their symbol selection even further. For instance, the parallels of the equations for electric field : electrostatic force :: electric potential : electric potential energy, don't just share their equation structure. My AP physics teacher asked the class to notice other similarities, and I was pleased to be first to realize that the symbols are visually analogous:
Nature is a beautiful subject, but the painter has license to be artistic too!
My relative ease in understanding visual/spatial structure compared to written work makes ideographic language particularly salient for me. Why is the structure of writing so… ugly? It is difficult to make a picture/visual that illustrates something precisely. Great mathematicians’ and physicists’ notebooks have all sort of elegant diagrams, but then get published with sentences and paragraphs to explain. This fascinates me.
Science fiction has also raised this point in exciting ways in the excellent trilogy Remembrance of Earth’s Past by Cixin Liu and in a limit case in the film Arrival.
Ceci n'est pas une footnote. Go read Godel, Escher, Bach
This is not a perfect analogy for a number of reasons, but a primary technical consideration is that hashes are fixed length, so it's more like they hash to idea addresses. The difference is immaterial for simplistic one-bit hashing function (eg. good or bad), but is not general.
Peackeeper missiles were intentionally designed with extra warheads to make it tactically relevant as a retaliatory strike where only a limited number of the weapons would be expected to survive the first-strike. This sounds quite gruesome or vengeful, but is just a practical fact of being a credible deterrent.
The two dimensional map's edge is pointed at opposing parties. Funny enough, it's geometrically dual to Eric's coin on its side language (coin on its side seen from two perspectives vs the true map seen edge-on).
This still isn’t quite a direct comparison since the coin goes to further to say the both sides are receiving different messages. More on this part, a bit later on in the essay.
The fact that people seem to fall for the trick shows how strong the urge is to consider “the other” not just wrong, but to see “the masses” from a snobbish, pretentious perspective as less moral.
Anecdotally, animosity can come from an indignation or belief in vengeance on behalf of someone else more easily.
XYZ group has suffered, group ABC can bear the consequences to balance it out, they’re _insert insult here_ anyways. This kind of thinking, no matter how well-intentioned for group XYZ, does not put you on the maximally good edge of the basic 2D idea domain.
I think seeking revenge for others’ sake somehow dramatically helps you miss the unwisdom of revenge. Forgiveness is the orthogonal solution.
An open social media graph (or big tech insider) with LLM-powered interpretation and tagging should be able to generate some killer correlation metrics.
“Anti” might also be improperly used to mean “non.” While 1, 0, and -1 are next to each other, they are very different. Natural, Whole, Integer. Furthermore, no rotation around the origin gets from 0 to 1.
This is the same confusion as the “=” of math vs. programming. Labeling something is more like instantiating an object and pointing something into it. As previously mentioned, everyone is running their own hashing function, but in this footnote’s analogy is more like everyone is running a different compiler. There is no global way to throw errors for all the collisions.
Yet another code analogy might be a datatype mismatch. It’s easy to write negated labels. It might be impossible, undefined, or a simply an ill-formed request to try to invert a big object.
x = giganticNDimensionalArray(idea)
antiX = not(x)
antiAntiX = not(y)In exchange for discovering and characterizing a truth of the universe, scientists often get a discovery named after them. I think this is a fair prize, but over time the library of historical terminology without literal verbal ties to the content grows. The problem can certainly end up like the character/hieroglyph memorization problem.
Godel's Incompleteness Theorem has a much better ring to it than Symbolic Recomprehension Proof of the Mutual Exclusivity of Consistency and Completeness
In an early version of this essay’s illustrations I went looking for this type of diagram
I could have hand-drawn, but I knew that this type of graph was common enough to exist in a way that I could make it with a python library, but I couldn't find the name of it to search for it!
I searched for "percent flow graph", "input output graph", "flow splitting graph", and others before switching exclusively to image search and trying "percentage flow graph category" where I scrolled until I found an example of what I needed.
It turns out that it's called a Sankey diagram.
An alternate name is alluvial graph, which does have a literal meaning close to "flow splitting" graph, but scrolling through Google images and letting my eyes do the work to find an example was definitely faster than trying to find “alluvial” in a thesaurus.
Because science does the work of explaining new and often invisible things, I think made up words are difficult to avoid and even warranted, making the names of their discoverers seem more reasonable.
Now I like the term alluvial graph, but either way once you know the term “alluvial” or “Sankey” you can start recombining to name other new things: “Income-Expenditure Sankey diagram”, “Evolutionary alluvial diagram“
Footnote footnote:
There is one other problem that can arise from naming things after the people who discovered them: Leonhard Euler.
Honestly, this is ridiculous. Leave something for the rest of us, Leon.
The stability of the billion-year-long evolutionary arms race of life is just incredible. The fossil record notes extreme extinction events of macroscopic life, but I wonder if the early microbial world had any close calls. Perhaps the density and volume of the ocean was enough of a curtain to wall-off sections of the then-small biosphere to avoid crashing the experiment. Here I am back at another Eric Weinstein question, now on the necessary threshold of isolation for life to run in parallel unaffected tracks.
This has been a long overdue entry to work in explaining function findings and to operate on the warnings of Eric Weinstein on the paucity of proper sense-making. Thanks in the same vein to Peter Thiel and Balaji.