Towards Semiotic Physics


While on a run, Achilles sees his friend the Tortoise reading a thick book under a palm tree. As he approaches to say hello, he notices the Tortoise also has an elaborate network of papers and strings on a cork board leaning against the tree.

Achilles: Hello my dear friend! What is it that you're doing with this tome? And this cork board - is it a piece of abstract expressionist art?

Tortoise: Achilles! What a wonderful surprise! No it's not art - it's a diagram. I am working on a new scientific theory I call semiotic physics.

Achilles: Semiotic physics? What on Earth does that mean?

Tortoise: A wonderful question! Semiotic physics is the study of the dynamics of sign production. In a way, it tries to provide a general answer to the question "What on Earth does that mean?"

Achilles: I'm not sure I follow.

Tortoise: Let me give you an example. Before you approached me today, I imagine the phrase "semiotic physics" was senseless to you.

Achilles: Indeed it was.

Tortoise: But now that we've had some conversation about it, your understanding of the meaning of "semiotic physics" has changed. The image that the term "semiotic physics" evokes in your mind may be blurry, but a blurry picture is much different than no picture at all. Semiotic physics is concerned with how the meaning of terms, or more generally the meaning of signs, evolves as they are used to communicate.

Achilles: Ah I see! The picture I conjure up in my mind when I hear "semiotic physics" is becoming less blurry by the second! Can I give you an example to test my understanding?

Tortoise: Certainly.

Achilles: When I approached you, I asked if that cork board was some kind of abstract expressionist art. Now that we have talked, I can identify it as some kind of scientific diagram. In this way, the meaning that the cork board conveys to me has been changed by our conversation.

Tortoise: Perceptive as always Achilles! That is exactly right. This cork board is a diagram of a communication network. I am using it to reason about how meaning might evolve as several entities communicate with each other.

At this moment, Achilles' and the Tortoise's friend the Crab happens by

Crab: Achilles! Tortoise! What a pleasant surprise! Are you making some kind of abstract expressionist art on that cork board?

Tortoise: Crab! How marvelous! I wasn't originally intending the cork board as a piece of abstract expressionist art, but I'm starting to think it might be one.

Achilles: Hello Crab! How comical - I said the same thing about the cork board when I first saw it.

The Tortoise picks up his notebook and pen

Tortoise: Achilles, want to explain what we've been talking about to our friend Crab?

Achilles: Of course! The teaching process is one of the best ways to learn, after all. We're talking about semiotic physics, which studies how the meaning of signs evolves as they are used to communicate.

Crab: How fascinating! Please tell me more.

Achilles: Well, when I first happened upon Tortoise, I didn't know what semiotic physics meant. But after talking with him, I gained some idea of what it means. Now, I'm communicating that meaning to you. Semiotic physics tries to describe how the meaning of signs, like the term "semiotic physics," spreads and changes as more people communicate with each other.

Crab: Ah! Can I give you an example to test my understanding?

Achilles: I see you're a student of Feynman as well! Please do.

Crab: Both of us have described the cork board as a piece of abstract expressionist art. While this was not Tortoise's intention, the fact that both of us have described it as art to Tortoise has changed the way he thinks about it. In this way, the meaning that the cork board conveys to Tortoise has changed as a result of our conversations with him.

Achilles: How insightful! I would say that's an accurate example. What do you think Tortoise?

Tortoise: I think that's quite right. I'm also thinking I should show my cork board to our dear friend Jackson Pollock.

Crab: I think that's a great idea.

Achilles: I have another question, Tortoise.

Tortoise: Ask away.

Achilles: Now that I have talked with Crab, all three of us have "pictures" in our mind of what semiotic physics means. How do you intend to take measurements of the similarities and differences between these pictures? Science requires measurement, after all, and the pictures are stuck in our heads.

Tortoise: Astute observation. The beauty of semiotic physics is that we don't actually have to measure the "pictures" to understand what a sign means. If you accept that the meaning of a sign is based on its use, as Wittgenstein does, we simply need measurements about how signs are used to understand their spread and distortion.

Achilles: So regardless of what pictures Crab and I have in our heads of "semiotic physics," the fact that we are using the word tells us something about its spread.

Crab: And the fact that Tortoise is now using the term "abstract expressionism" to refer to his cork board tells us something about how his picture of "abstract expressionism" has been distorted by our conversation.

Tortoise: Precisely. Based on how you are using "semiotic physics," I'm quite confident that you understand the spirit of my theory.

Crab: I'm still curious about your book, though. What are you reading?

Tortoise: This is a copy of A Thousand Plateaus by Gilles Deleuze and Felix Guattari. It's most certainly a piece of abstract expressionist art.


Introducing Semiotic Physics

Semiotic physics is the study of the dynamics of sign production. As we saw in the prelude, ideas like "semiotic physics" and "abstract expressionism" have a tendency to spread and evolve by way of communication. The fundamental postulate of semiotic physics is that we can extract precise dynamical laws describing the spread and distortion of signs by measuring how they are used.

Language stabilizes around a parish, a bishopric, a capital. It forms a bulb. It evolves by subterranean stems and flows, along river valleys or train tracks; it spreads like a patch of oil

Sign Producing Machines

The fundamental unit of study in semiotic physics is the sign producing machine. These machines can be biological (e.g. humans), mechanical (e.g. language models), or cybernetic (e.g. a human-language model assemblage).

In general, the boundaries of any particular sign producing machine are arbitrary; the thing that we care about is the flux of signs through a machine. In the same way that Gauss' theorem can be used to study the flux of a vector field through any closed surface, semiotic physics can be used to study the flux of signs through any sign producing machine.

A Research Agenda

Semiotic physics is a very, very young research field, with a plethora of open questions. Below, I present four open problems that I believe will form a strong basis for further work (if we can manage to answer them). If you have interest in working on any of these problems, please reach out to me at brandon [at] nomic [dot] ai and I can see about getting you research resources.

1) Develop and Analyze a Space-Time Language Model

Two of the axes that I want to understand sign spread and distortion along are space and time. A critical tool for performing this analysis is a Space-Time Language Model, or a language model that explicitly conditions on the geographic location and publication date of its input text. As I mention in Violence, The State, & Empirical Ethics, such a model could be used to investigate the likelihood of statements across space and time by varying the conditionals. I believe that this model would yield several empirical insights that could ground further formal work.

2) Characterize the Semiotic Flux Through a Single Language Model

UPDATE 11/10/23 : There is a research group actively investigating this question

Perhaps the most important sign producing machine of the next hundred years is the large language model (LLM). A precise characterization of how LLMs distort the language that passes through them will therefore be critical to understanding the process of sign evolution going forward. There are several ways to approach this question. One is to characterize the precise relationship between prompts and responses. Another is to characterize the relationship between training data composition and decoded sequences. In the limit, we will want to have a strong understanding of both.

3) Simulate Semiotic Procession in a Population of Language Models

UPDATE 11/10/23 : There is a research group actively investigating this question

Semiotic procession is a process by which the meaning of a sign shifts as it is continually used in discourse. The simplest example of semiotic procession is the game of telephone, where a message becomes increasingly distorted as it's passed down a line of people. A simulation that demonstrates this phenomenon using language models would serve as a basis for further experiments on semiotic distortion.

4) Create a Taxonomy of Existing Language Models

UPDATE 11/10/23 : Daniel van Strien has made first draft taxonomy based on text embeddings of Hugging Face model cards!

I conjecture that, in the same way that human beliefs can be taxonomized into schools of thought, language models can be taxonomized into genera. Moreover, I believe that language models that are close in this taxonomy will have similar distorting effects on the language that passes through them. Therefore, a taxonomy of language models would allow semiotic physicists to focus their research on models that sit at strategic points within the taxonomy.

Addendum: Why Study Semiotic Physics?

If Gaia is the organism, then we are her cells, and our roads and rivers and cafes and discord servers are her nerves, and our words are the action potentials that drive her behavior.

As Deleuze and Guattari say, "the elementary unit of language is the order word." In this way, linguistics is indistinguishable from pragmatics. Words are mechanisms for driving coordinated action.

Historically, several regimes of signs fought for control over Gaia. The tribe, the church, the state, the Allies, the Axis, the West, the East - and when new tools for the engineering of language and thus the engineering of action are created, the existing order is vulnerable to revolution. The printing press was the axe that Martin Luther used to fell the church, and as Herbie Hancock says about syntesizers in jazz: "It's a tool - the way an axe can be a tool to build a house, or it can be a tool to slaughter your neighbor."

Recently, a new genus of cell has been introduced to Gaia's body - the large language model. It's clear to all of us that this cell is quite powerful. It is the most powerful engine humans have ever conceived for the engineering of language. It is therefore the most powerful engine humans have ever conceived for the engineering of coordinated action. If we want to have any hope of protecting our body, of using this cell to build houses and not slaughter our neighbors, we must understand it. We must understand what words, and thus what actions, it promotes and suppresses. We must understand what species of these new cells exist, and how each is sickening or healing our planet.

Who do they make righteous? What actions do they call ethical? How does this differ from the social contract we are engaged in right now? Does it differ in a way that we call good? Does it differ in a way that someone calls bad? And perhaps most importantly, who gets to participate in the revolution?