In Axiomatizing the Ambiguity of Goodness, I presented a short argument for the ambiguity of goodness. The post contained several examples of traits that seemed unambiguously good on the surface, but became ambiguously good in certain contexts. One particular (trait, context) pair I mentioned concerns the use of violence.
I first offer pacifism as a trait that, on the surface, seems unambiguously good. I then present the situation of using violence to defend your child from an assault. In this situation, the goodness of pacifism becomes ambiguous. Why does it feel justified when we use violence to defend our child from assault?
In its most abstract form, the moral question we hope to answer is: “When is the use of violence Truly good?” Unfortunately, as we saw in A Short Proof of the Ambiguity of Goodness, an unambiguous definition of good, which we might refer to as True good, does not exist if you assume the axioms of optimism and conflict. Thus, we cannot hope to answer the question: “When is the use of violence Truly good?” Instead, we must settle for answering the question “When is the use of violence truly good?”
The distinction between “capitol T Truth”, which gestures at a noumenal, platonic, ineffable, unambiguous fact, and “lowercase t truth”, which concerns the level of justification we assign to an idea, is both subtle and critical. In a short video that merits its own (eventual) post, renowned neopragmatist Richard Rorty eloquently outlines the distinction between Truth and truth, as well as how truth relates to justification.
We know how we justify beliefs. We know that the adjective true is the word we apply to the beliefs that we’ve justified. We know that a belief can be True without being justified. That’s about all we know about Truth. Justification is relative to an audience and a range of truth candidates, Truth isn’t relative to anything.”
Equipped with the distinction between Truth and truth, as well as an understanding of the relationship between truth and justification, we can formulate a very precise version of the question we are hoping to answer: “When is the use of violence justified?”
Max Weber, the German political economist credited with founding the field of Sociology alongside Karl Marx and Emile Durkheim, offers an exact answer to this question: violence is justified whenever the state says so. In fact, Weber sees the connection between the justification of violence and the state as so profound, that he uses the power to justify violence as the defining characteristic of the state.
“A state is a human community that successfully claims the monopoly of the legitimate use of physical force within a given territory.”
Recall that Rorty describes justification as “relative to an audience and a range of truth candidates." Using Weber’s definition of the state, we can identify this audience as precisely the polity controlling the territory where the violence occurs. Critically, this relationship enables us to transform questions about the justification of violence into questions about the opinions of the people who make up the controlling polity.
This equivalence between justification and the consensus of the controlling polity hints at several deep connections between morality and ideas such as the state of exception, the overton window, the engineering of consent, and schizoanalysis, which I intend to explore in subsequent posts. For now, I want to conclude by remarking on the implications of this equivalence between justification and consensus.
Let’s return to our original question: “Why does it feel justified when we use violence to defend our child from assault?” The abstract answer is that the controlling polity deems it as justified. But who exactly is this controlling polity? As I am writing this in New York City, candidates for the controlling polity include the state and federal government, which both broadly recognize defending your child as part of your right to self-defense. Another candidate is my local community, which not only recognizes the right of self-defense, but also upholds the values of childhood innocence and familial support. A third candidate is the polity consisting of only myself, who by nature or nurture or both feels that defending my child from assault is right. In this situation, all the candidates for the controlling polity come to the same conclusion - that the use of violence in defending my child is justified.
The relationship between justification and consensus has broader implications on the field of ethics; namely, it provides a template for a purely empiricist ethics. In combination with the advent large language models (which are the primary object of study at Nomic), we now have the ability to analyze massive corpora of political speech across broad swaths of space and time.
It is already known that LLMs can be biased towards particular sets of geocultural values. In particular, Anthropic recently found that their LLM outputs are biased to agree with western geocultural values. Tabling the implications of this fact on value lock in for a moment, consider what were to happen if we inverted the setup. Instead of investigating the values learned by a LLM on a particular corpus, what if we tested what values were present in different geospatial corpora by investigating the different values an LLM learns when trained on them.
Concretely, I believe we can use a language model conditioned on the place and time of its training text (which I will hereafter refer to as a space-time language model, or STLM) to perform comparisons of ethical opinions across cultures. One such study using a STLM would involve measuring the perplexity of ethical statements across space and time. For instance, I conjecture the statement "emancipation is a good idea" would have lower perplexity when conditioned on New York City in 1863 than it would when conditioned on Atlanta in 1863.
From an information cartography standpoint, the visualization to be made here is a heatmap overlayed on a world map with a time slider. For a given statement, the heat of any point in space/time would equate to the perplexity of that statement in the STLM at the given place/time.
Oftentimes when discussing these ideas, I find that others mistakenly believe that I myself must have an ethics that is purely relativist. This is certainly not the case. I am subject to the place and time of my culture, and thus, I have an ethics that is shaped by the particular norms and ideas I have experienced. The following clip from Moonrise Kingdom, one of my favorite movies of all time, does a great job at summing up the personal ethics I hold:
Suzy: Was he a good dog?
Sam: Who's to say? But he didn't deserve to die.
I was fortunate enough to run into Chloe Bakalar at a dinner party recently, where we discussed several of the ideas in this post at length. The biggest takeaway I had from the conversation is that conditioning on the place and time of a text may not capture the functional ethics of that place and time's controlling polity. This is due to the nontrivial gap between the public opinion of a constituency and official policy enacted by its government. I intend to explore this phenomenon, and its relationship to random graph models, in a future post.
Tangentially, my favorite fact from our conversation is that Schole, the Greek word for leisure, is the etymological root of the English word School.