Most arguments aren’t really about what they appear to be about.
Two people disagree about a decision. They go back and forth, evidence, counter evidence, position, counterposition, and the argument doesn’t resolve. Because they are arguing about the same words while meaning completely different things. Because underneath the surface disagreement is a deeper disagreement that nobody has named.
They’re treating the same thing as different kinds of things.
One person is treating the decision as a technical problem, something to be optimized, measured, solved with the right data. The other is treating it as a value question, something to be weighed, deliberated, decided from commitment rather than calculation. Neither is wrong about their own framing. Both are wrong to assume the other is using the same framing.
The argument cannot resolve until someone asks the prior question. What kind of thing is this?
That question is ontology. And like epistemology, it sounds more abstract than it is. Ontology is simply the inquiry into the nature of things, what kind of thing is this, what category does it belong to, what does that mean for how it should be approached.
It is the thinking tool that prevents the most common and most consequential error in reasoning, treating something as the wrong kind of thing and then being confused when the tools appropriate to that kind of thing don’t work.
Category errors are everywhere once you learn to see them. And they are almost always invisible to the person making them, because the category you’re using feels like reality rather than a chosen framework.
Here are the category confusions that matter most for thinking clearly.
The first and most pervasive, treating assessments as observations.
An observation is a description of what is present. An assessment is a judgment about what is present. They sound identical in ordinary speech. “This code is inefficient.” “This strategy is wrong.” “This person is unreliable.” Each statement sounds like a report of something out there in the world. Each is actually a judgment made from inside a perspective, using standards that could be examined and questioned.
When you treat an assessment as an observation you make it unchallengeable. Observations can be verified. Assessments need to be grounded, in standards, in evidence, in the explicit perspective from which the judgment is being made. Flores was insistent on this distinction. The assessment that goes unacknowledged as an assessment becomes the unexamined background that determines every subsequent thought in that territory.
AI outputs are almost entirely assessments dressed as observations. The model produces statements that have the grammatical form of objective description and the epistemic status of pattern-completion. The ontological question, is this an observation or an assessment, and if an assessment, what standards is it using and where did those standards come from, is the question that reveals what the output actually is.
The second, treating feelings as conclusions.
A feeling is information about your current state. A conclusion is a claim about the world that can be examined, tested, revised. Feelings are real and important. They are not conclusions. The anxiety you feel about a decision is real information about your relationship to uncertainty. It is not evidence that the decision is wrong. The enthusiasm you feel about an idea is real information about your current orientation. It is not evidence that the idea is good.
This confusion produces more bad thinking than almost any other. Because feelings are immediate, vivid, and certain in a way that conclusions rarely are. They feel like knowing. They feel like evidence. They feel like the body telling you something true about the world. Sometimes they are. Often they are the background talking, the inherited pattern activating, the historical fear surfacing, the old assessment running in a new context.
The ontological question, is this a feeling or a conclusion, and if a feeling, what is it information about, is what separates the thinker who is using their emotional responses as data from the thinker who is being used by them.
The third, treating patterns as causes.
A pattern is a regularity in observed events. A cause is the mechanism by which one thing produces another. Patterns are enormously useful for prediction. They are not explanations. The sun has risen every morning of recorded human history, that pattern is reliable for prediction but it doesn’t explain why the sun rises. More consequentially, the pattern that X follows Y in the data does not mean Y causes X. Does not mean the pattern will hold in new contexts. Does not mean understanding the pattern constitutes understanding the phenomenon.
This is the deepest ontological confusion in the AI moment. Models are pattern systems. Their outputs are pattern completions. When the output feels explanatory it is because the pattern of explanation, the structure, the vocabulary, the logical form, is being reproduced convincingly. But the reproduction of the pattern of explanation is not an explanation. The ontological question, is this a genuine mechanism or a pattern dressed as mechanism, is what prevents the category error available in the current moment.
The fourth, treating the map as the territory.
The model of the thing is not the thing. The description of the experience is not the experience. The framework for understanding the domain is not the domain. This is perhaps the oldest ontological warning in philosophy, Korzybski’s the map is not the territory, Zen Buddhism’s the finger pointing at the moon is not the moon, Magritte’s this is not a pipe.
AI outputs are maps. Extraordinarily detailed, often very accurate maps. They are not the territory. The person who mistakes the AI’s description of a domain for direct access to the domain has made the map-territory confusion at the highest possible level. The output is so good that the map feels like the place.
The ontological question restores the distinction. This is a map. A good map, possibly. A map nonetheless. What would it take to verify that the map corresponds to the territory?
The ontological habit changes how you encounter everything.
Before you think about something, before you engage with it, argue about it, build on it, accept it or reject it, you ask what kind of thing it is. The category. What it is.
Is this an observation or an assessment? If assessment, what standards, from what perspective, with what grounds?
Is this a feeling or a conclusion? If feeling, what is it information about, and does that information constitute evidence for the claim being made?
Is this a pattern or a cause? If pattern, does the pattern hold outside the training data, the observed cases, the contexts where it was established?
Is this a map or the territory? If a map, how do I verify correspondence, where might the map be wrong, what would the territory reveal that the map doesn’t show?
These questions don’t slow thinking down. They prevent the kind of thinking that has to be completely redone when the category error surfaces, which it always does, eventually, usually at the worst possible moment.
Applied to AI outputs, the ontological habit is transformative. Because AI outputs are almost always category-ambiguous, they have the form of observation while being an assessment, the structure of explanation while being a pattern, the confidence of territory while being a map. The ontological question cuts through the form to the substance.
And applied inward, to your own thinking, it is the question that reveals where your most confident conclusions are actually assessments you’ve stopped examining, where your strongest intuitions are actually patterns you’ve stopped questioning, where your clearest picture of reality is actually a map you’ve stopped comparing to the territory.
Socrates kept asking, what is this thing, really. What is it?
That question sounds simple. Sitting with it honestly is one of the most demanding practices available to a thinking human being.
What kind of thing is this?
Ask it about everything. Especially the things that feel most obviously clear.
Those are usually the ones most worth asking.
@ Lee Greene