Truth is that which is preserved in transformation

I like to think of models as being physical structures that actually exist rather than pure abstractions. The model is embodied in the relationships among the parameters of the model, and the changes that the physical structure goes through are the model’s updates.

An object A is a model of object B if we can look at object A to make predictions about object B. For example, a footprint in the mud is a model of the foot that made it because you can look at the footprint to know things about the foot. The footprint actually exists as a physical structure—it’s a thing you can look at and poke and stuff—so we can study it empirically as an example of a model to better understand models themselves, with all the usual benefits associated with empiricism.

Thinking of models this way lets us see models in places we wouldn’t ordinarily expect to, such as bodies. It also helps us to recast first-person experiences as third-person objects of study, such as preferences and values. We can also think about how models, and the elements of models such as preferences, are constructed: put together by parts through a real physical process. And other properties that seem very abstract and maybe even inherently disembodied can be understood in relevantly concrete terms, like the notion of truth: truth is what is preserved when the model changes.

Truth is a really important concept in economics because the economy is all about getting people to share truthful messages with each other so that people can form plans based on accurate expectations about how those plans will fit with everyone else’s plans. If people don’t share truthful messages about their plans, then it’s hard to coordinate your plans with everyone else’s.

When a model interacts with evidence, it transforms. Evidence is defined analogously to interoception: an interoceptive signal is interoceptive because of its role in allostasis, not because of anything about the signal itself. Evidence is similarly any signal that interacts with a model in such a way that the model’s transformation is an update. The things the model does are its predictions.

It stands to reason that if the model doesn’t change its predictions, then those predictions weren’t falsified by the evidence. Therefore, the predictions are true. So anything that doesn’t change when the model changes is true, and anything that does change is false. So truth is that which is preserved in transformation.

One of the useful things about this view of truth is that it is more relaxed, more embodied, and less binary than how we typically think of truth. Normally, we think that somethings are just true, and some things are just false, even if we don’t know which things are true and which things are false. But in this view, nothing is abstractly true or false: instead, it’s all about real interactions that produce changes. Some things will survive some interactions, but not others. Other things will be preserved across all interactions we can currently throw at it, like the empirical regularities we call the laws of physics. This lets truth be relationally real rather than attributed to a proposition. And it fits very well with thinking about physical objects as models that try to predict their environments. Note that it’s not as simple as saying that good models stay pretty much the same while bad models change (or die): the stock market is a really good model, but it changes constantly: the constant change is because it’s constantly updating in response to new evidence, which is what a good model should do! Bad models ignore the evidence for a long time.