I grew up in a sharp landscape of expert systems - AKA models with few variables - that for my whole life has been melting into smooth statistical models.
Today's world is much prettier; much more accurate; but oh how annoying diffuse causality is to our linear minds!
Now, when I hear people talk about changes in nation-wide measurements like interest rates, inflation, unemployment as if they have "a cause" or even "a main cause", I judge the truth-value of those words as well below ChatGPT.
The "real answer" today - based on our best (most predictive) tools - is more like: "Well, I can list the 10 inputs that capture 24% of the change in inflation; or the 100 inputs that capture 53%; or my entire model that captures 67%; 🤷🏻♂️?🤷🏻♂️? about the rest."
And while I know it's way more true, my inner 3-year-old is very dissatisfied at how the world is now so often unable to answer my simple, endless question: "Why?"
True answers are now long and awkward to express in my native language of words.
In the new world of models, which are opaque to formal verification, our trust for the people and processes who train the models becomes much more important.
I need zero trust for any mathematician showing me any proof done with math I know, on one piece of paper. But as our models of reality grow from verifiable formal logic to large statistical models, that changes.
Whereas to trust a model, I need to trust the person, the process, and the data set used to train it. (Or be able to test it rigorously).
So, in conclusion, this melting into large models with diffuse causality is more true, and makes everything more annoying...which is much like many parts of growing up!