The obvious thing is knowledge uses math. Knowledge uses mathematical objects to model the world. You can keep seeing patterns from one knowledge domain to another, but it doesn't mean they are connected by the model intrinsically, that is, that the model is the thing generating the knowledge. The model is just a model. Math is filled to the brim in ways of making models of things. The model can be representative of a full data set and closed under the data set, or the model can be representative of a prediction. In an open system, there is no mathematical closure. So in an open system that is producing knowledge, something has to be the ground. The mathematical models.
I don't know how this relates to bias, or digging oneself into a bias, but I do know that after enough cycles of going in a feedback loop, it can converge.
Worldviews are always difficult. You never have all the information. There's always more information. If there's not more information, people will invent new ways of making more information.
This is always the dilemma with data versus theory.