In AI, pragmatism treats workable solutions and measurable gains as more important than settling metaphysical debates. It allows researchers to make progress by bracketing questions about what intelligence ultimately is. section 9.3.7
Prediction and understanding
When a model predicts data correctly, calling that “understanding” can be a pragmatic label for its utility and its encoding of data regularities—not a claim about the nature of understanding. section 9.3.7
