How to Use Causal Inference to Distill the Complex Pathways of Giant Deep Neural Networks

I was curious about this idea on whether you can use causal inference to distill the complex pathways of these giant deep neural networks. Could we kind of do causal structure discovery within these activation graphs of these big deep neural networks? Yeah, I mean I don't know is the answer right but like that certainly it would be a cool thing for people who say they need to understand the neural network better.