The main context is that current ML models although seeem to be doing very well across many tasks however, do not have any mechanisms to explain why they are taking those decisision. Are they just complex mathematical systems that somehow learn to capture patterns from data? This also leads to a very natural question: why do these systems fail spectaculary in unfamiliar circumsatances or situations it has not encountered during training ?. Clever way of fooling Neural Networks ( and other types of model ) using adversarial examples cast more doubts wheter these models have truly generalized / learned the hidden meaning in datas rather than superficial correlations.

However, in the same context we humans are very good at adapting to unseen scenarios. We are good at making decisions based on our intiutions which is probably shaped by our own experinces, imagination or even learned from others. In a way, we can somehow explain why we do certain things or at least make an informed guess.

Learning links: