[chgeswcr] Learning chess tactics

Chess has a well documented set of effective tactics discovered over its history: pin, fork, skewer, and many others.  Apply machine learning to chess in a way that it can be read off from the learned model the tactics that the computer has learned, and we can verify that the ones above, well-known to humans, are among the tactics it has learned.  This is likely a holy grail of machine learning.  What other tactics did it learn?  Did it discover any novel tactics unknown to humans?

Then, apply the same machine learning to many chess variants, e.g., different piece types, seeking those which have interesting tactics not seen in orthodox chess.  The goal is to find good chess variants.