[meroes]

Can I take this opportunity to ask a stupid question?

Is ML anything more than pattern recognition? If I could tally nearly every game state vs game state -> win %, then run a simple if [state], then {} program, then what do I miss out on vs an ML approach? Is the magic just in how we feed good data sets to an ML algorithm so it can efficiently mimic the above much quicker?

[rsrsrs86]

This is a very clever question.

If your problem's domain is not too big then you are basically right, but in practice things are more complicated, larger, noisier, etc. I am conflating many things here of course.