2/ I believe this saga started when
@MITresearchers found Go problems that are easy for humans, hard for AI. In this example, black has a guaranteed win if it simply chains together black columns. Yet KataGo, playing against itself, looses as black 60% of time.