In the games against Lee Sedol, AlphaGo played some odd moves when it appeared to be in the lead. A human would not have played those moves because we typically want to win by a lot. But AlphaGo only cared about winning, even if it was by the smallest margin, and chose different moves that gave the highest probability of winning in the end.
This reminded me of Gerrymandering in the US electoral system because you don’t want to “overdo” a win.