Why Zero

Zero means no human knowledge added. No one has ever taught Leela Chess Zero even it's good to have more pieces instead of less.

There are two schools of thought on that.

One for its possibility and style. Two for its maximum strength.

Anything it learns from humans could be wrong, since even SF makes mistakes. It wouldn't work to try various strategies against SF or a patzer, because trying two strategies would either both win or both lose, and you wouldn't know which was better.

Only self play lets it gradually find a path to perfect chess, assuming there is a gradual path. But Google's DeepMind shows there is a path to at least very very good chess.