You should check out the accomplishments of Jack and WBridge5.
Here is a wiki, programs defeat world-class opponents 10 years ago.
This paper, although limited in scope, combines deep learning with bridge and have good performance.
Deepmind is a research organization.
Deepmind's Alphago does not learn by "observing go games", but by learning games records and self-play to become stronger. It is nothing like getting videos of professionals playing games.
However, it does learn atari games (breakout is one of them) by observing ...