The documentary film “AlphaGo” has been released and can be seen on Google Play, iTunes and Amazon. This film takes us into the world of Google’s DeepMind Developers as they are ramping up the AI program AlphaGo to compete against the worlds best Go player Lee Sedol in a five-game match.
The company DeepMind originally started building an agent for the 1975 game Breakout written by Steve Wozniak and Steve Jobs of Apple. They made a simple AI that could move the paddle and was instructed to figure out how obtain the highest score possible. Within 600 episodes of training the agent finds a tunnelling strategy by hitting the ball behind the wall that was not previous realized by human players. The AI is teaching us new strategies.
While the film is deeply entertaining, technical and exciting there are many things we can learn from this round of human vs AI challenge.
- The two Go Players featured in the film Fan Hui and Lee Sedol went on winning sprees after playing AlphaGo. Fan Hui won the European Professional Go Championship and Lee Sedol did not lose a Go match for the next two months.
- Even within the DeepMind AlphaGo team there were several people sort of rooting for the human player.
- The AI tactic of winning by being taking over 51% of the board area dramatically changed the game. Normally humans tend to go for a larger win.
- Who is the expert? During several points of the five game series it was unclear if AlphaGo had come up with an amazing strategy or was just embarrassing itself with bizarre moves. Even the developers were not good enough Go players to interpret the choices AlphaGo was making.
- It was clear that one important roll for the AI was that humans setup all the right conditions for the machine. Plug in the cables, load the software and monitor the agent to make sure everything is working perfectly. There are several scenes where humans are “serving” the AI. We’ll make great pets.
- Through out the film professional Go commentators are referring to the AlphaGo AI as “he” or “she” thus anthropomorphizing the agent.
This film will be enjoyable to any geek. The use of open source tools is clearly depicted (Linux, XEmacs and shell consoles) help to create a link to the development teams reality. The documentary is well shot exposing the complex moves made during gameplay with constant switching to different sources of expert interpretation to help the viewer understand “what just happened”. There is no need to be a Go player or machine learning expert this documentary will create a lot of questions and discussion about the future.