Successful computer Go programs typically rely on a mixture of tree search, knowledge-based systems, and machine learning. Implementing these, especially the pattern matching components, is no small task.
This AI has a minimalist design. It is an experiment on the failures of a previous project, which used Monte-Carlo methods. Even though its skill wasn't as bad as it had been expected, it was suspected, and noticed, that trivial AIs which worked thousands of times faster --because they performed no look-ahead-- were as good, if not better than our statistical approach.
So this AI does just that: no look-ahead. It has simple knowledge of stone formations, and acts upon this: improves the liberties of current groups, and attempts to expand these by joining nearby stones. This is interesting because such knowledge is typically non-trivial to construct if redone from scratch at each move, but building it along during the game is very simple.
This is a game that was played between this AI (black) and the montecarlo one (white). The result as a total victory for black.
Other games have been well balanced, or even total white victory.