I'm guessing it generates the tree (2 levels deep) and then uses an evaluation function to choose the best branch.

Not too far off from a modern algorithm, except that it lacks even basic optimizations like alpha-beta pruning, etc. and an incredibly fine-tuned evaluation function and a huge library of opening and closing moves.

Here is one summary I found:

http://uscfsales.wordpress.com/2011/07/01/an-“easy”-engine-f...