It just seems to play a very safe blocking game, which is fine but not always the best plan against good players. At least it doesn't make the classic noob mistake of playing ONLY the safest move which is bad, you end up with very few points to block with then, and lots of checkers on only a few points (known as "candlesticks", for obvious reasons). You want make connected blocks of points (a prime) which it does quite well (so did I but I did not get chances to hit back when I had good blockers, I was usually stuck behind decent blockers from CPU).
I should have started counting pips earlier than I did - it's not worth running if you are a fair way behind in the pipcount, it's better to stay back, work on your blockers, wait for a hit. I think I ran too early before I started taking the CPU opponent seriously enough
![Smile :)](./images/smilies/icon_e_smile.gif)
EDIT: My timing was good in the last game though I thought I only ran when I had a decent chance (after it left a blot in its inner board), unfortunately I wasn't able to escape or hit the move after - although it was lucky I didn't get hit and CPU could make itself safe there). Then it ruined all my good planning with that 66 and 33 when it was bearing off, lol
![Smile :)](./images/smilies/icon_e_smile.gif)
If you hang back you may need to get yourself hit to provide better tempo (i.e. don't crunch your board and waste all those blockers you spent time building). It's better to be hit and provide chances to hit later on when CPU is bearing off or about to than to crunch your board.
EDIT: Since it's a 16K game I expect the AI to be not too sophisticated. It might do a 2-ply evaluation of each position (i.e. each move it works out what will happen on all possibilities of next 2 rolls perhaps)? But that would probably use a lot of memory so maybe not? Maybe it just gives bonus points for making connected blockers/making more blockers etc. I expect it works out chance to be hit if it leaves blots though and plays the safest move which does leave blots. I don't know if it is then able to work out chance to become safe next move if it isn't hit etc.
Backgammon is quite hard to do AI for you can't do minimax, you have to do Monte Carlo (roll out a lot of positions and work out what happens on average) but that uses a lot of CPU time.