An unbeatable game of Tic Tac Toe. The AI uses a tweaked Alpha-Beta Pruning algorithm for the decision making.

The AI cannot be defeated. Every game either ends in a draw or the AI winning.

Vanilla Alpha-Beta Pruning sometimes leads to unexpected results such as the following:

You can see in the above gif that the AI could have won by playing an O in the bottom left corner but it instead chose the middle cell in the first column. This may seem a little strange but no matter what move I choose afterwards, it still wins.

What I found was that even though Alpha-Beta Pruning based AI plays perfect games, occassionally it chooses to make a move where the outcome will be a slower victory or a quicker loss. By tweaking the algorithm a little bit and including search depth in its board evaluation I was able to get it to always choose the quickest victory or the slowest loss.

The following is what the tweaked Alpha-Beta Pruning algorithm does:

As you can tell, this time it chooses the path of quickest victory.

When using a regular MiniMax algorithm the AI checks 59,704 possible board combinations for its first move (this number obviously decreases throughout the game as less cells are available). The Alpha-Beta Pruning on the other hand brings this number down to 2,337. Adding the tweak that I mentioned above brings the number up slightly to 2,787. This increase is miniscule in comparison to the advantages that the algorithm brings to the table.

After downloading, navigate into the folder that contains the packages ArtificalIntelligence, Assets, TicTacToe, and Results.txt. Then type the following commands to run the game in Window mode:

javac Window.java
java TicTacToe.Window

Typing in any parameter will run the game in Player vs. Player mode. Example:

java TicTacToe.Window -pvp

To run the game in the console, without a GUI, type:

java TicTacToe.Console

The console version takes in the player input by index. This means that to select position (1, 1), the index would be 4 since it is the 5th square but we are using zero based indexing.

Console mode does not support Player vs. Player.