Game play is controlled and regulated by Public Sub PlayCoordinator() located in the MainSubs.bas file. This procedure calls up the AI Engine with the function call nProgramMove = AIEngine located inside the PlayCoordinator SUB procedure.
Public Function AIEngine() As Integer is located in AIEngineMod.bas. It returns an integer corresponding to the cell position within the grid that AI wants to play.
The heart of the AIEngine rests upon nPlay = GetFactBasedMove. This function located inside the AIEngineMod.bas module is responsible for evaluating current information regarding game play, location of the program and teacher pieces and history pertaining to what constitutes a win. If the function returns a zero to the AIEngine that means it could not decide a good move based upon facts alone. Therefore, AIEngine then executes nPlay = GetRandomMove which is also located inside the AIEngineMod.bas module.
This can be done by selecting the AI Knowledge Tab. Each play by the AI program and by the teacher is recorded to this location.
Figure 1
The diagram above shows how the cells are numbered within the grid. The data to the right of the grid represents winning patterns that can be played by the program (left column) and by the teacher (right column). NOTE: This information represents the programs level of knowledge. It usually takes several games for this knowledge base to increase.
Each column of binary data has the digits 9 through 1. This indicates that all bits '1's and '0's below the respective cell number are applicable to that cell number. Observe the program information column on the left side. Look at '9'. Count the number of '1's. There are three of them. This means there are three winning patterns that can be constructed with a piece (X or O) at cell location 9. This makes sense since a horizontal, diagonal and vertical row can be played from that cell.
Each binary pattern consists of 9 bits. To the right of the binary patterns is an equal sign '=' followed by two numbers. Observe the first binary pattern for the teacher. It reads ' = 3 (2)'. The three gives the horizontal sum (H Sum) of the bits within the 9 bit binary number. Count the 1's to the left of the equal sign to see that this is true.
The number to the immediate right of the H Sum is located inside parentheses. This number indicates how many times this binary pattern has result in a win during game play for either the teacher or program.
At the bottom of the two columns of data, are numbers called vertical sums (V Sum). They represent the sum of the bits immediately above them. For example, the cell '9' position for the program data (left column) is equal to '3'. This can be verified by adding the bits directly above the '3'.
One additional point. The bits inside each binary pattern that is set to '1' represents an empty cell on the grid. All moves made thus far by the program or teacher are no longer necessary.
Three pieces of information are critcal: H Sums, Wins and V Sums. The relevance are explained as follows.
H Sums (Horizontal Sums) represent the number of cells that must be played to win a game. As the program makes its moves, on a particular winning pattern, the value of H Sum decreases. When H Sum is equal to 1 that means that the program, for example, can win the very next move by playing that cell indicated by the last remaining bit in the binary number.
Figure 2
In grid A above, the top row is empty and therefore available to 'X'. This is represented by the binary pattern 000 000 111. Be careful reading this. The least significant bit (LSB) is the first cell which is in the top left corner. The binary number in grid A is equal to 3. This means that three plays are required for this pattern to be won. Assume the program plays an 'X' in cell 3 as shown in grid B.
The binary pattern changes giving a H Sum = 2. 'X' is played again as in grid C. Now the binary pattern shows only one '1' bit. This pattern is equal to a '1', meaning that in one more play this program will win.
The function GetFactBasedMove does in fact check for a H Sum =1 for the program. If it exists it plays it. On the other hand, if 'X' cannot win in 1 play, then it checks the teacher's binary pattern. If that H Sum=1 then the teacher's play must be blocked. AIP does this.
Wins provide a measure of how often this pattern has resulted in a win in the past. Let's assume that the lowest H Sum is equal to 2 and that thee patterns have this value. That means there are three possible ways for 'X' to win in two moves. Which should it play? AIP looks at history and picks the binary pattern with the most number of wins. If it doesn't find any history it picks the first binary pattern in the list.
V Sums help to select the specific bit which corresponds to an actual grid cell. Once a binary pattern is chosen, it will consist of 2, 3 or more bits that are set to '1' (for 5x5 grid games). The V Sum tallys all of the bits and tells AIP which cells result in the most wins.
Observe the program binary data in Figure 1. Note that a binary pattern with H Sum=3 has 7 wins. Therefore AIP has chosen this pattern. AIP must now decide which particular bit to play. It decides this by find a V Sum with the highest value that also corresponds to a bit of '1' in the binary pattern. Notice the two flatten red circles. They intersect on the 5th cell on the binary pattern with 7 wins. This is the center of the grid. As all Tic Tac Toe players know, this center cell or location can result in four possible winning patterns as denoted by the V Sum=4.
Last act of desperation. When none of the above data is not meaningful such as at the beginning of a training session AIP yields to the function named GetRandomMove which produces a random legal move.