When programming artificial intelligent systems to make decisions and or evaluate there are a number of theories that should be explored before choosing a specific path. Writing code is tedious and time consuming indeed and it is appropriate to do as much up front work as possible. First and foremost you need to decide on which model your AI system will take in its decision making process. If you intend to build a decision matrix system then you may wish to do a little philosophical discussion with yourself or your team. For instance think of ways in which mankind normally chooses things?

For instance in a beauty contest each woman would be given a numerical value on each event. Then after three to five events the woman with the highest score wins? In the Winter Olympics in 2006 each judge might have their own method of figuring things out and perhaps give consideration either objectively or subjectively to things such as technical work, creativity, uniqueness, performance and difficulty and come up with a score and thus use that to judge. All we see is a little card which says 9.8, 9.9 or a perfect 10 you see?

Now then on these two evaluation processes you can see how you might design your program to do these things. In the event of Winter Olympics Ice Skating with many judges you might mimic this in your artificial intelligence program by simply running several slightly different programs of numerical evaluations (each one similar to a judge) and then average the scores for a total score. This might work good for choosing one product, person or concept over another. On the beauty contest you might have many judges with similar ways of grading and slightly different subjectivity over various things within each event. This too can be duplication with programming for an artificial system quite easily.

Another humanly way we do things is to have contests where we pit two different teams against each other and the winner advances and the loser does not. Kind of like basketball playoffs or soccer champion ships. Or the NFL to determine at the end of the year who is the best. This form of evaluation can easily be programmed and such code is readily available. Using techniques of double or triple elimination you can better evaluate and remove much of the randomness of chance or luck and improve probability for coming up with not necessarily the absolute answer, but statistically speaking the best probable answer down to a point in which you can feel good that your AI computer system is fair, honest and in the end has chosen the best possible answer.

Now then when evaluating evaluation decision matrix artificial intelligent systems one must also realize that you are only limited by your creativity to pick the best system to evaluate. You might even write separate systems which uses all three of these systems or more and then average them together, thus getting closer to the unattainable best possible perfect choice. Why is this important? Well, if you expect humans to follow the orders of artificial intelligence systems then you should make sure than their answers that they come up with are truly 99.999999999% the best possible answer. Think on this in 2006.