It's a network of artificial neurons arranged in layers. Each "neuron" has many connections to other neurons. A neuron "fires" if the sum of signals received hits a certain strength. It thereby sends many signals to all the connected neurons, usually in other layers. You set up an input layer of neurons and an output layer of neurons and zero to many layers in between. The human brain uses about 5-7 such layers. So the basic paradigm is that you train it to recognise patterns by rewarding the good results and/or punishing the bad. You do this by altering the strength (the signal level) of the connections. In a real brain the connections are called dendrites, and in a real brain there are different frequencies of signals. A real brain alters the strength by using "drugs"!. Also you need to customise areas of the "net" to perform well defined functions. You can't just set up a billion neurons and hope it becomes intelligent. A neural net doesn't know how it works! So it's susceptible to all sorts of tricks, just like we are. For example, my hypothesis on the enjoyment of music is that our brain recognises thousand of patterns so it believes it's learning something very useful and provides "happy" drugs as a reward. In fact it's not, but it can't tell! Good programmers are very good at recognising patterns too. That's how we decide which functions and algorithms to write. And that's why a high percentage of programmers also enjoy music.

The good thing is that unlike a real brain a VFP model can insert new neurons into overloaded areas thereby improving the AI's IQ dynamically. Of course keeping control of all this is quite a challenge! ROFL.

Dave


>>>I've got a rough schema for pattern recognition vai a neural net type arrangement that can be stored in .dbfs.
>
>What's a NEURAL NET TYPE arrangement?? Sounds like something from Star Trek! :) Or if you tell me you'll have to kill me??
>
>>>You really have to purpose build the "net" with much of the heuristic (rules of thumb) level of intelligence pre-built. That's hard :-)
>
>Yeah, in my chess/othello game, I started writing AI that basically assigned different values to different strategic positions on the board. The computer would look at where it's pieces were, what the value of those positions were and calculate which moves would result in higher numbers. It actually worked ok for offense but the real problem was defense. I wanted to have it "learn" different offensive tactics from the human player so that it would know how to defend itself. But it got a little over my head and I haven't really picked it up again since. But if I recode it in VFP, I could use tables to store information that it learns and could scan that info very quickly if needed.
>
>Later.