2 taliesin:
I'm sorry too if my words had rude or offensive shades. I see your efforts to make all sentences clearer, thank you. But i know that i can't do this as i want. Although, i'll try too...
It's great that you have theoretical knowledge on this subject. Although i don't like measures or any kind of compares, i'll write my background here too. So i have a standard russian education on applied mathematics and computer sciences. Since our education is wide-orienting, not problem-orienting, we haven't such separate topics as 'Artificial Intelligence' (it was a part of wider course).
I don't remember exactly, how long i know on practice what is programming. Probably i think that this time is about 21-22 years, maybe it's not so long. Maybe i'm an old-style algorithmic programmer; for now it's not valuable for commercial purposes -- most of our companies need for 'button-pushers'.
I know that theoretical knowledge is quite valuable for developing purposes, but it needs a practice too. Returning to task: building a decision tree is not a problem if we knows when we have to cut its branches; position evaluation is not a problem too if we have correct function for it. But my experience tell me that it's not a solution. If we have a problems with realization on 'calculative' method even with 1-2 hp, we'll have much more problems with 10-15 hp.
To extend our minds till solution, to break old dogmas, to make live perception -- it's my way. I don't call others to this way, it's hard there to go.
Prolog is very interesting language, but i don't use it in real tasks 'cause it's not a problem for me to build searching or 'go-through' alrorithms in classic languages. I think that this language weak from point of time-evaluation of method-working and space-evaluation of it. From the other side, it's a great platform for quick logic solutions (we need quick realization? i'm not sure about it)
Neural nets is an element of data-mining methods which are widely use in my current work. My opinion on neural nets: strong classification and unification characteristics; weak in multi-factor analysing; hard to teach and use.
I'm not an expert in AI too. I never finished game with fully-done AI (except seabattle AI many years ago), but i make some logical programs (and games too) where was the elements of AI-players.
Undoubtable. I never reach this level 'cause i have number of limitations for doing it. One of them is unskillful in strategic perception, another is inability of deep-analysis. For example, Alaric, who widely use deep combinations, making the game with me quite easily.
It's very interesting for me too. But i have no time to make even a part of it.
Can you explain me a basis of 'state-space search' term? I'll try to compare it with all i know in this subject.
Yes, they are. We can use them for making a decision when there left 5-10 input parameters and 1 output parameter; in the other words, for making a final decision.
Using neural nets for 'Warlocks' was not my idea; this idea was spoken by nawglan and i said that his method (let input of net will be code of current position and output of net will be move code) is impossible to making.
Maybe. But i don't know how to convert my knowledge to words. I suppose that your theoretical knowledge will be quite useful in this situation.
I think that unheard-of bridge building -- to teach workers how to construct a bridge -- via phone, for example. You're trying to do something like this. You need 3-5 centuries for it, i think.
If you want to get a real result, you need to organize developer group with nawglan or vilhazarog, or both. Your theory and their practice may produce a sort of solution...
Oh, i see. Exactly for this purpose i plan to gather statistics of opponent moves. And, for example, neural nets may be quite usable for player-behaviour prediction... Yes, it needs to run 'calculative' method before using the neural net.
So which situation will we have when arm AI with 'minimax' method? AI can say which move is most valuable so it can predict most of moves. Situation of that ogre extends to more common variant with making a sacrifice: you going bad for a short time, then you make the position much stronger than you have now.
Oh, yes, sure. Position evaluation and move evaluation must be done prior any statistics gathering...
It's a very partly situation. I think there are much more positions where predicting is important.
For example: 'DP/PS' vs 'DS/PD' on Maladroit. Theory reads that it's useful to target DPP on player 2 so you don't disrupt yourself with amnesia. Ok, i use it. I'll target DSF on myself and being sure that my opponent plays DPP. For computer this information may be get from statistics.
Additional, you can know most likely openings for each player. It's important too for move prediction.
Ok, with extension you prefer after DSF opening? DSFW? DSFFFc? DSFPS? Other? In any case, you have your own behaviour and it's important to predict variants when you change your tactics suddenly.
Surely, and many other variants too. But you can't change your strategy infinitely. Each time when you use new strategy, you'll be in risk soon. I don't said that all moves will be get from statistics; it's important to take in consideration classic analysis of possible situations too.
Oh, yes. Stable function of situation perception is a great problem too. To make a difference or find a common in two positions...
It's a serious barrier for making a solution, i know.
...easy?
It's not easy, and you know it...