| |
| |
| |
|
Page: 1 2
Comments:
<0> Ok, heres another philosphical proposition:
<0> Is it fair to state there are two types of programs in AI, 1. Those that interact with the environment and 2. Those that observe the environment without the means to interact.
<0> Although the first would require the second, the second may not require the first.
<1> by "interact with" the environment... do you mean, "attempt to change the state of" the environment?
<0> athena: Exactly
<1> well, we should probably talk about what you mean by "AI"... but, yes, you could make such a cl***ification
<0> athena: For example: We have four sets of bit arrays, our desired states, our motor states "hands feet", and our perceived states "eyes, fingers"
<0> Wait make that three sets...
<0> So we observe with our eye bit arrays and compare them to our desires, if they match we store the sequence leading to that state.
<0> But if they don't we just observe what our new state will be by effecting our "hands and feet" bit arrays of motor output.
<0> We can learn to make a series of "hands and feet" states to achieve our desired states.
<0> I'm probably not making any sense, this is all on the fly and completely unrevised.
<1> ai_coder: yeah... have a bit of trouble seeing where you're going with this :)
<0> athena: Well you've probably noticed I'm not going anywhere, I'm just trying to mock human behavior, after all The standard definition of AI is that which we don't understand.
<0> It seems as though humans are attracted to observable things that change with some frequency that are pertinent to their perceived success. This seems all unrelated, however.
<1> humans have goals
<1> they seek to fulfill them
<1> i thought you don't believe that such a thing exists or is coming
<0> athena: I don't know if I do or not.
<0> It's pretty tough to predict the future.
<2> that's how he plans to be immortal, by waiting forever for it :)
<0> Interesting to think about 20 years from now if we will have something replace us. Besides chinese troops or korean missles.
<1> chinese troops, korean missiles... don't focus too much on the "threat du jour"
<2> if not singularity, then maybe thing about phase change, what will the next stable configuration or organizational principles be ?
<0> Yeah, I think of it like a card game. Everyone wants to win and they'll usually try to do brinksmanship, but they're really just after respect and resources, they don't actually want to destroy the world.
<2> destroying the world is bad for business
<2> everyone worrying about it on the other hand ....
<3> problem with human war is it aint over till its over.
<0> type_T_: I don't really understand how killing someone proves that you are more right than someone else.
<0> Yet that is the way it's played.
<1> i think that stuff is fairly well understood in evolutionary psychology
<4> Hi I've read a bunch of stuff on Minimax, and am pretty sure I understand the basically. However I'm not sure how I am supposed to define a game state, build a node of states, or create an evaluation function.
<1> some context pls? :)
<4> Basically its a game where you have 6 gold nuggets worth 1 to 6. So does your opponent. Each round the dealer puts out a nugget from 1 to 6 and the players must do a blind bet using one of their nuggets.
<1> not the specific rules, but what are you trying to accomplish?
<4> Minimax the game. For any round, I want an AI to pick the best bet.
<1> ok, what don't you understand about defining a game state, bulding ply trees, etc.?
<1> are you asking about data structures?
<4> Yes
<1> graphs, trees
<1> root node is start of the game
<1> each possible move results in a new state, which are the children of the root node
<4> So before the game
<4> I should generate the entire tree?
<1> evaluation function basically computes the value of any given state
<4> Which will be used for searching later?
<1> different ways of doing it... for a game like tic-tac-toe, with small number of possible state you could easily do that
<1> for chess, you'll run out of time/memory
<4> How about 9000 possible states?
<1> depends on your resources, but that seems small enough to pre-compute and cache
<1> and then you can confidently make the best possible move (and know already what the best possible outcome of the game will be)
<4> How do I compute a value for a move?
<4> Subtract the two player's scores?
<1> you'd evaluate states
<4> What does evaluate mean?
<4> here anyway.
<1> take the example of chess
<5> if you have ready to use scores for the two then yes the difference between the two is a good evaluation value
<1> a state is (a) position and (b) who's turn it is
<4> both players move simultaenously
<1> value of position is likelihood of you reaching your goal
<1> value of a move is the utility of it
<4> So my plan of action is. Define a game state structure. Generate the game tree. Run minimax over the entire thing as to cache it.
<4> Sound good?
<1> sure
<4> Thanks.
<5> minimax is usually calculated as you move through the tree. the tree is not stored afaik
<1> although it won't be a tree, since some states can be reached through different paths
<5> still be a tree. the tree doesn't recognize same state
<5> but minor detail
<4> For any state, I'll just have a list of possible states that can be moved from that one state right?
<1> danf: it would be better if it did
<5> you generate it as you go
<4> DanF_DrC: Why?
<5> athena, sure but can be expensive
<4> Wouldn't precaching it just be better?
<4> Since you'll be regenerating the same conclusion each time you play with an AI?
<5> ldlework, to save memory. usually there is nothing gained by storing it
<1> his state space is tiny
<4> There is if I want the AI to be as smart as possible.
<1> it doesn't make the AI less smart... it's a cpu(time)/memory tradeoff
<5> ok if the game is simple you can precalc everything before it starts. you can then render a perfect strategy such that you'll never need to calculate anything more
<5> each move is an instant lookup
<5> but the game would have to be simple
<4> 9000 moves?
<4> states
<1> danf_drc: this has already been covered... see above
<5> k
<1> note that 9000 moves doesn't necessarily imply 9000 states
<5> ah that game
<4> Its 9000 states
<4> I mistyped
<4> Heh
<4> I don't even need minimax do I?
<4> Since each state's value is simply the difference in player scores.
<4> I just look up the best next possible state
<5> still be minimax
<6> hey ya'all. what's the topic?
<4> Minimax
<1> chessguy: atm, optimal heuristics for state space search
<6> for any particular application?
<4> simple betting game
<1> a game of Idlework's design
<5> athena, r u comp sci?
<1> i have a paper that says something about that, yes :)
<5> : ) and female?
<5> I understand
<5> relatively rare
<1> rara avis, i know
<5> never heard that expression before
<5> r u smart smart?
<1> sorry, you quoted latin earlier
<5> don't apologize :) I now know what it means
<5> how long have you been coming here?
<1> few days? i think
<5> ok. was just wondering my attention had been amiss : )
<5> if
<1> looking for some serious AI discussion... what other fora do you know of?
<5> none. but then again you have found me : )
<5> are you still in the university environment?
<5> or just as an interest
<5> keep in mind Einstein was just interested
<1> left uni a few years ago... current interest in AI is my own
<5> ok. what's your take on the field? if you have yet formed an opinion. where to look
<1> mmm... still a discouraging amount of thought wasted on "GOFAI"
<5> I think it's worse than that but sure :)
<1> have the feeling that to build general AI, we'll need to incorporate findings from serveral disciplines... not just comp sci
<5> kinky :) are you thinking some sort of reverse engineering of the brain.. or nothing specific as of yet
<4> lol okay I have my game state defined. Now how do I generate the tree?
<5> systematically
<5> which options. visit each one in turn
<5> draw a tree to help see it
<1> danf_drc: yes, basically... it's an interdisciplinary effort
<5> ok. personally I don't think that is needed
<1> ldlework: an edge is a move, leading to a vertex (state)
<1> danf_drc: what do you suppose is sufficient?
<5> machine learning. of the unsupervised learning persuasion
<5> learning on it's own so to speak
<5> do you yet appreciate the importance of learning?
<5> I equate AI with learning
<5> learning is fundamental is my mantra
<5> although we are not machines I think the analogy holds. think of a baby. is it intelligent or is it its potential. its learning potential
<5> make sense?
<1> yes, makes sense and i agree more or less
<5> in what sense do you disagree?
<1> in the sense that i'm too tired to think carefully about what you said and therefore i qualify my statement :)
<5> : ) quite tired too. I should sleep
<5> talk to you later
<1> k, bye
<7> athena are you a greek goddess?
Return to
#ai
or
Go to some related
logs:
#lisp
#perl
gentoo stop unicode
ubuntu hdparam
Unknown format +ffmpeg +jpg -printf
get rid of dynamic outlines in Internet Explorer
php func_*
#kde
#perl
#gentoo
|
|