1.5 request

[Fosprey]

Dave, really i don't understand your point.
Yes, the characteristic of your team is relevant, because if you were 120 Genius, with all the time , joy, will and money in the world, with access to all the techonology and resources of humans at your disposal, this work would be possible in a decent framework of time.
i'm pretty sure this aren't the characteristics of your team.
So it is related to the characteristics of your team.
A build of a decent AI in a incomplete information game, one that can beat at least poor players, it's a project bigger than wesnoth, if you do, it will be 100 more relevant to most of the programming world than wesnoth. As i said, there is plenty of people working on it with no success at all. And it would be a brillant advance to achieve that. (to be honest i hope it never happens, it will become a new way to cheat, people will make bots play for them in competitive games, exactly what happens in games of complete information like chess and checkers)
AS i said, your already stated that you don't want to impose yourself utopic objectives, but now you seem to defend the idea that is productive and logical to work on the ai. I'm really confused.


I just didn't get what you did you want to tell me with the rest of the sentence, but i already knew it, and doesn't change the validity of what i say in any way i know.
Yes compared to commercial games wesnoth AI es very good, Commercial games AI as I said, are not decent AI, most AI can be beaten by any player that can move the pieces or units.

Sorry for my very poor english.

PD: Is true that Wesnoth doesn't have TONS of incomplete information like other games, since is very easy to scout, and puzzles form a major part of the neccesary skills to play well in wesnoth. While it means is a lot easier than other games, it's not easier enough.

[Dave]

Dave, really i don't understand your point.

My point is your comparison of Wesnoth being a "small team" relative to "big teams" like Blizzard is I think erroneous. I think that we can and do make AI's competitive with those made for Blizzard games. I think we have a team with some extremely talented people.

Yes, the characteristic of your team is relevant, because if you were 120 Genius, with all the time , joy, will and money in the world, with access to all the techonology and resources of humans at your disposal,

This is irrelevant because it is completely different with your comparison of us vs Blizzard. Blizzard doesn't have anything near this either. Sure if you go this far then having that many resources might help. *shrug*

this work would be possible in a decent framework of time.

What work? Making an AI of what magnitude? If you mean making a general Wesnoth AI that can play competitively with the best humans on any arbitrary Wesnoth scenario, then no I don't think it's necessarily possible in a decent framework of time.

A considerable amount of effort has been put into making an AI that plays Go well. Yet the best Go AI's are nowhere near world-class. Wesnoth is MUCH more computationally complex than Go.

David

[anakayub]

A build of a decent AI in a incomplete information game, one that can beat at least poor players, it's a project bigger than wesnoth, if you do, it will be 100 more relevant to most of the programming world than wesnoth.

I have serious concern over where you got that idea. A good AI improves both campaign and MP, and is relevant to Wesnoth. It allows AI difficulty to not be overly dependent on AI gold (and thus number of units).

As i said, there is plenty of people working on it with no success at all. And it would be a brillant advance to achieve that. (to be honest i hope it never happens, it will become a new way to cheat, people will make bots play for them in competitive games, exactly what happens in games of complete information like chess and checkers)

Are you sure of this, especially in a Wesnoth community where there are no rankings except in the unofficial ladder? A community where it's more of a friendly community rather than a wrestling arena for people to overly compete just to get on top?

AS i said, your already stated that you don't want to impose yourself utopic objectives, but now you seem to defend the idea that is productive and logical to work on the ai. I'm really confused.

No need to be. Working on a good AI is just as productive and logical as working on art and balancing for Wesnoth.

[Fosprey]

Anakyub, i meant that if wesnoth team would advance in building a powerfull AI, be sure that that advancement will be used in other games, of course i don't expect people to use bots to play for themselves on wesnoth.

PD and off topic: I played 3 levels of scepter of fire, the game is not hard at all even in higher difficulty.
At least not for now (33% of the game). Well that depends of course, i lose two times, because, 1.nobody expect a door to kill your hero, and two because in the level 3, it shows 24 turns to complete the objective, BUT WHEN YOU do, you are given another objective, but not extra turns.... I delay objectives on porpouse to maximize my exp gain, how i'm supposed to know that i will give a new objective with no extra turns?
Since the campaign isn't any hard but frustrating, very frustrating, i will quit.
BTW i doubt ANY campaign can be difficult, since all of them seems to lack a clock, thus, you have all the time in the world to think, so it's very hard not to do a very good move at each move.
Basically it's very hard to lose since you have all the time in the world to think and thus make all excellent moves.

[Aethaeryn]

Anakyub, i meant that if wesnoth team would advance in building a powerfull AI, be sure that that advancement will be used in other games, of course i don't expect people to use bots to play for themselves on wesnoth.

PD and off topic: I played 3 levels of scepter of fire, the game is not hard at all even in higher difficulty.
At least not for now (33% of the game). Well that depends of course, i lose two times, because, 1.nobody expect a door to kill your hero, and two because in the level 3, it shows 24 turns to complete the objective, BUT WHEN YOU do, you are given another objective, but not extra turns.... I delay objectives on porpouse to maximize my exp gain, how i'm supposed to know that i will give a new objective with no extra turns?
Since the campaign isn't any hard but frustrating, very frustrating, i will quit.
BTW i doubt ANY campaign can be difficult, since all of them seems to lack a clock, thus, you have all the time in the world to think, so it's very hard not to do a very good move at each move.
Basically it's very hard to lose since you have all the time in the world to think and thus make all excellent moves.

Obviously you're really skilled for the even higher-level campaigns. Play Northern Rebirth... on Nightmare... and every time you're in a bad situation and have to reload, make sure to reload from the start and not from an autosave (saveloading abuse is a waste of time, if you want to save time cheating through hard use debug mode to give yourself more gold - either way your victory doesn't really count). Good luck.

[Fosprey]

I think the problem isn't that i'm skilled, but i'm patient, and since there isn't a clock (like when you play online multiplayer) i have enough time to think the best moves, that's very hard to fail, just that.
That's the kind of thing you should add on on the campaigns to make them more intersting instead of adding a lsightly better AI, Adding a clock (this should be an option of course) and maybe a final score when you ended the campaign ( I don't know what should it be based on, but a scored based on certain achievements during the campagin, like how fast in turns, clock time, and maybe other stuff, did you finish the campaign)
A clock, a timer, optional, and maybe also a score, would help people like me (i'm more an online multiplayer player) to enjoy the campaigns. To be honest i was really enojoying the story of scepter of fire, but the lack of those features make me quit the campaign. I know other people who would have the same problem.
I think adding a timer it's very easy and would have great benefits since it will welcome campaigns to a kind of players that otherwise wouldn't, adding a score would improve it even further, Both things are a lot easier to implement than AI and i personally would benefit a lot more than a slighly better AI (and if it helps me, it means it will help others that think like me, yes they exist, i know some...)

[Grand Marshal Aditya]

I think that AI can be developed into a formidable machine. At least enough to give us humans a challenge. Of course it will take a lot of time, but it will work out in the end.

We don't have "Wesnoth Grandmasters" yet to make the AI useless...and there is always handicaps...

[Velensk]

When I origionaly posted there were two things I was thinking of, one of which I'm pretty sure isn't a problem, two of which that I'm sure is.

A minor thing that would help the AI in multiplayer would be if it could tell the diffrence between tanks, and fighters. If it plays loyalists it will always get masses of heavy infantry, and ignore spearman. Playtesting the AI with the Era of Myths is made even more useless , by the fact that for most factions it will olny recruit 3 units from it's arsenal, the one scout, fighter, and archer it likes best. An alternative might be to have it choose a random unit that fills the roll it's recruit list states.

The second thing that I think would help allot is if you could get it to understand specials such as backstab/posion. This is probably much harder to impliment than the first one. Possibly if position based abilities like backstab sent some sort of signal that the computer that makes it take an extra consiteration into it's placement. I've got no clue how it'd understand posion tacticaly since it does not seem to think ahead, and posion effects are specificaly set in the future.

[mowerpower]

Wesnoth is MUCH more computationally complex than Go.

This might be true, but I don't see how we can be so sure. Games of Go involve many more moves than typical Wesnoth MP games, and so have much deeper search trees.

I can imagine a research proposal that starts along the lines of "Complexity theorists have paid little attention to understanding the computational complexity of Wesnoth"...

[CarpeGuitarrem]

Wesnoth is MUCH more computationally complex than Go.

This might be true, but I don't see how we can be so sure. Games of Go involve many more moves than typical Wesnoth MP games, and so have much deeper search trees.

I can imagine a research proposal that starts along the lines of "Complexity theorists have paid little attention to understanding the computational complexity of Wesnoth"...

Wesnoth is invariably more complex for a computer to handle. The complexities of Go stem from the intense focus of attention on the simple gameplay. Both players know how to play, and can plan completely. Everything is out in the open, and there's no randomness. In Wesnoth, it's more realistic, and has more variables. Both are complex in their own way.

Wesnoth is worse, because of all of the variables that have to be taken into account.

[turin]

This might be true, but I don't see how we can be so sure. Games of Go involve many more moves than typical Wesnoth MP games, and so have much deeper search trees.

Not really. A game of go averages (according to Wikipedia) 150-200 moves. A game of Wesnoth never goes 150 turns, but a turn in Wesnoth is not the same as a "move". There's no clear unit of measurement to take as a "move" in Wesnoth, but the best choice is probably an action taken with a unit - each move in Go has at most 200 possibilities, but usually much less, and each unit in Wesnoth can move to probably around 60 hexes, depending on what its movement is, and then choose whether or not to attack and if so which unit - the number of possibilities for a single unit on a single turn in Wesnoth is roughly equal to the number of possibilities for a single move in Go.

So, if a side has 10 units, a reasonable amount to take as an average, then each turn in Wesnoth is equivalent to 10 moves in Go, and the total length of a Wesnoth game in moves would be a few hundred, depending on the length of the game. Wesnoth is thus at least as complex as Go, if we're just going from a search tree persective.

[Velensk]

Heh, another thing is that all results in go are predeterminable, in wesnoth each combat has a veriety of results so then there has to be a certain factor of calculating risks something that requires an algorithim for determining an absolute answer too if a risk is worth it or not, and some way of determining/calculating every single factor in such decisons, which is another problem because the AI thinks one unit at a time, and a large part of those factors is what you can do with your other units present.

[mowerpower]

This might be true, but I don't see how we can be so sure. Games of Go involve many more moves than typical Wesnoth MP games, and so have much deeper search trees.

Not really. A game of go averages (according to Wikipedia) 150-200 moves. A game of Wesnoth never goes 150 turns, but a turn in Wesnoth is not the same as a "move".

Quite so. I meant turn.

There's no clear unit of measurement to take as a "move" in Wesnoth, but the best choice is probably an action taken with a unit - each move in Go has at most 200 possibilities, but usually much less, and each unit in Wesnoth can move to probably around 60 hexes, depending on what its movement is, and then choose whether or not to attack and if so which unit - the number of possibilities for a single unit on a single turn in Wesnoth is roughly equal to the number of possibilities for a single move in Go.

So, if a side has 10 units, a reasonable amount to take as an average, then each turn in Wesnoth is equivalent to 10 moves in Go, and the total length of a Wesnoth game in moves would be a few hundred, depending on the length of the game. Wesnoth is thus at least as complex as Go, if we're just going from a search tree persective.

Not so fast! This analysis shows what the branching complexity (call it "B") of a position is; what I said earlier concerns the depth. AI algorithms will try to limit the number of positions they look at by pruning, and sometimes by spotting that different positions are tactically analogous.

Go is hard because good human players routinely perform analyses that involve making judgements about lines of play that don't pay off until 20 or so moves ahead, which are hard for AIs to match without doing lookaheads that deep. Say the depth complexity (D) is the amount of needed lookahead the AI needs to perform to be competitive.

To oversimplify, the search space is something like the exponential B^D. You get an equivalent increase to doubling D by squaring B. So the relatively high B for Wesnoth rapidly vanishes if its D can't keep pace with Go's.

Heh, another thing is that all results in go are predeterminable, in wesnoth each combat has a veriety of results so then there has to be a certain factor of calculating risks something that requires an algorithim for determining an absolute answer too if a risk is worth it or not, and some way of determining/calculating every single factor in such decisons, which is another problem because the AI thinks one unit at a time, and a large part of those factors is what you can do with your other units present.

I've not really thought this through, but it looks to me that Wesnoth's nondeterminism affects the size of B, but not D.

I mentioned nondeterminism as a possible carrot in Wesnoth for AI researchers...

[Dave]

Go is hard because good human players routinely perform analyses that involve making judgements about lines of play that don't pay off until 20 or so moves ahead, which are hard for AIs to match without doing lookaheads that deep. Say the depth complexity (D) is the amount of needed lookahead the AI needs to perform to be competitive.

Well, in Wesnoth humans do the same kind of thing. As an example, often a human will consider tactics such as attacking an enemy with a weak unit that has a powerful attack, such as a mage, and then follow in with fighter units who guard the mage, and typically also attack the enemy units.

The human must consider whether the fighter units can be positioned adequately to defend the mage, and whether the enemy attack will break through to the mage.

In all, if you count a unit move as one 'ply', this analysis goes very deep. It starts at the mage's move and attack, goes through the analysis of all the fighter moves, and then possible counter-attacks and so forth. Not to mention analyzing potential reinforcements from either side, and so forth.

For an AI to think this far ahead also is of great difficulty.

David

[mowerpower]

In all, if you count a unit move as one 'ply', this analysis goes very deep. It starts at the mage's move and attack, goes through the analysis of all the fighter moves, and then possible counter-attacks and so forth. Not to mention analyzing potential reinforcements from either side, and so forth.

Does the current AI (is it chesslike?) really branch on possible enemy moves when evaluating each of the unit moves that makes up the turn? That would be very interesting! I can't think, offhand, how I would go about constructing such an algorithm that wouldn't very wastefully reevalute the same enemy moves over and over again.