AI question - would you mind if... ? (Full Version)

All Forums >> [New Releases from Matrix Games] >> Empires in Arms the Napoleonic Wars of 1805 - 1815



Message

JanSorensen -> AI question - would you mind if... ? (6/19/2008 9:35:46 PM)

Which of the following three options do you find most reasonable for the AI?
a) The AI is allowed to cheat and always know the exact number of factors in every corps? OR
b) The AI is allowed to cheat and always know the total number of each type of factor each MP has? OR
c) The AI should not be allowed to cheat by knowing such things - instead it needs to be coded in such a way that it keeps track using the same means a human player has.



NeverMan -> RE: AI question - would you mind if... ? (6/19/2008 9:46:58 PM)

Ideally?

C



pzgndr -> RE: AI question - would you mind if... ? (6/19/2008 10:25:33 PM)

This is not a simple question.  Depending on how Marshall sets up a decision matrix or something, (b) may be necessary for the AI to make decent choices.  (c) may or may not work too well.

Instead of the "most reasonable" option, consider implementing things like this as part of difficulty levels.  Easy could have (c), no help for AI.  Medium could have (b), which should allow for better strategic decisions.  Hard could have (a) and (b), for better operational and strategic decision making.

Other AI options would be to give leaders some bonuses to their default ratings.  Maybe +1 strategic rating on medium, and +1 tactical rating on hard.  When the depot supply limit gets implemented, AI could get +1 on medium and +2 on hard.  Things like this would help compensate for AI weaknesses in other areas. 



baboune -> RE: AI question - would you mind if... ? (6/20/2008 12:53:50 AM)

B)

With a bit of A.

The AI at starts should not know how the factors are distributed but (like human player) should know the total forces of the human player (how many factors are available and how many corps). When the AI fights or during foraging, the AI should remember and build a map of the counters and forces available.

Of course each reinforcement phase would introduce a bit of chaos but that would be a "fun" part of the game. Following that the AI could start rebuilding his map of the player's forces.



eske -> RE: AI question - would you mind if... ? (6/20/2008 1:49:51 AM)

Anything at all, that makes the AI interesting to play against [8D]

Let it know all, and then add some random 'mistakes', if there is too few surprises.
So what if you can't bluff the AI.
Read somewhere 'You can't bluff the daft'. So what exactly do you loose, by letting it know it all ??

/eske



NeverMan -> RE: AI question - would you mind if... ? (6/20/2008 2:50:40 AM)

The problem is that right now it gets VP/PP that it shouldn't get and it's STILL WAY TOO EASY to destroy.

I'd rather play against an AI that was making decisions based on some logic and strategy rather than one who gets free VP/PP, factors, etc and/or one that beats me every fourth battle because it knows which chit I picked and/or fixes the die rolls.

That's not AI. That's not what AI does. If that is what the AI is going to be they should not call it AI, they should call it playing against the computer.



delatbabel -> RE: AI question - would you mind if... ? (6/20/2008 3:59:54 AM)

(a) this isn't a discussion about AI strategy. That's being worked on entirely separately by Marshall. There have been significant discussions within the test team about that, and hopefully Marshall's taking on board some of those ideas. Hopefully the AI's strategy will get smarter over time, and even if it does need a help getting VPs and PPs on the board, ideally in future it will not be so easy to destroy. This is about guiding the AI's thinking in one particular area, which is whether it should "know" or "guess" the composition of an enemy stack as the basis for its decision making process.

(b) Otherwise I agree with pzgndr. I think that perhaps (a) should be the case for "easy" level AI, (b) should be the case for "medium" level AI, and (c) for "hard" level AI.

Note that in "base mode" EiA all corps strengths, but not their locations, are public knowledge. Hence all players have (b). A lot of gaming groups, especially PBEM ones, tend to go with complete fog of war as a house rule, however. This means most players have (a) unless they are good at guessing their opponents' build strategies (and that builds of ships are always made public) in which case they have (b). Essentially (a) is what is implemented on a player by player basis in EiANW.

The number of corps in an enemy stack is always known, and therefore the AI could make some kind of decision making process inside (a) where it perhaps guessed that all enemy corps were approximately 15 factors, maybe 20 factors for the French and 10 for the Turks and 12 for the Russians, but about 15 on average. If (b) were implemented then it would say "well, I know his army has 80 factors, I can see 10 corps on the board, so I'm going to guess that each corps contains 8 factors on average". If (a) were implemented then the AI would count your factors in each stack.



NeverMan -> RE: AI question - would you mind if... ? (6/20/2008 6:04:26 AM)


quote:

ORIGINAL: delatbabel

(a) this isn't a discussion about AI strategy. That's being worked on entirely separately by Marshall. There have been significant discussions within the test team about that, and hopefully Marshall's taking on board some of those ideas. Hopefully the AI's strategy will get smarter over time, and even if it does need a help getting VPs and PPs on the board, ideally in future it will not be so easy to destroy. This is about guiding the AI's thinking in one particular area, which is whether it should "know" or "guess" the composition of an enemy stack as the basis for its decision making process.



It shouldn't know or guess, it should make a decision based on the facts. There is a difference between guessing and making an informed decision, and by informed I don't mean informed of data that would otherwise not be available to a human player, I mean making an informed decision just like we (humans) do when we play the game. That's why it's called AI.

An AI does not "guess" or "know", it makes an informed decision. This is what I would hope for.



eske -> RE: AI question - would you mind if... ? (6/20/2008 11:34:58 AM)


quote:

ORIGINAL: NeverMan

The problem is that right now it gets VP/PP that it shouldn't get and it's STILL WAY TOO EASY to destroy.

I'd rather play against an AI that was making decisions based on some logic and strategy rather than one who gets free VP/PP, factors, etc and/or one that beats me every fourth battle because it knows which chit I picked and/or fixes the die rolls.

That's not AI. That's not what AI does. If that is what the AI is going to be they should not call it AI, they should call it playing against the computer.


I would prefer 'playing against the computer' that is a challenge over against an AI thats a walkover.
Let me refer to the famous Turing test: If you - by its behaviour - can't tell if its a machine or a human, you have artificial intelligence. (..or something like that..). But remember humans make mistakes.

Playing EiANW against a system, thats knows all your secrets and controls everything in the game and always reacts optimally would hardly be any fun. You need to feel you got a chance to win and your decisions can influence that chance. And a good decision against a human player should also be a good decision against an artificial player. If not its turned into a different game.

So I actually agree with neverman. The 'AI' can't be allowed to cheat, in a way that changes or reduces the relevance of a human players decisions. Examples of cheats that would be off-limits is die-rolls, chit-picking, DoWs, control of minors and of course all basic game mechanics.

The 'AI's cheats should give it advantages that you can respond to, or at least doesn't change you playing goals. Extra resources, PP or VP are ok, I think. Or advantages that increases its abilities to mimic a human player. And this is what this thread is all about, isn't it.

An information gathering system could in most cases determine the location of a nations forces. A human player would be able to do that and/or guess it. Simply giving that information to the 'AI' increase its 'human likeness' so I'm all for it. Especially if you allow it to 'guess wrong' by randomly switching the location of a few of the corps. This way you still have a bluffing chance [;)]

/eske



pzgndr -> RE: AI question - would you mind if... ? (6/20/2008 3:16:19 PM)

The original idea of each MP AI having a personality could be reconsidered. This would provide a little more uncertainty and variability in AI performance.



NeverMan -> RE: AI question - would you mind if... ? (6/20/2008 7:49:43 PM)


quote:

ORIGINAL: eske


quote:

ORIGINAL: NeverMan

The problem is that right now it gets VP/PP that it shouldn't get and it's STILL WAY TOO EASY to destroy.

I'd rather play against an AI that was making decisions based on some logic and strategy rather than one who gets free VP/PP, factors, etc and/or one that beats me every fourth battle because it knows which chit I picked and/or fixes the die rolls.

That's not AI. That's not what AI does. If that is what the AI is going to be they should not call it AI, they should call it playing against the computer.


I would prefer 'playing against the computer' that is a challenge over against an AI thats a walkover.
Let me refer to the famous Turing test: If you - by its behaviour - can't tell if its a machine or a human, you have artificial intelligence. (..or something like that..). But remember humans make mistakes.

Playing EiANW against a system, thats knows all your secrets and controls everything in the game and always reacts optimally would hardly be any fun. You need to feel you got a chance to win and your decisions can influence that chance. And a good decision against a human player should also be a good decision against an artificial player. If not its turned into a different game.

So I actually agree with neverman. The 'AI' can't be allowed to cheat, in a way that changes or reduces the relevance of a human players decisions. Examples of cheats that would be off-limits is die-rolls, chit-picking, DoWs, control of minors and of course all basic game mechanics.

The 'AI's cheats should give it advantages that you can respond to, or at least doesn't change you playing goals. Extra resources, PP or VP are ok, I think. Or advantages that increases its abilities to mimic a human player. And this is what this thread is all about, isn't it.

An information gathering system could in most cases determine the location of a nations forces. A human player would be able to do that and/or guess it. Simply giving that information to the 'AI' increase its 'human likeness' so I'm all for it. Especially if you allow it to 'guess wrong' by randomly switching the location of a few of the corps. This way you still have a bluffing chance [;)]

/eske


This is my problem:

With added VP/PP the game could look like this:

You destroy the "AI" and are riding high in the dom zone almost every eco phase. You have between every country at least once and gotten unconditionals every time. You own most of the map. You lose because one of the MPs has managed to get more VP than you.

How dumb is that? Why would anyone want to see this outcome? This isn't AI, it's just plain silly.

I agree with those that think each MP AI should have a few different personalities (one of which it will pick to use for each game). So if you were playing solo, there are 6 MP AIs. Of those 6, let's say that each has 7 personalities to choose from. That makes for a good amount of combinations of games, particularly if you take into effect that each personality will not always make the same little move every time but be adhering to an overall strategy, thus making the combinations of games almost limitless.



Marshall Ellis -> RE: AI question - would you mind if... ? (6/20/2008 8:47:41 PM)

You're right Neverman. I could make the AI beat all of you today with that change but that would not make the game very fun :-) or you guys very happy.

I don't think (And will not add) the VP/PP thingy without an improved AI. It's not the only answer but maybe just a part of the equation.



Jimmer -> RE: AI question - would you mind if... ? (6/20/2008 10:53:32 PM)


quote:

ORIGINAL: Marshall Ellis

You're right Neverman. I could make the AI beat all of you today with that change but that would not make the game very fun :-) or you guys very happy.

I don't think (And will not add) the VP/PP thingy without an improved AI. It's not the only answer but maybe just a part of the equation.

Right. At the present time, we can add it in ourselves by simply adding a larger bid than the default 0.

As someone pointed out to me, the bidding process can really even out a game with players of different levels. However, the computer has to be at least equal to a poor human player before the bidding would me it acceptable vs. the AI.



NeverMan -> RE: AI question - would you mind if... ? (6/20/2008 11:10:29 PM)


quote:

ORIGINAL: Marshall Ellis

You're right Neverman. I could make the AI beat all of you today with that change but that would not make the game very fun :-) or you guys very happy.

I don't think (And will not add) the VP/PP thingy without an improved AI. It's not the only answer but maybe just a part of the equation.


I guess I can see how having the VP/PP as part of the equation would help balance out a weak AI, although ideally (which I know is very hard to achieve) I would still prefer a pure AI.



Marshall Ellis -> RE: AI question - would you mind if... ? (6/21/2008 12:24:15 AM)

I too would prefer a "pure" AI but reality trims my dreams!
I'll be trying for a "pure" AI until I die ... if it means anything :-)

BTW: I actually think that the VP thing (IMO) makes the AI more real because nobody ever played France without bidding 100 VPs (OK not 100 but you know what I mean).




Jimmer -> RE: AI question - would you mind if... ? (6/21/2008 12:58:50 AM)


quote:

ORIGINAL: Marshall Ellis

I too would prefer a "pure" AI but reality trims my dreams!
I'll be trying for a "pure" AI until I die ... if it means anything :-)

BTW: I actually think that the VP thing (IMO) makes the AI more real because nobody ever played France without bidding 100 VPs (OK not 100 but you know what I mean).



One step at a time. Small steps, too. If the AI gets only twice as good as it is, players will have to change tactics. It's still only maybe 1/20th of a human, but, if it moves forward slowly.

Once you get some AI changes in place, I intend to try my pet tricks and see which ones still work. I'll be reporting how I get around them, too. As a hacker, there's no way I won't be able to find loopholes. But, if they are hard to find or hard to implement, that makes the AI stronger.



JavaJoe -> RE: AI question - would you mind if... ? (6/21/2008 1:53:48 AM)

Marshall,

Can you persuade the AI to always conspire to defeat the Human?

Instead of trying to code for 6 individual nations just make them all one evil demented nation mindset.

Let's say for example I'm GB. The whole world hates me and will not help me in any way. They all go out of their way to kill my ships and invade my Island. I surrender to them but they will have none of this they want to crawl through the computer screen and kill me.

That would be the level of AI that would satisfy me.

It would then be playing like a normal person.




Marshall Ellis -> RE: AI question - would you mind if... ? (6/22/2008 8:04:17 PM)

Joe:

I did add a small mod to decrease the chances of the AI going after another AI opponent. The end result (Although not direct) should be more AI players going after humans. This effect would be less visible the more with more human players but more visible in solo vs AI.



j-s -> RE: AI question - would you mind if... ? (6/23/2008 9:22:06 AM)

One big broblem is that it's impossible to co-operate with AI. You can't play with GB and use some german help to beat up France, you have to beat france up with just GB army (sonds crazy!). This is a problem, becouse Prussia can't play well if it will not co-operate with Austria, for example.

I would develope combinent movement something like that: You can offer a move to your ally (AI), who will accept or reject your proposal for AI move. That would be a big advantage, for example Combine movement with me and attack Napoleon with all present corps.

Then I would add dominant powers to the game. That would give one clear strategy to AI, aim to the dominant power.



pzgndr -> RE: AI question - would you mind if... ? (6/23/2008 3:49:29 PM)

I'm more with j-s in wanting an AI that behaves more like a human opponent.  Each MP computer opponent should be playing to win, cooperating with human and AI players as appropriate, and also going after human and AI players as appropriate.  In the course of a game if an AI opponent is going after me and another AI opponent, then fine give it a slight bonus to go after me a little more vigorously.  But I wouldn't want say GB to just go after me as Prussia or Austria just because I'm a human opponent.  So I'm a little leery about the suggested mod, unless it's geared towards the higher difficulty settings for each MP?  This is something we should reserve judgement on until we see the AI enhancements implemented and determine whether we need this or not. 



gwheelock -> RE: AI question - would you mind if... ? (6/24/2008 3:08:37 AM)

I would recommend a "gang up on the LEADER (largest vp %)" - no matter who/what it
is ... THAT is the way that rational HUMAN players play.



Edfactor -> RE: AI question - would you mind if... ? (6/25/2008 8:49:09 PM)

I would say use option A for the AI



Page: [1]

Valid CSS!




Forum Software © ASPPlayground.NET Advanced Edition 2.4.5 ANSI
0.609375