The unfair player : the AI

[unmerged(10761)]

I wonder if the artificial "intelligence" for HOI will be improved, in that it is FAIR play.

I get horrendously frustrated when I see ,for example , in EU2 when my flottilla attacks a foreign flottilla with invasion army on board his ships, and ready to invade my provinces, that my OVERWHELMING flottilla will lose (panic) or even get sunk !! whereas when his ships are empty I DO get a fair battle result (I WIN..).
Also regarding producing new armies, ships, etc..the AI opponent can build and DOES build far more than a human controlled opponent.

WHEN WILL THIS NONSENSE STOP ???

When is it possible to REALLY create IN-TEL-LI-GEN-CE for the AI ?
It musn't be too difficult to figure out for a computer to find weaker defended regions of the (human) opponent and make the right alliances ??

 

[Peter Ebbesen]

Originally posted by snailtrailer
I wonder if the artificial "intelligence" for HOI will be improved, in that it is FAIR play.

I get horrendously frustrated when I see ,for example , in EU2 when my flottilla attacks a foreign flottilla with invasion army on board his ships, and ready to invade my provinces, that my OVERWHELMING flottilla will lose (panic) or even get sunk !! whereas when his ships are empty I DO get a fair battle result (I WIN..).
Also regarding producing new armies, ships, etc..the AI opponent can build and DOES build far more than a human controlled opponent.

WHEN WILL THIS NONSENSE STOP ???

Not in the forseable future.

When is it possible to REALLY create IN-TEL-LI-GEN-CE for the AI ?

Not in the forseable future.

It musn't be too difficult to figure out for a computer to find weaker defended regions of the (human) opponent and make the right alliances ??

It is trivial to find a weakly defended province. It is even trivial to find a weakly defended region.

However, it is non-trivial to project the results of invading that region some months on in game-time, and it is non-trivial to project the results on other fronts of withdrawing the forces needed to attack a certain area. Additionally it is non-trivial to project an enemy's reaction.

And it is non-trivial to arrange logistics.

And most of all, it is bordering on the impossible to write an AI which can assess a given world-situation overall in winning/losing terms, because any rules-based AI lacks vision.

The major reason that players of strategy games find an AI easy to beat isn't that they are good at strategy or logistics - merely that the AI is even worse at solving complex multi-variable non zero-sum games than they are.

With a very few exceptions (Chess comes to mind, since it is relatively simple), you are never going to see fair play by an AI in computer games. The question isn't: does the AI cheat, but rather: how transparently does it cheat.

 

[unmerged(10761)]

Re: Re: The unfair player : the AI

Originally posted by Peter Ebbesen


The major reason that players of strategy games find an AI easy to beat isn't that they are good at strategy or logistics - merely that the AI is even worse at solving complex multi-variable non zero-sum games than they are.
[/I].

(..sigh..) <<< and and I was already under the (mis)conception that I was Napoleon reincarnated >>>
<<< worse still, there are hundreds of them walking around freely today!!>>>

(sigh)

 

[unmerged(10704)]

Re: Re: The unfair player : the AI

Originally posted by Peter Ebbesen

Not in the forseable future.


It is trivial to find a weakly defended province. It is even trivial to find a weakly defended region.

However, it is non-trivial to project the results of invading that region some months on in game-time, and it is non-trivial to project the results on other fronts of withdrawing the forces needed to attack a certain area. Additionally it is non-trivial to project an enemy's reaction.

It IS trivial! AI in EU and EU2 sucks! A typical example: AI has 100k army and cant move it into enemy's province which
1) is completely defenceless
2) is just next to a province with 100k army
3) would fall from assault in several weeks

Another example: AI is siegeing a province. When it is almost taken - it... withdraws without any reason comes back and lays siege again!

And it is non-trivial to arrange logistics.

AI in EU2 during war moves its 100k armies from one proince to another and losing thousands of men from attrition. There is no TARGET for armies - they move without any reason!
Is it hard to implement such algorythm:
When war breaks out:
1) Find weak opponent province
2) Assign every army status: defensive or offensive and dont change this status to avoid senseless moves.
3) Assign weak province as a target for offencive armies and try to take it.
4) When enemy army enters AI's prowince assign it as a target for defensive army and go after it until it is destroyed.

In EU armies move in a very chaotic way. Directions of movement seem to be in 90% random. In 9 of 10 cases AI overlooks 100% opportunieties to destroy weaker opponent army or take over a province. Armies have no target or change targets too often and that's why AI in EU is very weak.

I really hope that in HOI AI will be _much better_ than in EU or at least making mistakes will be much harder.

And most of all, it is bordering on the impossible to write an AI which can assess a given world-situation overall in winning/losing terms, because any rules-based AI lacks vision.

The major reason that players of strategy games find an AI easy to beat isn't that they are good at strategy or logistics - merely that the AI is even worse at solving complex multi-variable non zero-sum games than they are.

With a very few exceptions (Chess comes to mind, since it is relatively simple), you are never going to see fair play by an AI in computer games. The question isn't: does the AI cheat, but rather: how transparently does it cheat.

AI can be very difficult (not neccesary unbeatable) opponent even without cheating - it's just the matter of good algorythms. Simple algorythms are often very effective.

 

[unmerged(2695)]

I disagree. The AI is pretty good at lifting sieges.

The AI is not, however, intelligent like a human being - thank God.

 

[Peter Ebbesen]

Re: Re: Re: The unfair player : the AI

Originally posted by comrade


It IS trivial! AI in EU and EU2 sucks! A typical example: AI has 100k army and cant move it into enemy's province which
1) is completely defenceless
2) is just next to a province with 100k army
3) would fall from assault in several weeks

Another example: AI is siegeing a province. When it is almost taken - it... withdraws without any reason comes back and lays siege again!

Hardly typical behaviour in EU2, but it does happen, and I agree that the AI would be better if it assaulted more often. However, even as a player I find it difficult to evaluate when an assault is likely to be succesful unless I have truly overwhelming odds on my side - how then should the AI evaluate the feasibility? Does it always make sense to move the 100K army in to assault, or is it only in some situations, in other words, does the AI need larger situational awareness before it sends those 100K to help in the assault?

AI in EU2 during war moves its 100k armies from one proince to another and losing thousands of men from attrition. There is no TARGET for armies - they move without any reason!

They move with reason, and they stay together. The programmers dilemma: Move armies in big stacks and see them wither from attrition, or split them in more manegable stacks and see them defeated in detail by the player, who is much better at gathering forces for a decisive battle.

Is it hard to implement such algorythm:
When war breaks out:
1) Find weak opponent province
2) Assign every army status: defensive or offensive and dont change this status to avoid senseless moves.
3) Assign weak province as a target for offencive armies and try to take it.
4) When enemy army enters AI's prowince assign it as a target for defensive army and go after it until it is destroyed.

No, that is fairly easy. Unfortunately, such lemming-like behaviour with rigid rules is exceedingly easy for a human player to exploit, once he has identified the offensive/defensive armies.

In EU armies move in a very chaotic way. Directions of movement seem to be in 90% random. In 9 of 10 cases AI overlooks 100% opportunieties to destroy weaker opponent army or take over a province. Armies have no target or change targets too often and that's why AI in EU is very weak.

No targets? 90% random movement? Could have fooled me. As far as I can see the AI operates with gathering points (which explains the recruited troops marching towards a common goal) and conquest targets, as well as a conservative bit of self defense.

The major issue I see with the EU2 AI, is that what with governing 100-200 countries and thousands of armies, the time-slice for individual unit AIs becomes very small, making it all but impossible for individual units to exploit temporary weaknesses.

I really hope that in HOI AI will be _much better_ than in EU or at least making mistakes will be much harder.

AI can be very difficult (not neccesary unbeatable) opponent even without cheating - it's just the matter of good algorythms. Simple algorythms are often very effective.

Well, since HOI will have higher minimum specs it seems likely that the AI can be improved in some ways just by getting a larger time-slice.

Out of curiosity, can you name a good strategy game that doesn't cheat? Just wondering.

 

[Murmurandus]

Re: Re: Re: Re: The unfair player : the AI

Originally posted by Peter Ebbesen

...
Out of curiosity, can you name a good strategy game that doesn't cheat? Just wondering.

Battle Chess...

 

[unmerged(358)]

And that is the problem... People actually think the program running Deep Blue (the computer that beat the reigning chess champion in repeated matches) is good AI, when it's not. First of all is that chess is a lot simpler game and with only a fraction of the options of EU and thus it's easier to make solid good algorithms for it (not to forget chess computers have been around the last 15 years). And then the important point of Deep Blue is that it's simply processing all the possible moves 10-15 moves in to the future. But since it's a speedy machine it can crunch all those within a normal timeframe. Essentially it is brute-forcing it's way through the problem And that is the only instance where an AI actually can beat a competent human.

Cobos

 

[unmerged(10704)]

Originally posted by Peter Ebbesen

Hardly typical behaviour in EU2, but it does happen, and I agree that the AI would be better if it assaulted more often. However, even as a player I find it difficult to evaluate when an assault is likely to be succesful unless I have truly overwhelming odds on my side - how then should the AI evaluate the feasibility?
Does it always make sense to move the 100K army in to assault, or is it only in some situations, in other words, does the AI need larger situational awareness before it sends those 100K to help in the assault?

EU2 is a game where agressive style of play (when in war) is the best style of play. I often attack enemy army twice as strong as mine just to "soften" them. Besides: battle results (when difference in tech levels isnt large) are often very unpredictable - you can relatively often defeat much stronger opponent. I don't calculate odds - I just attack and I dont care if I lose - I can always withdraw beaten army, wait until morale goes up and try again - AI is just too stupid to make any use of battle it has just won - it is much too passive. The only good thing in AI is assaulting - AI's assaults on fortresses are very effective.
Armies in EU2 can't stay in place - they must MOVE either on enemy's army or to siege enemy's province. And I can observe a lot of AI's _large_ armies standing in one province during war and doing nothing.

They move with reason, and they stay together. The programmers dilemma: Move armies in big stacks and see them wither from attrition, or split them in more manegable stacks and see them defeated in detail by the player, who is much better at gathering forces for a decisive battle.

There are no decisive battles in EU2 (exept from battles fought by one-province countries - loser is finished). When you lose a battle - you withdraw try again. Battles are in 99% won by reducing morale rather then killing all enemy. And defeating 30k army with a 15k army happens very often.

No, that is fairly easy. Unfortunately, such lemming-like behaviour with rigid rules is exceedingly easy for a human player to exploit, once he has identified the offensive/defensive armies.

And I ask - so what? This simple algorythm would make wars with strong countries a real challenge. When I play some minor (say Naples) and enter the war with major power (like Austria) - I can easily beat them because of AI stupidity. If the algorithm I mentioned was apllied - Austria would move ther bigger armies and crush me - this would be realistic. Instead - they just stay somewhere in the fields doing nothing or splitting, merging, moving from Wien to Tirol etc... and i laugh. Thanks to such stupid AI behavior Naples can conquer whole world in 300 yrs.
My basic objection against AI in EU2 is: large country with large army led by AI simply CANNOT crush smaller country.

No targets? 90% random movement? Could have fooled me. As far as I can see the AI operates with gathering points (which explains the recruited troops marching towards a common goal) and conquest targets, as well as a conservative bit of self defense.

Armies are gathered in one province just to be splitted and move in some weird direction. AI seems to be afraid to move army to battle.

The major issue I see with the EU2 AI, is that what with governing 100-200 countries and thousands of armies, the time-slice for individual unit AIs becomes very small, making it all but impossible for individual units to exploit temporary weaknesses.

This may be true.

Well, since HOI will have higher minimum specs it seems likely that the AI can be improved in some ways just by getting a larger time-slice.

Well let's hope HOI won't need P8 6 Ghz and 16 Gb RAM

Out of curiosity, can you name a good strategy game that doesn't cheat? Just wondering.

That's a good question
I can name some games that are good and AI doesnt cheat but it doesnt mean that AI is good For example: Steel Panthers, Close Combat, Clash of steel (one of my favourite games).
In games with hexagonal map and turns AI usually doesnt cheat.

 

[daedalus]

I think the main problem with the AI (in any game) is that she is unable to learn. This is even a bigger problem in strategy games like EU.
The human player, in the other hand, is a master on this. He learns from past experiences in game, he learns from game to game, and he can come to this site and learn strategies from other players. With all that, it is almost impossible fot the human player to lose.
The AI can't learn. She has be pre - programmed with a set strategy, that was prob OK at the begining, but it is rather quickly surpased by the Human player, once that he has learned her behaviour. And everytime you start a new game, she will 'forget' the last game, and start anew with the same strategy. And this behavoiur is repeated for every AI, in every computer where EU is played. Talk about an unfair player, she cant learn, and even if she could, she cant communicate to others AI what she has learn.

 

[HisMajestyBOB]

The million dollar question then, is how do you make an AI learn? If Paradox can do this, they'll be richer than Bill Gates.

 

[Darkrenown]

Originally posted by daedalus
I think the main problem with the AI (in any game) is that she is unable to learn.

There are a few games with learning AI, Fields of Battle comes to mind. Its a WWI TBS. The AI has a neural net which updates every turn and saves from game to game.

 

[unmerged(5120)]

Paradox's AI really ain't that bad, I even think it's pretty good.

The crux is, do you play "gamey" or "in character"?

If you try to make the kind of decisions a monarch of the times would make, then the AI will give you a run for your money.

On the other hand, If you rely on the fact that you can be Napoleon for 400 years and that for those 400 years your opponents won't change their mindset, of course you'll trash them.

With the former playing style you're playing a (very good) simulation, with the latter you're just cracking a puzzle.

Choose your own poison, but AFAIC, I find the simulation path by far the more enjoyable.

 

[daedalus]

Originally posted by Sire Enaique
Paradox's AI really ain't that bad, I even think it's pretty good.

The crux is, do you play "gamey" or "in character"?

If you try to make the kind of decisions a monarch of the times would make, then the AI will give you a run for your money.

On the other hand, If you rely on the fact that you can be Napoleon for 400 years and that for those 400 years your opponents won't change their mindset, of course you'll trash them.

With the former playing style you're playing a (very good) simulation, with the latter you're just cracking a puzzle.

Choose your own poison, but AFAIC, I find the simulation path by far the more enjoyable.

I cant agree more.

 

[unmerged(2833)]

Re: Re: The unfair player : the AI

Originally posted by comrade


And I ask - so what? This simple algorythm would make wars with strong countries a real challenge. When I play some minor (say Naples) and enter the war with major power (like Austria) - I can easily beat them because of AI stupidity. If the algorithm I mentioned was apllied - Austria would move ther bigger armies and crush me - this would be realistic. Instead - they just stay somewhere in the fields doing nothing or splitting, merging, moving from Wien to Tirol etc... and i laugh. Thanks to such stupid AI behavior Naples can conquer whole world in 300 yrs.
My basic objection against AI in EU2 is: large country with large army led by AI simply CANNOT crush smaller country.
[/B]

Ummm. Yes... you probably aren't aware of the most popular EU I strategy, are you?

Yes, i'm speaking of letting AI gather it armies in one your province, and wait till 90% of them will die due to attrition. What you propose is in reality a big step back

 

[unmerged(3168)]

Originally posted by Sire Enaique
Paradox's AI really ain't that bad, I even think it's pretty good.

The crux is, do you play "gamey" or "in character"?

If you try to make the kind of decisions a monarch of the times would make, then the AI will give you a run for your money.

With the former playing style you're playing a (very good) simulation, with the latter you're just cracking a puzzle.

Choose your own poison, but AFAIC, I find the simulation path by far the more enjoyable.

Couldn't agree more, the most enjoyable EU games I played were the simulation type.

 

[Thanak]

The biggest problem with EU AI is that it doesn't know how to properly cover provinces

 

[unmerged(10155)]

re

I can think of two games with truly intelligent AI.
First is Black&White where the creatures LEARNED what you thaught them, and the second is Dark Reign, a game overshadowed by Starcraft but with an AI times better than the aforementioned piece of crap. I remember playing skirmish and setting up defenses right in the direction of the enemy because it was my experience from Red Alert that the computer always strikes in the same place...

And I recevied a very nasty flank attack.

 

[unmerged(5228)]

This is all very interesting, but I think we are overlooking one important aspect: will we, the players be able to tell the difference between "AI mistakes" and how the programmers intentinally programed the AI to act. I can picture some of the complaints some people may send to Paradox:

"I was playing a game as Germany in the Grand Campaign and I was annexing minors left and right and the Allies did NOTHING to stop me. Not one solitary thing, even though they outnumbered me 10:1. Jesus, I even took Czechoslovakia, which Britain and France had an alliance with and they still refused to move their butts. Now it is 1939 and I have some serious military power. They AI should have acted while I was weak, but instead all I got was an occasional message from the League of Nations telling me to stop. Is this some kind of joke? You guys really need to fix the AI."

"OK, Hearts of Iron is a great game but the AI is not all that smart. I was playing a game as Britain and I kept on noticing the German AI making really STUPID mistakes. When the Germans conquered France (granted, the German AI was good enough to take on the French, but that really isn't much of an accomplishment. This is the French we are talking about here) and I had to evacuate my army from the continent. Now, as my guys are sitting on their butts at Dunkirk waiting for my fleet to pick them up, the Germany army is just sitting there right next to mine. Just sitting mind you- not attacking. WTF? They could totally destroy the BEF and they are just sitting around! Are they stopping to admire the French countryside? And that is not the biggest error. For a while the AI tried to cut off my supplies and destroy the airforce, but then suddenly out of the blue, Germany invades the Soviet Union!?! I'm sorry, I could't help but laugh at this one. Please, please, please do something about the AI!"

 

[unmerged(1847)]

Originally posted by Richard Nixon
This is all very interesting, but I think we are overlooking one important aspect: will we, the players be able to tell the difference between "AI mistakes" and how the programmers intentinally programed the AI to act. I can picture some of the complaints some people may send to Paradox:

"I was playing a game as Germany in the Grand Campaign and I was annexing minors left and right and the Allies did NOTHING to stop me. Not one solitary thing, even though they outnumbered me 10:1. Jesus, I even took Czechoslovakia, which Britain and France had an alliance with and they still refused to move their butts. Now it is 1939 and I have some serious military power. They AI should have acted while I was weak, but instead all I got was an occasional message from the League of Nations telling me to stop. Is this some kind of joke? You guys really need to fix the AI."

"OK, Hearts of Iron is a great game but the AI is not all that smart. I was playing a game as Britain and I kept on noticing the German AI making really STUPID mistakes. When the Germans conquered France (granted, the German AI was good enough to take on the French, but that really isn't much of an accomplishment. This is the French we are talking about here) and I had to evacuate my army from the continent. Now, as my guys are sitting on their butts at Dunkirk waiting for my fleet to pick them up, the Germany army is just sitting there right next to mine. Just sitting mind you- not attacking. WTF? They could totally destroy the BEF and they are just sitting around! Are they stopping to admire the French countryside? And that is not the biggest error. For a while the AI tried to cut off my supplies and destroy the airforce, but then suddenly out of the blue, Germany invades the Soviet Union!?! I'm sorry, I could't help but laugh at this one. Please, please, please do something about the AI!"