Are Wargame AIs Fated to Suck?

[sIg3b]

Umm . . . anyone up for some thread necrophilia?
It's kind of unusual to see the last poster in the thread before it went dormant 6 years ago also be the one to resurrect it.

Man I started the thread. Stands to reason that I am qualified to resurrect it.

[pzgndr]

Third Reich does an excellent job at limiting the outcomes on the strategic level to what was historically possible. It also does a good job of showing what was most important (the economy!). It does this by consistently throwing out the irrelevant little details. Such as one German infantry corps doing 10% better than another. They are all the same perfectly interchangeable 3-3s. And that´s what I like about Third Reich: It does away with the random anecdotical stuff and focuses on what was important.

Agreed. Regarding wargame AIs, I spent years working on and improving the AI and Event scripting in my Advanced Third Reich mod for Strategic Command WWII - War in Europe. I wanted challenging and competent AI for both Axis and Allies, to include variable strategies. I have also spent years on code development improving the AI in Empires in Arms. This stuff is nontrivial!

I'd argue that it's not impossible to program challenging and competent AI, but it takes a very long time to do so and get it right. The major problem with wargame AIs is that developers and players want games released as soon as possible, so that developers can make money and players can get their new toy. AI development is usually not a priority prior to release, and then game patches attempt to make up for shortcomings. Until the developers need to start work on a new game, to make money, and then the old game AI is left in whatever state it was in. Like I got Empires in Arms when I did, and then I needed to figure out somebody else's code and start making some enhancements and improvements. Again, it takes a very long time to do all this. Years! At least for single developers. It's not like any of our niche wargames have staffs and budgets on the scale of World of Warcraft. If only...

[Karri]

However, my attitude to all this is: If a game does have (unintentional) exploits, it´s a sloppy design and I don´t like it.

That´s why I think the Heart of Iron series is basically crap.

The only games with no exploits are games that come with a very strict set of rules and a very limited number of actions that can be taken, or things that can happen in general. And the very basic exploit you are going to run is simply knowing what the AI will do.

To add to the discussion, most games don't even use an AI that is programmed to win. If it was, players would more often than not hate it.

As for Hearts of Iron series, there are so many moving parts that I consider the AI pretty good for what it is. It still sucks, but at least you can have a bit of fun with the game.

[sIg3b]

Third Reich does an excellent job at limiting the outcomes on the strategic level to what was historically possible. It also does a good job of showing what was most important (the economy!). It does this by consistently throwing out the irrelevant little details. Such as one German infantry corps doing 10% better than another. They are all the same perfectly interchangeable 3-3s. And that´s what I like about Third Reich: It does away with the random anecdotical stuff and focuses on what was important.

Agreed. Regarding wargame AIs, I spent years working on and improving the AI and Event scripting in my Advanced Third Reich mod for Strategic Command WWII - War in Europe. I wanted challenging and competent AI for both Axis and Allies, to include variable strategies. I have also spent years on code development improving the AI in Empires in Arms. This stuff is nontrivial!

I'd argue that it's not impossible to program challenging and competent AI, but it takes a very long time to do so and get it right. The major problem with wargame AIs is that developers and players want games released as soon as possible, so that developers can make money and players can get their new toy. AI development is usually not a priority prior to release, and then game patches attempt to make up for shortcomings. Until the developers need to start work on a new game, to make money, and then the old game AI is left in whatever state it was in. Like I got Empires in Arms when I did, and then I needed to figure out somebody else's code and start making some enhancements and improvements. Again, it takes a very long time to do all this. Years! At least for single developers. It's not like any of our niche wargames have staffs and budgets on the scale of World of Warcraft. If only...

The road they took with AlphaGo was to just teach it the rules and let it play against itself a zillion times to learn from experience until it could defeat every human player.
I wonder if one could do something like that with a game like 3R, provided one had a supercomputer and unlimited resources.

[sIg3b]

However, my attitude to all this is: If a game does have (unintentional) exploits, it´s a sloppy design and I don´t like it.

That´s why I think the Heart of Iron series is basically crap.

The only games with no exploits are games that come with a very strict set of rules and a very limited number of actions that can be taken, or things that can happen in general. And the very basic exploit you are going to run is simply knowing what the AI will do.

To add to the discussion, most games don't even use an AI that is programmed to win. If it was, players would more often than not hate it.

As for Hearts of Iron series, there are so many moving parts that I consider the AI pretty good for what it is. It still sucks, but at least you can have a bit of fun with the game.

Strict set of rules and limited number of actions is good.

I have no fun with a game where I can conquer Europe with Rumania. It should be tremendously difficult to do with Germany. A historical game needs some accuracy. Leave the fantasy and science fiction stuff to fantasy and science fiction.

[pzgndr]

The road they took with AlphaGo was to just teach it the rules and let it play against itself a zillion times to learn from experience until it could defeat every human player.
I wonder if one could do something like that with a game like 3R, provided one had a supercomputer and unlimited resources.

I am not an AI programming expert, by any means. I simply have a perspective regarding the AI necessary for WWII ETO like Third Reich and the Napoleonic Wars based on what I've been working on for about 15 years. There are diverse strategies required for early-, mid-, and late-game. Unit builds cover many different types of forces with different attributes. Throw in diplomacy, research, and economics and the games become quite complex. Go has only black and white pieces (not even different, as in chess), a relatively limited game board, and simple rules. I can see where a supercomputer could run Monte Carlo calculations to assess what strategies work better than other strategies. But I think our more complex grand strategy wargames would require more non-linear optimizations and a lot of exceptions to rules to achieve desired realism and historical accuracy, and play balance. That's a tall order, assuming you have the "perfect" set of rules. If we had unlimited resources, I could see better use of them for developing competent and challenging computer opponents for the games we already have. Particularly the classic AH and SPI board wargames with relatively simple CRTs and TECs. Instead, we create ever more complex games that leave us wanting for a good solitaire gaming experience. Perhaps someday. Maybe.

[Dabo]

Third Reich does an excellent job at limiting the outcomes on the strategic level to what was historically possible. It also does a good job of showing what was most important (the economy!). It does this by consistently throwing out the irrelevant little details. Such as one German infantry corps doing 10% better than another. They are all the same perfectly interchangeable 3-3s. And that´s what I like about Third Reich: It does away with the random anecdotical stuff and focuses on what was important.

TR gives a very chess like vibe, it’s a perfect information system and luck plays a very small role: you can almost always predict the result of an attack: you only have 3 possible outcomes A, D and EX and even at 1-1 odd there’s only one in six chance that the attack could fail, 1 in 36 at 2-1, die rolls are pretty much used to determine the entity of the attacker losses, something that I find extremely apt for a strategic level game and historically accurate: very few major operations in WWII failed because of botched initial attacks. (Kursk, 1st El Alamein, Market Garden, Rezhev which could be all considered 1-2 attacks).
Even TR successors, Advanced TR and a World at War while adding more uncertainty (mobilizations, shipbuilding, secret research and diplomacy) have maintained the same predictability at the tactical level, actually it even increased it with the revisited CRT (a 2.5-1 attack always guaranties the defender elimination)

[sIg3b]

I love the 3R CRT. The predictability is just right, with 2:1 giving you a victory nearly always. I also like the small chance of catastrophic failure, which will happen to you about once or twice per game, so you have to take the possibility into account.

[Dabo]

The TR family would really deserve a computer version (there was an old AH version of TR with a terrible AI still playable in dosbox) but I fear it would require a WiF level of commitment, IICR there was an initial agreement between Bruce Harper (aTR and AWAW designer) and Matrix to produce an AWAW computer version but it came to naught. Warplanner (a sort of dedicated VASSAL for ATR and AWAW) is great tool for PBEM and solitaire (and even FtF you don’t have the room for the massive maps) but of course it doesn’t have an AI and doesn’t enforce the rules.

[altipueri]

I am beaten by most game AIs unless I play on the really simplest level.

Last night I played Advance Tactics and got a win on AI- and had a couple of hours good fun.

I also played an hour of Sid Meier's Alpha Centauri and a bit of Conquest of the Aegean (Command Ops).

Oh, and I played the Gettysburg scenario from Ageod's American Civil War.

There's so many good enjoyable games if you aren't obsessed with showing how brilliant you are.

[ernieschwitz]

I am beaten by most game AIs unless I play on the really simplest level.

Last night I played Advance Tactics and got a win on AI- and had a couple of hours good fun.

I also played an hour of Sid Meier's Alpha Centauri and a bit of Conquest of the Aegean (Command Ops).

Oh, and I played the Gettysburg scenario from Ageod's American Civil War.

There's so many good enjoyable games if you aren't obsessed with showing how brilliant you are.

Did you find the new Phoenix patch for ATG fun?

There has been some AI work on it.

[altipueri]

@ ernie

I'll look at it - thanks for the tip. I think I played one of your scenarios a few years ago. Time flies.

[wodin]

Thankfully in an odd way I suck so I rarely have been let down by AIs!!

[Kuokkanen]

Thankfully in an odd way I suck so I rarely have been let down by AIs!!

I suck too, and often AI suck even more. In tactical scale games I get pwned mostly by AI's superior numbers and/or better equipment. In X-Piratez (modded OpenXcom) I had my operatives surround a door, AI goon opens it and walks through, turns around to walk back, and gets shot by opfire to its rear. I used to do that in the original UFO: Enemy Unknown. In SPWAW I managed to stonewall a mechanized battalion with just a reinforced jaeger platoon (Finland vs Soviet Union). In TOAW's Tutorial/Korean scenario with AI playing North Korea barely gets over the border. All in all, not impressive.