AI Development and Our Hobby

[jcorbin]

Guys,

I've been looking over several of the posts and I think it's time for everyone to face the music: It's a real SOB to create a really good AI opponent for these types of strategy games. Consequently, it sounds like we are literally years (if not decades) away from a developer ever being capable of creating a really good AI opponent for said games. Someone in another thread mentioned the "Hearts of Iron" series, which I own, and is in its zillionth iteration, dating back to the initial Hearts of Iron 1.0 release. And they still can't get it right, even when the computer player is allowed to dramatically cheat. And don't even get me started on the "Civilization" series, which has some of the worst AI I've ever seen, and relies almost entirely on its pedigree, combined with tons of chrome, to continue to sell millions of units. Meanwhile, the "Rome: Total War" series is only marginally better at maneuvering stacks of infantry on its rather handsome strategic level map. (I think all of the eye candy in these latter two games is specifically designed to mask the fact that the AI is rather mediocre, at best.)

I just think this stuff is simply too difficult to program to the level that the typical 25-year armchair general demands, due to an overwhelming number of variables that must be taken into consideration. I've been in software development as a technical writer for 15 of those 25 years and I have known literally hundreds of developers, including several in the computer entertainment field. I don't envy their task at all. Coding Deep Blue to eventually beat Kasparov at chess in 1997 is one thing; coding a superlative AI at this level of strategic gaming, with literally hundreds of units on the board and thousands of variables on any given turn, is entirely different. Now add to this the pressures of releasing a product on schedule, so the people who put up the money can attempt to realize, at the very least, some marginal return on their investment within a reasonable amount of time, and you have the present state of affairs.

I remember the "old days" where I couldn't find anyone locally to play games like "The Longest Day" and "Third Reich." They'd take one look at the Rule Book, then a second look at the hundreds of unit counters, and promptly run away with their hair on fire. And so, I would end up personally playing out various scenarios as the Axis and then the Allies. It looks like I might be doing the same with this game, or finally break down and carve out enough time to play some MP.

Someone find me a game where a strategic level AI is genuinely capable of:

1) Launching a coordinated large-scale amphibious invasion.
2) Effectively responding to a large-scale amphibious invasion.
3) Changing a nation's military unit production to counter what is actually occurring on the battlefield.
4) Changing a nation's technological research path(s) to counter what is actually occurring on the battlefield.

...and I might change my mind. (There are dozens more; I'll stop here.)

I've also been following the World in Flames thread, which is one of my favorite board games of all time, and will purchase in its PC incarnation when it eventually arrives. Years ago, when they got started with the computer port, there was some speculation in an Australian Design Group thread regarding whether a computer player AI was even necessary--that the game could conceivably stand on its own two legs as a pure multiplayer game. Obviously, they're presently trying to code an AI for WiF, having long since realized that the game would not sell in sufficient numbers without a single player mode.

Maybe someday we'll have what we all want in single player mode for our favorite hobby. But, it would appear, it's going to take an AI programming genius with lots of caffeine to get us there...

[YohanTM2]

Very well said, the issue is many players do not play agianst other humans and in order to sell viable quantities of a game you must have an AI.

As one of those players who never really plays the AI after one pass and the tutorials I would much prefer the development time and effort being spent on a better TCP/IP and most importantly PBEM. You need things like a solid replay and heavy anti-cheat provisions including some of the new online die rollers.

But, economics rear their ugly head except for the War in Europe effort.

[Warfare1]

jcorbin:

Very thoughtful insights.

I know that Strategic Command 1 does indeed have large Allied invasions of France when I play as the Axis.

AI development (or Programmed Opponent) is slowly gaining ground. But as long as graphics and chrome dominates PC games, then few developers will invest the time and money to improve it (which can be done if the will is there).

I am always surprised when I see games shipped without campaign/scenario/unit/tech editors. Since a gamer would be able to use these editors to make up for many AI faults.

Anyone familiar with Civ2's event scripting will know just how versatile and useful this editor is in creating some truly tough user scenarios. Anyone who has played Nemo's Red Front v1.4, for example, will experience some awesome game play using the game's original 1996 AI. Nemo pushed the game engine to its limits by exploiting all of the game's easy to use events and editors.

No game, and no developer, especially when it comes to making strategic games, can possibly cover all the areas that impacts the AI. That is why these types of games NEED editors, with the ability to mod and tweak every facet of the game.

It is not as though we live in a vacuum. Hundreds of games have come before our present time. We should learn what made the great games, great; incorporate their useful ideas, menus and interfaces; and finally, build into present games some of the same types of editors that made those classic games so playable.

[O`Connor]

Someone find me a game where a strategic level AI is genuinely capable of:

1) Launching a coordinated large-scale amphibious invasion.
2) Effectively responding to a large-scale amphibious invasion.
3) Changing a nation's military unit production to counter what is actually occurring on the battlefield.
4) Changing a nation's technological research path(s) to counter what is actually occurring on the battlefield.

Civilization IV.

OK, it's not a wargame, but it has a lot to deal with (more factors I would suggest than a number of wargames). In its latest incarnation, and (especially) modded with the Better AI mod, you can't take anything for granted. You definitely have to keep an eye on the coastline. And if there is a counter to your force, the AI will find and use it.

[Dave Ferguson]

I applaud Decision games decision NOT to include a AI in the upcoming war in europe, the game mechanics are simpler than CEAW and even so a competent AI would be almost impossible to develop. It seems I prefer playing wargames ON a computer rather than computer wargames, I must be one of a small number who do so.
 
Dave

[jcorbin]

Very good points, Yohan. I don't do much MP gaming, since I never seem to have enough time to hook up with other gamers, but I should probably start heading in that direction...

[IainMcNeil]

One of the issues with AI is that they tend to be developed for a specific game. Virtually nothing from CEAW would be any use in another game, even one that used the same basic mechanisms.

E.g. if you make a change that starts to charge a reasonable PP cost for units loading on to amphibious transports. This sounds like a trivial change, but the AI has to now know that it must save production points for an amphibious attack. This means it has to prioritise this versus other things like extra rail moves, research, recruitment and repair. If you weight any one of these badly you end up with the AI being screwey. It can a long time to get one tiny change like this balanced and this time is of zero value in any other game.

Now think about the AI required for weather, politics etc.

I think this is the problem - you can't build up a nice AI code base and the vast majority tends to be so game specific it's of no use in anythign else and is thrown away at the end of each game.

[jcorbin]

You're right. Editors can make a difference. For example, I've used the "Hearts of Iron: Doomsday" editor extensively and incorporated some AI mods. This combination has certainly improved gameplay, but we're not there yet...

[jcorbin]

I haven't tried the Better AI mod yet for Civ IV, but I will certainly look into it. Of course, it's not a wargame in the classic sense, but I am a fan of the series dating back to the original. I enjoy taking a break from some of the more realistic/historical games to do the "beer and pretzels" thing...

[jcorbin]

Great to see the game's developers responding to posts here.

Excellent point. The AI has to be game-specific. That means solid sales must drive continuing investment in that one product via point releases and, later on, sequels, in order to advance the AI. I wish you guys very robust sales.

Again, I give you guys all the credit in the world for tackling such a difficult task. And I never even bothered to mention play balancing in the original thread. It's not enough to create a competent AI; you then have to play balance.

Existing AI issues aside, you appear to have a terrific start here. I hope the product does well. We need more developers like you to continue to push the hobby, or we will all eventually end up with nothing but "Chrome Wars: Part X" from the publishing behemoths...

[jcorbin]

No AI? I was unaware. Now that's interesting. Most dev firms wouldn't even consider traveling down that bumpy road, due to the presumed negative impact on sales in forcing one to exclusively go the MP/PBEM/Hot Seat route.

So scrap the AI and put all one's emphasis on a terrific UI and ironclad MP options, eh?

Never mind interesting, that's downright ballsy...

[targul]

Computer War in Europe has never had AI didnt in the last version and wont in the next.  Game of that scale would be near impossible to write AI code.  Decision made that decision because he doesnt hire people to make his computer games he gets them to make he game and then he puts it out.  I believe he then gives some money to the designer but that was never made clear by the writer of CWIE 1. 
 
Game is very good but it is a historical port of the original War in Europe so it is really a 70's game on a computer without much advancement in gaming tech.  I love the game and still play it regularly. 
 
SC2 has extremely good AI if you play historically.  Most players tend to try to find the flaws in the AI and then abuse them.  They then state the AI is broken.  Personnnally I think the player is broken but that is for another discussion.
 
CivIII conquests has the best AI of any game of its like I have ever or probably ever will see.  It is very competitive and seems to have alot of inititive.
 
I too am awaiting World in Flames but that has been going on for many years and I am at the point that I doubt its actual release ever.
 
This game has good AI in Russia.  But if it has AI in the Med it has the worst of any game I have seen in many years.  Tried 3 games and it still has not invaded France or even tried so I hear there is a landing but it must be very rare. 
 
Anyway adequate AI can be written but if you want it to do everything it will not.  It can be written to recreate many WWII actions but players who avoid history it will never be able to beat because you cant program against odd actions only preceived actions.

[IrishGuards]

What's and AI [8|]
IDG

[targul]

An AI is the thing we are not using in our games.

[Zakhal]

Galactic Civilizations 2 has a really tough nonscripted AI that doesnt cheat (only little in the very last AI level). There are like 10+(?) AI levels in it and I was only able to beat like first 5 of them. The game doesnt even have multiplayer and they continiusly patch the AI to be even more better (not like it allready is almost unbeatable).

The difference is that the whole series is built with AI in mind. Game doesnt even have features that would be hard to code for AI to use (allthough they have managed to do som like beliavable AI diplomacy and there are planetary invasions). Also space is easier for AI as gameboard because the only units that move in it are spaceships and they can go anywhere.

As for pbem the only game I really enjoyed it was VGA planets. That was mostly because:
(1) You always got a set amount () of gameturns per week. Whether som player wasnt able to deliver his moves the host still made a new turn.
(2) There could be up to 11 players and players could quit/be replaced with zero disruption to the game.

Longest vga planets game I had going lasted like 1 1/2 years with 9-11 players.

Steel panthers is a great example of good TCP/IP gameplay. I have always enjoyed it. I hate to do like one turn and then wait for 28 hours for another turn. Life is too short for that.

I think the optimum multiplayer system would be:
(1) More than two players
(2) Has both TCP/IP and pbem
(3) Ability to switch multiplayer type every turn. If you have pbem game and want to do several turns today through TCP/IP all you have to do is change the mp type. Later on you can change back to pbem.

[pad152]

You know what you call a wargame without an AI? a board game.[8|] I think developers need to include some way of allowing users to edit the AI? WITP is a great game, but the AI is scripted and ends up doing the same thing in every game. It's tuff enough to create a good playable game let alone a good AI.

It's been said many times there are only three types of AI; dumb, cheats, and (dumb + cheats).

[Warfare1]

I've done a lot of modding of games, and some pretty old ones at that. And with only the use of editors, and with no additional tweaks to the AI, I have turned out some tough scenarios to play.

My general feeling is that I don't think the AI can play on an even playing field with a human player - cheats ARE needed - but make them subtle.

For example, if I want the AI to invade somewhere, and if it has to decide how to use PPs between rail, tank movement, and unit builds, etc, then, I would allow the AI to invade without the need to use PPs. Simply make AI invasions FREE.

A cheat yes - but the important thing is to get the AI moving... After all, the USA in the real war had virtually unlimited resources and it was supplying war material for the war in Europe AND the Pacific (say "100 aircraft carriers by 1945" 10 times fast).

AI invasions in a WWII European game are far too vital and important to be left to the whims of the AI and its resources. Remember, we are talking about American and Canadian production.

While AI scripting is game specific - the editors that are created with the game are also game specific.

Let's take an example of what happens in games that are scripted but have NO editors.

For example Uncommon Valor. It is a great game, but it has virtually NO editors (well there is sort of one but I won't go there for now). It has invasions but these are tightly scripted - you can't change them. Forces are historic, and once used up, that's it.

Now, what would happen if it had a decent editor along with additional force pools? Well, the player would be able to select a land hex, select units to be included in an invasion force (including ships and troops), select the experience levels, then select a date or turn number for the invasion to occur for that scenario, and then press the "save" button. Do this a few times, and the player will now experience multiple enemy AI invasions throughout the scenario at the locations he selected.

Unfortunately, scripted games without editors have virtually no user created scenarios available.

On the other hand, games such as the original Strategic Command does come with a great editor and there are currently more than 100 user created campaigns. A game not tough enough? Open the editor, make a few tweaks, and presto-chango, the difficulty goes up.

As long as an AI is competent, good editors can do the rest...

[targul]

Why do you care if the AI cheats?  I certainly do not.  I only care if the AI is good making the game fun and playable.  If it must cheat to give a good, fun, playable game do it.  I will buy it and it will be fun.

[ijontichy]

CEaW will give you a great and fun game in the eastern front. No doubt about that. It's just in some of the strategic aspects where the AI looks really stupid. I would say the developers need to focus their efforts on improving the sea-land invasion AI, and the Mediterranean needs to be looked at as well. Other criticisms of this game I would give to all computer games: not enough innovation (hey, that's why we have board games, right? [;)])

[Marc von Martial]

I think the problem is also that some wargamers expect they get an AI that is as capable as their best wargaming buddy (that does not cheat). This is nonsense and way to high expectations. Scientists burn billions each year to let a robot do things a 2 year old could do better, and no the robot does not do it on his own too [;)]

If a game AI "cheats" that is no bad thing at all. You can not copy human cheating over to computer cheating. Human cheating is negative, for sure. Computer "cheating" is necessary. Computers only do what you tell them, they do not do the thinking.
Right now the main problem with computer game AIs is that they hardly "learn from mistakes", unlike humans. Your MP buddy will do this. Once we are at a stage where computers game AIs can learn certain things we will see better artificial opponents.

There is certainly room, sometimes a lot of room for improvement with some games AIs, but the expectations should not be set too high too. We have had also several cases where gamers simply did not bother to raise the difficulty level and then complained about a bad AI when they play on "easy level". Some people are extremely good at finding patterns of behavior within games and thus have no problems to defeat the AI even when it cheats extremely. It will be extremely hard to please these people with a good AI until scientists and coders have developed a real AI that has human patterns of thinking and learning