The challenges of developing a good AI to ever more complex games

[Panzer76]

First off, this is not a stab at WitP, more a comment of a industry problem. And that problem is the development of decent AIs.

The problem lies in the ever more complex nature of the games. Many of the old strategic games had out right good "AI", becasue the choices you had was very limited. Within this very strict framework, the AI could do senisble choices, less parameters to worry about.

These days, and for the past recent years it has become more and more obvious that the main obstacle facing good war games is the AI. You now have games that functions on the tactical / operations / strategy / operational levels, and while an AI usually can do OK when confined to tactics or even operations, the holistic perspective is lost.

Even tactical games, like Combat Mission, where the AI does not cheat, has problems coping with its "bigger issue", but does ok on a unit to unit basis.

This layering in the games are problematic for an AI which is uncapable of learning and rememebering. As any developer can tell you, the development of the AI is a time sink, because it never gets finished. You reach (hopefully) an acceptable level and then leave it. After that the returns on further development time is limited and shrinking.

So, where does this leaves us, the players, the users?

Well, as I can see, there are a few different possibilities.

1. Multiplayer only. Saves alot of time, focus on game mechanics instead. Works great for short attention span games like shooters, not so good for longer, turn based games. Right out unaccaptble for huge games like WitP due to the huge time requirements. (for most people) If you start playing the grand campaign in WitP over PBEM, you will almost for certain never finish it.

2. Accept a bad/mediocre AI. Well, seems this is pretty evident in many games already. And in an effort to compensate for the lacking AI, it gets various bonuses, which is anoher word for it cheating.

3. Development of an industry standard/engines? Would it be possible, in the same sense as you have audio and graphical standards, like DirectX one could develop a AI standard which developers could use?

Alternativly, could a AI engine be developed and others could license it in the same sense as you would lisense the Havoc physics engine or Source gfx engine?

This would really be more of an business idea (Panzer makes a note of it).

Anyhow, ultimatly we have to accept that an AI will perhaps never, and atleast not for a long time be as challenging as a skilled human opponent. But, does that mean we should accept bad AIs? Ofcource, what is a bad AI is a matter of discussion.

Again, before I get flamed too much, this post is not specifically targeted at WitP, but on a growing industry problem.

Cheers,

Panzer

[Ron Saueracker]

Probably easier to cure cancer. Big, big undertaking. Compensation forhuge effort probably not worth it.

[The Dude]

totally agree with you
while we are wishing, can i get hangover free liquor[:)]

[steveh11Matrix]

Not quite as bad as that, surely. I like the idea of the licensed 'plug-in' module.

For me, PBEM is out of the question. FTF/Hotseat or (more likely) vs the ai, so an ai that I haven't 'solved' or that doeas something a little different each time is a good thing!

Steve.

[Ron Saueracker]

ORIGINAL: steveh11Matrix

Not quite as bad as that, surely. I like the idea of the licensed 'plug-in' module.

For me, PBEM is out of the question. FTF/Hotseat or (more likely) vs the ai, so an ai that I haven't 'solved' or that doeas something a little different each time is a good thing!

Steve.

I used to play hotseat games of GG's Guadalcanal Campaign, Bomb Alley, North Atlantic '86, Carrier Strike...list goes on. I used to be quite a fan of his games...best on the market at one point. Had a buddy who was totally hooked and we'd just play for hours on end. PBEM is the only way to go now, though. Hard to find players who are into hard core games. It's all flight sims or real time crap now.

I actually play a little game called Tropico to kill time once in awhile. One of the few games which has endless replayability. Not exactly a wargame but I find it endlessly alluring. Don't flame me for this!!![8D]

[steveh11Matrix]

Hi Ron! For me, playing against another guy is a social thing, so FTF is the only way to go. I have one player near(ish)by and we 'indulge' occasionally. For FTF play however I prefer good old low-tech miniatures or counters, though.

I've heard a lot of people saying "Play PBEM!", but what they don't understand is that I want to play on my schedule, and that I get as much fun playing with the game(s) as actually playing them. It's a classic case of "What floats your boat".[;)]

Steve.

[Ron Saueracker]

ORIGINAL: steveh11Matrix

Hi Ron! For me, playing against another guy is a social thing, so FTF is the only way to go. I have one player near(ish)by and we 'indulge' occasionally. For FTF play however I prefer good old low-tech miniatures or counters, though.

I've heard a lot of people saying "Play PBEM!", but what they don't understand is that I want to play on my schedule, and that I get as much fun playing with the game(s) as actually playing them. It's a classic case of "What floats your boat".[;)]

Steve.

Got that right. FTF is a hoot when the combat replay is on and the pair of ya are tied to the screen, hooting and hollering like your at a college football game. Some of the language got quite impressive. Sign of a good game. PBEM is OK but it does not have the immediate element.

[Sneer]

I do not know mechanics of WITP
but I always thought that such scale game has multi- layered scripts.
one for aiming strategic targets ,defining threats and assigning troops. - this one is responsible for elastic play
second for carring out operations ( landing, defence, conwoys etc)
third defininig how actually troops behave.
Looks like there is nothing like the first one in WITP where a defined list of second stage scripts run AI.

I'm not an expert but this way I look at strategic games like this

[Captain Cruft]

It seems to me the only practical answer to this is to allow users to script their own AIs. The main reason developers churn out cr*p is because it's not worth them expendng the time and effort. On the other hand, it may be worth it to the users.

You're still not going to match a human player like this but I do believe there is scope for a lot of improvement.

The Lua embeddable scripting language might be a good technical solution as to how to provide the functionality ...

[Moquia]

Direct AI already exist. Don’t know how useful it is for strategic games though.

DirectAI

[Kitakami]

Having actually had to program a learning AI for the simple game of Pente, I have an appreciation for what it takes to program a system that actually learns from its own (and the enemy's) mistakes.

The programming side of it is a large undertaking, and I do not think you could do a generic AI that covers every type of game... it would have to take too much into account. The problem with that is the fact that a specialized AI would, although technically a success, not be an economic success

And to the above you have to add the fact that the knowledge databases would grow, and grow, and grow... the more you play, the bigger they'd get (that's how you store knowledge for the system).

And if you want the system to be efficient, you'd have a server online, where all the registered copies of the game would dump what they have learned from the player(s) that use them, and distribute the knowledge to all the copies by means of AI updates. Otherwise a registered copy would only learn from the local players, and a player with a different style might throw it into a fit.

Doable? Yes. Worth the while of whoever does it? Don't know. But if it ever gets done, count on an industry award, because a good AI is almost a dream that can't be reached... yet.

[Mike Scholl]

But it would be nice if you could get it to the point where if it "gets hurt" a couple
of times in an area, it would try to avoid that area while beating down whatever
unit(s) were causing it's pain. If a child can learn to avoid a hot stove, it would
seem that an AI could be programmed to avoid sailing TF after TF into harms way
without taking the slightest action to suppress or eliminate the source of the harm.
It does have access to ALL the information in the game.

[Mr.Frag]

But it would be nice if you could get it to the point where if it "gets hurt" a couple
of times in an area, it would try to avoid that area while beating down whatever
unit(s) were causing it's pain. If a child can learn to avoid a hot stove, it would
seem that an AI could be programmed to avoid sailing TF after TF into harms way
without taking the slightest action to suppress or eliminate the source of the harm.
It does have access to ALL the information in the game.

Mike, that logic would be completely valid if the AI tracked multiple turns worth of events. It could see that there was a problem "Stove is hot" and remember "I got burned here last time" coupled with a reverse trigger event "stove is cold now". Thats a reactionary system where things become learned based on things going wrong.

I don't think the WitP engine runs as a reactionary system. It seems to be an evaluation type system where it checks each turn to see if a set of conditions exist that would warrant a change "Air Bal = xxx" and chooses to do something else.

To get such a system, the game would have to track trends. That tracking obviously increases the memory and horsepower requirements in an exponential scale. (the farther you track, the more it increases). I doubt very much that type of system (although it would be fantastic) can be done on a PC based platform beyond more then a hundred units or so. WitP has roughly 12,000 ships/planes/ground units.

I'm all for it, but I think we are still years away from the horsepower required.

[Knavey]

ORIGINAL: Mr.Frag



I'm all for it, but I think we are still years away from the horsepower required.

But you know what...at one point we were still years away from being able to make a game like WitP...so I suppose that probably within the next ten years we might see something that comes closer to the expectations that players have for an AI enemy.

[Mr.Frag]

ORIGINAL: Knavey

ORIGINAL: Mr.Frag



I'm all for it, but I think we are still years away from the horsepower required.

But you know what...at one point we were still years away from being able to make a game like WitP...so I suppose that probably within the next ten years we might see something that comes closer to the expectations that players have for an AI enemy.

Certainly hope so, but have been watching a rather negative trend in new software development over the years. The ever increasing flood of piracy is bringing us closer and closer to the end of all PC based gaming software. I normally buy 20+ games a year. This year has been sad. 5 total. I play all types of games *except* FPS (really tough to play a 2 handed game when only one hand works - did I mention strokes are fun?)

It is a great thing that so many developers have found a home at Matrix. I look back over the years and remember such greats like 360, atomic, SSI, talonsoft and come here seeing most of the skill hiding in some new company that Matrix supports. Probably why I support these guys so much ... afraid of what will happen once they are gone.

[Thayne]

For the AI problem, I do not think that any solution would work short of a PBEM server where players email their games to the AI computer (which then learns from multiple games with different players), who then resolves the turn and emails back the moves.

This, of course, is only going to be possible for the more popular games -- it will not be cost effective otherwise.

Such a system would not need so much heavy programming. The game designers themselves can see what is happening in different games and make adjustments -- not necessarily waiting for the AI to learn from its mistakes, but instead forcing adjustments on the computer. Which, then, will affect all of the games that server is playing.

I can see no other cost-effective way for that portion of the game industry to advance.

[Mr.Frag]

ORIGINAL: Thayne

For the AI problem, I do not think that any solution would work short of a PBEM server where players email their games to the AI computer (which then learns from multiple games with different players), who then resolves the turn and emails back the moves.

This, of course, is only going to be possible for the more popular games -- it will not be cost effective otherwise.

Such a system would not need so much heavy programming. The game designers themselves can see what is happening in different games and make adjustments -- not necessarily waiting for the AI to learn from its mistakes, but instead forcing adjustments on the computer. Which, then, will affect all of the games that server is playing.

I can see no other cost-effective way for that portion of the game industry to advance.

Now thats a nifty thought ...

A PBEM dumping ground where you pick a side ... stored there are a whole bunch of games for that side to enter their orders.

You get the file, enter the orders, them submit it back ...

Other players download that turn ... do their stuff, pump it back up.

No one actually owns any game, they all just play that turn to the best of their abilities

[:D]

[Popoi]

I'm thinking multilayered scripts that are USER programmable... that would give the modding community something interesting to think about.

Using neural networks or other semi-non-determinitstic technologies i do not see as viable.. the testing would have to be insane to make sure the AI all of a sudden didn't decide to stack all units into korea, convert all planes to kamikazes and scuttle all his ships.

[ZOOMIE1980]

ORIGINAL: Captain Cruft

It seems to me the only practical answer to this is to allow users to script their own AIs. The main reason developers churn out cr*p is because it's not worth them expendng the time and effort. On the other hand, it may be worth it to the users.

You're still not going to match a human player like this but I do believe there is scope for a lot of improvement.

The Lua embeddable scripting language might be a good technical solution as to how to provide the functionality ...

This is the next silver bullet possibly. I think wargame developers, though, are going to eventually have to adobt structured disk assess data management where volatile data resides in a structured disk based system, like an RDBMS. Turn based strategy are not particularly high performance, high throughput designs, but they tend to cart around comparitively larges amounts of data. With something like that you can use sort of an "Expert System" type approach to make much more sophisticated AI's without unduely beating up system performance. And you can provide abilities for players to design their own AI's as well. And system performanace is always skyrocketing. You can now by 2GHz 512MB+ RAM systems for under $1000 (US) and there are a number of good, high perfomance, public domain database engines out there so software toolkit costs are not an issue.

Speaking of performance, on a side note, does anyone have any idea why, sitting totally idle, WitP consumes 85-95% of the CPU cycles on any machine it is installed on???? I've know they probably cache up all their bitmaps into memory at startup which explains the high memory utilization, but I don't get the CPU utilization? Sitting idle, the thing should be almost 0%, but it's not????

[ZOOMIE1980]

ORIGINAL: Mike Scholl

But it would be nice if you could get it to the point where if it "gets hurt" a couple
of times in an area, it would try to avoid that area while beating down whatever
unit(s) were causing it's pain. If a child can learn to avoid a hot stove, it would
seem that an AI could be programmed to avoid sailing TF after TF into harms way
without taking the slightest action to suppress or eliminate the source of the harm.
It does have access to ALL the information in the game.

Sanity checks. The AI needs to periodically ensure it is not going INSANE or it's opponent is not going insane.