AI

[AussieGiant]

Ah yes, the "AI can easily be fixed" topic has raised its attractive but subtly annoying head...again.

Of course the modding community can fix the AI, that's why they are all working for military or banking organisations in their spare time. Replicating the human brain is what you do when you have a few hundred spare hours of free time.

[Aradan]

You can't put the AI in the soft-code any more than you can put the graphics engine. What modders might be able to change is some the various thresholds, yes/no options and modifiers/multipliers the AI uses, but the main structure of the AI code needs to be hardcoded. And programming good AI, finding the perfect balance between performance and quality, is very hard (it's nothing modders can do, it's not like changing 2d or scripting, one would need a degree in informatics at least) and money-intensive. And of course it's a fact that AI is not as cost-effective for game developers as shiny graphics are.

[Lemur]

I'm a programmer.. nothing about AI for games like Total War is 'easy'.. nothing at all.

I'll never forget me and my friends stumbling through RoboWar, a 1992 game that asked you to program the AI for simple battle tanks. Brutal stuff, and not a hundredth as complex as TW AI.

I know programming AI is hard. And I seem to recall that there was some business where the dude who created the ETW AI left the company before release, or something like that. I'm under the impression that CA got royally gah-ed in the AI department and has been playing catch-up ever since.

That said, CA has a real budget, can hire real programmers, and has both a fiduciary and aesthetic obligation to get its house in order. That includes releasing its next generation of TW game with something better than Black Knight diplomacy (or omnium contra omnes AI, as I've come to think of it.)

[Gregoshi]

I'll never forget me and my friends stumbling through RoboWar, a 1992 game that asked you to program the AI for simple battle tanks. Brutal stuff, and not a hundredth as complex as TW AI.

I recall playing a game with the same theme but I don't think it was "Robowars". However I came away from that game with a similar appreciation for the complexities of AI. Trying to imagine what the AI programming must look like for any game in the TW series makes my head spin. There are just too many variables to consider.

Anyone who thinks AI is easy, doesn't understand AI. Try this: write down detailed instructions on how to get a drink of water. Everytime and wherever you are and you want a drink of water, follow your instructions TO THE LETTER. Remember, every decision you make in getting a drink of water has to be written down in your instructions. Do your instructions consider what kind of water is available? Sink, bathtub, toilet, bottled, pitcher, lake, ocean, stream/river, drinking fountain? Depending upon the source, do you really want to drink that water? (GAH! what if we have a choice of water sources?!) Let's assume we determine the availabe water is from a sink. Now you need a cup. What kind of cup is available? Is it clean? Is it broken? Where is it? How do you search for a cup? In your search for a cup, did your instructions included opening up the cupboard door if you search there? Which size cup to you get? Which hand do you pick it up with, i.e. is any hand unavailable? You have your cup now and let's assume your instructions including getting to the sink successfully (pathfinding!). Do you get hot water or cold water? What kind of controls are there for the water spigot? I hope your instructions cover every type of spigot you could possibly run into otherwise you will have to make a choice of operation actions which may look rather silly to anyone watching you try to turn it on. Etc, etc, etc. The simple act of getting water isn't so simple after all. We make a lot of decisions we aren't even aware of, but AI has to be aware of them.

No, AI is not easy.

[Vladimir]

And breathe.

[Intranetusa]

Ah yes, the "AI can easily be fixed" topic has raised its attractive but subtly annoying head...again.

Of course the modding community can fix the AI, that's why they are all working for military or banking organisations in their spare time. Replicating the human brain is what you do when you have a few hundred spare hours of free time.

I prefer to think of modders as geniuses who can memorize a million values of Pi and do complex differential equations in their heads...

Don't spoil my hopes and dreams...

[Crazed Rabbit]

If CA can't make a decent AI, they should make the AI non-hard coded...

Then some modders/programmers like the EB team will just fix up the AI and port Asia-ton-Barbarorum over ez peazy... :D

Soren Johnson (designer and leader coder for Civilization 4) addresses that an essay titled Seven Deadly Sins for Strategy Games.

5. Locked code/data

Protecting your code and data is a very natural instinct – after all, you may have spent years working on the project, developing unique features, pushing the boundaries of the genre. Giving away the innards of your game is a hard step for many developers, especially executives, to take. Nonetheless, we released the game/AI source code for Civ 4 shortly after shipping, and – so far – the results have been fantastic. Three fan-made mods were included in the game’s second expansion pack – Derek Paxton’s Fall from Heaven: Age of Ice, Gabriele Trovato’s Rhye’s and Fall of Civilization, and Dale Kent’s WWII: The Road to War – and so far, these scenarios have been heralded as one of Beyond the Sword’s strongest features. These mods would have been nowhere near as deep or compelling (or even possible) if we had not released our source code. For many PC developers, I’m preaching to the choir, so I’d like to be very clear that the problem is worst amongst strategy games. For whatever reason (perhaps the lack of a pioneering developer like id Software?), strategy developers have been much more closed off to modding than their shooter and RPG brethren. There are exceptions, like Blizzard’s fantastic scenario editor for WarCraft 3, but by and large, strategy modders do not have many places to turn for platforms on which to work, which was one reason we felt compelled to focus on modding for Civ 4. Giving stuff away can feel good. It should also feel smart.

Good AI may not be easy, but it is possible. CA has no excuse for not including it, or not unlocking code.

CR

[Gregoshi]

And breathe.

[Vladimir]

[Alexander the Pretty Good]

Soren Johnson (designer and leader coder for Civilization 4) addresses that an essay titled Seven Deadly Sins for Strategy Games.

Good link, CA is probably guilty in the past of the "too much stuff" sin as well. I think they may be turning back the clock a bit with STW2 but they could probably cut even more.

[hoom]

A part of the seemingly better AI in Shogun/Medieval was scripting I believe. Seem to recall mention of this in one of the CA interviews building up to RTW.

With the limited number of battlefields, the AI could be given per-battlefield hints that would help out the generic AI.
eg there was one battlefield in Shogun where the AI would nearly always hide a decent force in woods that were up a hill & on the flank of the easiest route to their main force.
By the time you came to them your units were a bit tired (or lots if you tried to flank it/ran) & then with the surprise attack morale penalty your army might get broken by the initial charge. That map was always a PITA unless you forced a Strategic retreat or caught them with a mostly Ashigaru army that could be broken fairly easily.

But the AI in later games has only the dynamic part & never uses surprise attacks like that :(

Another thing is that the units need to be properly balanced to bring out the AI algorythms.
There are a bunch of features that are MIA from the released version of the newer games but show up nicely in re-balanced mods.
If an algorithm is written expecting a certain range of values & works nicely within those values but a late balance change causes the typical case to be either outside those values or always at one end of the range, then to most players the entire algorithm will appear to be completely absent.

Also, on the Strategic level the Risk type is much more easy for the AI to understand what to do, which is a big part of why I want that back/separated from the Theatre level manoeuvring.

I played Shogun I demo once and AI totally destroyed me, where I was used to weak AI of Rome.

Ah yes. If you could win that battle, you could pretty much win any battle where you had a reasonable chance of winning.
A bunch of it is again in the balancing, the AI is very confident that it has superiority & comes straight for you. Your small number of Samurai Archers need to be fought very well to get in enough shots to reduce the AI numbers sufficiently to allow you to win the inevitable hand to hand fighting.
If the AI just milled around like it does when it doesn't know what to do you'll destroy them easily.

RTW AI is brought out excellently by the balancing in RTR:The Iberian Campaign. Lots of the battles there (including the first as Carthage!) have the AI confident that its got superiority & they come straight at you, engaging along your full front, scary.

[Monk]

A part of the seemingly better AI in Shogun/Medieval was scripting I believe. Seem to recall mention of this in one of the CA interviews building up to RTW.

With the limited number of battlefields, the AI could be given per-battlefield hints that would help out the generic AI.
eg there was one battlefield in Shogun where the AI would nearly always hide a decent force in woods that were up a hill & on the flank of the easiest route to their main force.
By the time you came to them your units were a bit tired (or lots if you tried to flank it/ran) & then with the surprise attack morale penalty your army might get broken by the initial charge. That map was always a PITA unless you forced a Strategic retreat or caught them with a mostly Ashigaru army that could be broken fairly easily.

I remember that map, vividly. It always seemed to be raining when my troops arrived as well.

I could picture my men just slogging up that muddy hill, only to see the flags of the enemy raise from the woods as the ambush sprung. Some of the toughest battles from STW belong to that map, well, that and any bridge battle.

[O'Hea]

I think the real difference between Shogun/Medieval and the later games was the AI's awareness of the environment. The computer was able to recognize advantageous terrain and take advantage of it, and if you maneuvered to take that advantage away they would respond. From Rome on, the AI either stands its ground or advances straight towards you. It really has no concept of maneuver or of the significance of a hill, or a river, or a valley. And the "Art of War'-inspired AI is unlikely to affect this, since even if they do manage to internalize some of the Art's principles, it'll mostly affect strat map behavior.

[TinCow]

I ran across an article on S2TW's "Sun Tzu" AI claims by an AI programmer. The general gist is that it's all marketing fluff because the basic elements of Sun Tzu's Art of War are already used by nearly all AI designers. So, essentially, CA is just claiming that they're doing the same thing that everyone else has already been doing for a long time. Nothing new but the marketing.

[Gregoshi]

I ran across an article on S2TW's "Sun Tzu" AI claims by an AI programmer. The general gist is that it's all marketing fluff because the basic elements of Sun Tzu's Art of War are already used by nearly all AI designers. So, essentially, CA is just claiming that they're doing the same thing that everyone else has already been doing for a long time. Nothing new but the marketing.

Yeah, in MW2, I remember seeing those birds flying over the trees where the sniper was.[/sarcasm]

Some good points in the article and maybe the Sun Tzu claim for S2TW is fluff - or maybe not. Any game wanting to use Sun Tzu also has to have the elements built-in environmentally for the AI to act upon his "common sense" - at least in some of the cases. The birds I mentioned in sarcasm being one such point. The game requires more than just AI programming to utilize Sun Tzu effectively. While some of Sun Tzu's rules are generic, others are more applicable and match perfectly to the older forms of warfare such as STW/MTW/RTW. And, while one can read Sun Tzu and say "yeah, no duh!" and roll one's eyes at the obviousness of it all, how many of us actively apply these rules to our battles (in history and in game) aside from the most generic of them? Knowing them is one thing, using them is something else. Same would hold to the AI/game programmers.

Besides, I think we have a host of STW players here aside from me who might contest the author's claims that everybody does it in there games.

Er, sorry if this comes off a bit snippy. Didn't mean for it to but it probably is.

[ReluctantSamurai]

I've seen several topics here debate 2D vs. 3D maps.

My own take on that is that it really doesn't matter which is used...obviously 3D looks much better and will more than likely be the type used for Shogun II. But there are two things about the 3D maps, as done since RTW, that greatly impact how the AI will perform, and there are several gross omissions by CA, as well.

1. The addition of roads guaranteed that combat will no longer be for control of a province, like STW I, but simply a set of skirmishes for the rights to siege.

2. The addition of cities, and the placing of income and population in them, guaranteed that the majority of battles fought will be siege battles for control of a provinces population and resources.

Both of these additions greatly increased the decision-making processes needed by the AI on the campaign map, and AFAIAC, the AI has never been able to fully cope with. I see no good way to compensate for #1 except to adhere to one of the most basic rules-of-engagement: if an army moves into any non-allied territory, it must stop its' movement upon entry and there is an immediate DoW by the country intruded upon. While this doesn't affect path-finding problems (which any good map-maker should be able to spot through play-testing) it will certainly put an end to some of the stupidity like dumping an army on a foreign shore where they stand around for years doing nothing because there is no automatic DoW for doing so.

If resources were placed outside of cities, and population loyalty could be controlled by having a garrison in the province rather than the city itself, perhaps there would be far less boring sieges.

If basic ZOC rules were applied, then there'd be less of those time-wasting battles in AI vs. rebels, where the rebel army continues to retreat two or three hexes. As a corollary, basic movement rules should apply, as well. A slower-moving army cannot retreat in the face of an attacker with more mobility, except where terrain might permit.

All of these things might go a ways towards reducing the number of decisions required by the AI each turn perhaps helping it to make better ones.

[Hooahguy]

If CA can't make a decent AI, they should make the AI non-hard coded...

Then some modders/programmers like the EB team will just fix up the AI and port Asia-ton-Barbarorum over ez peazy... :D

=(

no, not the EB team. Darth will do it!

[vartan]

If CA can't make a decent AI, they should make the AI non-hard coded...

Then some modders/programmers like the EB team will just fix up the AI and port Asia-ton-Barbarorum over ez peazy... :D

No offence to the EB dev team, but what makes you think they've any better a shot at making the AI "smarter"? There are plenty of research going on right now at institutions devoted just to this, and you want the AI to be made open source? Appalling.

[drone]

No offence to the EB dev team, but what makes you think they've any better a shot at making the AI "smarter"? There are plenty of research going on right now at institutions devoted just to this, and you want the AI to be made open source? Appalling.

I think you would be surprised. The AI often takes a backseat in game development, so time/budget constraints can ruin it. Modders working on their own time can do wonders. There are several RTS games where the mod community have vastly improved the AI for single player use.

[Marcus Caelius]

given the state of the AI in the tw series, i think it would be folly to consider it a core technology that must be protected to keep the competitors at bay. As it stands, no competitor would benefit significantly by having access to this AI.

CA, however, stands to benefit massively from having input from other people with regards to AI design.

It should also be possible to have multiple AIs, so that different factions/generals/maps can load up a different AI. Even if none of these is inherently superior to the standard AI, the very fact that you don't know which it will be, will limit your ability to set up knowing the weaknesses. But also tailor made AIs for different kinds of terrain and defenses could be made.

Furthermore CA could encourage AI vs AI battles with a cash prize for the winner in a tournament. This would rapidly spur development in battle AIs.

All of the development gains made can be used by CA in subsequent titles.

The result is that at best, they will have outsourced a major development area for nothing much while they can focus on other things like graphics.

At worst the job is too challenging for the community and nothing serious changes to the AI but the grumbling might stop.

But none of this is likely. What is likely is they will protect what is not worth protecting, rather like their own CAI and BAI in fact. There will likely be some more slight improvements, but sieges will still be ridiculous and battles easy and an AI simply unable to cope with the possibilities provided - and there will just be a few tweaks by modders.

The wishlist will be postponed for the next game.

[Hosakawa Tito]

Just think of a STW campaign without the Hojo Horde and suicide daimyos reducing your competition to rebels.

Two of my biggest pet peeves, especially the suicidal daimyo charge. My foggy memory seems to recall a flanking exploit with cav strung out into a single line that was rather silly & bogus as well. Fix that stuff and it should be as fun as the original.

Actually, the sengoku jidai period is one of the very few that makes sense for CA's Hobbesean all-against-all AI. Everybody wants to become Shogun, or be a close (preferably by marriage) ally of the dude who wins. Hmm, well, I don't know if CA can handle that second part. Allies who stick with you even when you're winning? Unthinkable.

Hehehe, allies who stick it to you, inconsiderate cherry blossoms. Human nature really stinks sometimes.

[Reenk Roink]

given the state of the AI in the tw series, i think it would be folly to consider it a core technology that must be protected to keep the competitors at bay. As it stands, no competitor would benefit significantly by having access to this AI.

CA, however, stands to benefit massively from having input from other people with regards to AI design.

I agree with this sentiment. If CA is implementing brand new AI stuff in S2:TW that is for sure better, then by all means keep it away from your competitors, but if it is very similar to what the older games have been using, why not open it up to modders to give you insights on how to make that brand new state of the art AI for the next game.

That is, if that's the way it works, not really sure...

[Tomisama]

Game AI will absolutely never be as good as facing a human player, even for practice, and especially not when playing with allies.

The ultimate experience to be found in any Total War game; is with other people.

Even as difficult as it may be sometimes to find persons who are as dedicated to honourable combat as you yourself are, it is still worth the effort to seek them out.

[Sasaki Kojiro]

I think you would be surprised. The AI often takes a backseat in game development, so time/budget constraints can ruin it. Modders working on their own time can do wonders. There are several RTS games where the mod community have vastly improved the AI for single player use.

Besides, wasn't it common wisdom that one of the reasons for the AI/unit balance problems was the CA wasn't that good at the game? Certainly in shogun there were a few very simple things that could have been done, for example on river defense:

1) Put a spear unit to cover for cavalry charging over the bridge and getting behind the defenders
2) Not charging the monks in to attack a solo spear year unit (and getting the monks decimated by archers)

But if you never figured out those tactics as a player while attacking a bridge, you won't end up coding them I guess...

[Hosakawa Tito]

I recall playing a game with the same theme but I don't think it was "Robowars". However I came away from that game with a similar appreciation for the complexities of AI. Trying to imagine what the AI programming must look like for any game in the TW series makes my head spin. There are just too many variables to consider.

Anyone who thinks AI is easy, doesn't understand AI. Try this: write down detailed instructions on how to get a drink of water. Everytime and wherever you are and you want a drink of water, follow your instructions TO THE LETTER. Remember, every decision you make in getting a drink of water has to be written down in your instructions. Do your instructions consider what kind of water is available? Sink, bathtub, toilet, bottled, pitcher, lake, ocean, stream/river, drinking fountain? Depending upon the source, do you really want to drink that water? (GAH! what if we have a choice of water sources?!) Let's assume we determine the availabe water is from a sink. Now you need a cup. What kind of cup is available? Is it clean? Is it broken? Where is it? How do you search for a cup? In your search for a cup, did your instructions included opening up the cupboard door if you search there? Which size cup to you get? Which hand do you pick it up with, i.e. is any hand unavailable? You have your cup now and let's assume your instructions including getting to the sink successfully (pathfinding!). Do you get hot water or cold water? What kind of controls are there for the water spigot? I hope your instructions cover every type of spigot you could possibly run into otherwise you will have to make a choice of operation actions which may look rather silly to anyone watching you try to turn it on. Etc, etc, etc. The simple act of getting water isn't so simple after all. We make a lot of decisions we aren't even aware of, but AI has to be aware of them.

No, AI is not easy.

And breathe.

Not without detailed instructions Vlady, sorry. Have to wait for the next patch, and then the fix will probably break drinking water.