Targeting efficiency [Archive]

Pocus

October 12th, 2003, 08:00 AM

Originally posted by Psitticine:
The missile AI seems to target Groups now, as that's more efficient, but it'll go after single targets it they are high-priority or simply the only things within range.

On a missile-related side-note, I had a pleasant surprise the other day. I was playing as Man against Ulm, and was rather surprised when the AI cast Arrow Fend since the opposing side had no missile Users. Right after that, my horde of Longbowmen opened fire on the clashing melee line, inflicting no casualties because of the Arrow Fend but absolutely demolishing the opposition. I'd never seen (or maybe just never noticed) it doing that before, and it was beautiful! http://forum.shrapnelgames.com/images/icons/icon10.gif arrow fend to prevent friendly fire? Thats cute indeed. http://forum.shrapnelgames.com/images/icons/icon7.gif

Mortifer

October 13th, 2003, 09:11 PM

Originally posted by Maelstorm:
I hope that the Dominions 2. AI will be lot better.You must be kidding, the Doms I. AI was lot better than the AI of the current strategy games. Try a medium map with 3-6 AIs on the highest level. You will be owned. Yeah a human can beat the AI in 1v1, but if more (4+) AIs are set, it is LOT harder to win, you need luck.
+ The Doms II AI will rock some booties if true, so maybe even the 1v1 AI will kick you hard! http://forum.shrapnelgames.com/images/icons/icon10.gif

[ October 14, 2003, 04:00: Message edited by: Psitticine ]

Jasper

October 14th, 2003, 08:13 AM

Originally posted by Mortifer:
<blockquote>quote:<hr />Originally posted by Maelstorm:
I hope that the Dominions 2. AI will be lot better.You must be kidding, the Doms I. AI was lot better than the AI of the current strategy games. Try a medium map with 3-6 AIs on the highest level. You will be owned. Yeah a human can beat the AI in 1v1, but if more (4+) AIs are set, it is LOT harder to win, you need luck.
+ The Doms II AI will rock some booties if true, so maybe even the 1v1 AI will kick you hard! http://forum.shrapnelgames.com/images/icons/icon10.gif <hr /></blockquote>No, I don't think he is kidding. I've talked to lots of people who have no problem stomping all over the Dom 1 AI, despite the fact it starts out with an advantage over human players. It's not bad, but it doesn't particularily appear to be better than other similar games either. I'd say both the Aow2 and Warlords 3 AIs are better.

I find the AI easy to abuse, and only difficult when you have lots of AI players + Ermor. The other AIs have no clue how to fight Ermor, which proceeds to have free reign and potentially grow quite large before you face it.

It's been a while since I've played solitaire Dom 1, but as I recall the AIs just don't expand very fast, and aren't tactically prepared when battle is joined.

[ October 14, 2003, 07:14: Message edited by: Jasper ]

DominionsFan

October 14th, 2003, 09:53 AM

Yes well, the Dominions II. supposed to be lot better, if true...we shall see. http://forum.shrapnelgames.com/images/icons/blush.gif

johan osterman

October 14th, 2003, 10:31 AM

The strategic AI can handle more situations in dom 2 than in dom 1, it is slightly better but it isnt Deep Blue. I hardly ever play single player TBS games so I do not know how it holds up against the AI's of other games. Shrapnels beta testers and Shrapnels personel thought the AI was very good, but an experienced dom player will still no doubt beat th AI in one on one matches.

Pocus

October 14th, 2003, 10:39 AM

Originally posted by johan osterman:
The strategic AI can handle more situations in dom 2 than in dom 1, it is slightly better but it isnt Deep Blue. I hardly ever play single player TBS games so I do not know how it holds up against the AI's of other games. Shrapnels beta testers and Shrapnels personel thought the AI was very good, but an experienced dom player will still no doubt beat th AI in one on one matches.the important things to consider, as it was the AI main weaknesses in doms I are :
is the AI aware that a supply rule exists.
is the AI prone to ward his armies, somehow intelligently.

in a pbem game we just finished, playing as Ctis, I got invaded by Pythium AI (Grana map). Even at 1 versus 5 I was able to wade thru their provinces, just because I casted each time foul vapor/poison ward, and they were totally caught off guard each time.

This is the kind oh problems that the AI have. The AI on the other hand is very tough when you counter them 'fairly', that is you attempt to have superiority on the battlefield without magic.

Wendigo

October 14th, 2003, 12:00 PM

Indeed, to follow up on what Pocus & others point out:

The Dom I AI cannot stand vs a vet player either strategically nor tactically:

Strategically:
-It doesn't build the right unit-mix to face its opponent (with lack of priest power when facing Ermor being the most notable failure).
-It disregards supply & usually fields starving hordes.
-It does not patrol, nor builds local defense so it is totally open to sneak attacks.
-Cannot equip supercombatants & mages in a competitive way.

Tactically:
-It just bunches its troops in a big mass, disregarding friendly damage due to auras & such.
-Has no scripts for specialized troops like fliers, cavalry or missile troops.
-Does not protect its mages & commanders.
-Cannot deal with ward+battlefield spell combos.
-Cannot deal with supercombatants.
-Does not bloodhunt.

I do think the tactical spell AI is competent though as I have said before, but a vet will likely script the 5 initial spells for maximum efficiency anyway.

Still, this is no different from any other 4X game in pc gaming industry, the more complex the game (and Dom is more complex than any of them), the harder it is for the AI to stand vs a competent player.

But that's why we have MP in Dom anyway, and seeing as I got 2 full years of enjoyable gaming from Dom I will gladly support IW by purchasing Dom II. I have great hopes in this game.

[ October 14, 2003, 11:02: Message edited by: Wendigo ]

johan osterman

October 14th, 2003, 02:02 PM

Originally posted by Mortifer:
Nice list about the AI weaknesses and I agree with those.
Lot of players are playing singleplayer mainly, so the AI must be upgraded. If the devs know that what was wrong with the AI, they can upgrade it. This list is valid, so these things should be fixed/updated.

I tell you something. If the mod tools will be out, you will be able to tweak the AI. How? Simply disable the tricky spells, what the AI cannot use properly. Than the players cannot trick the AI that much.

Anyways these issues with the AI should be fixed and than the AI will kick some ***. I am totally sure that the AI can be tweaked like that.
Just check the list and tweak/update the necessary parts of the AI. http://forum.shrapnelgames.com/images/icons/icon6.gif It is a little more complicated than that, for example making the strategic AI counter a player devised combo or supercombatant strategy is almost impossible since there are so many potentionally devastating combos and spells etc that are very powerful in special circumstances. Making an AI that adapts its strategy the way a human does in a game as complex as this is all but undoable, not only would the AI have to 'understand' the implications of the way spells items units interact but also 'understand' when a particular combo or spell etc was likely to appear etc, there is just to many factors to take into account. So making more than incremental improvements in the AIs ability to counter spell combos and supercombatants is very difficult.

Some but not all of the rest has been dealt with. The AI bloodhunts. Protecting commanders is not as important as it used to besince the attack commander/attack magic Users has been replaced with attack rear. Supercombatants might also be somewhat weaker due to changes in the strikeback effect and the removal of the attack commander orders.

[ October 14, 2003, 13:03: Message edited by: johan osterman ]

Pocus

October 14th, 2003, 02:10 PM

Originally posted by Mortifer:
snip : now that the problems are identified, its simple to fix them.
You are a bit too affirmative, in my humble opinion. I think you should try yourself to make a sizable software before saying that anything related to coding is simple.

Mortifer

October 14th, 2003, 02:16 PM

Originally posted by Pocus:
<blockquote>quote:<hr />
Originally posted by Mortifer:
snip : now that the problems are identified, its simple to fix them.
You are a bit too affirmative, in my humble opinion. I think you should try yourself to make a sizable software before saying that anything related to coding is simple.<hr /></blockquote>Aye, I know that it is not easy, but all of those parts can be upgraded with coding. In fact they should be. Well first lets wait for the game, and we can list that what should be upgraded in the AI mechanism.

Pocus

October 14th, 2003, 02:25 PM

Originally posted by johan osterman:
It is a little more complicated than that, for example making the strategic AI counter a player devised combo or supercombatant strategy is almost impossible since there are so many potentionally devastating combos and spells etc that are very powerful in special circumstances. Making an AI that adapts its strategy the way a human does in a game as complex as this is all but undoable, not only would the AI have to 'understand' the implications of the way spells items units interact but also 'understand' when a particular combo or spell etc was likely to appear etc, there is just to many factors to take into account. So making more than incremental improvements in the AIs ability to counter spell combos and supercombatants is very difficult.
true, but you should aim for proven recipes. Some games adapt their forces to the threat, but they generally does this after the threat appears, and not preemptively. For example, it is surely awfully complex to determine in advance if a player is heading toward having foul vapor / poison ward combos, or if he has the potential to field trampling gifted gargoyles with charcoal shields.

But a thing which is very doable is to tag each loss of the AI units with the origin of the loss. That is, the AI should store that it has losts so far 852 units to poison*, and 145 to trampling damages. These numbers can then be tweaked with a 'time distance', that is if the AI loose some 20 turns ago 150 units to poison, it should be less important than loosing these units just the Last turn. Having done that, you can sort the biggest threat, and have the AI focus on alleviating the problem. Solutions to poison can be to give a high priority to druid recruitments, a big incentive to search poison ward, a higher probability of having nature gems on nature mages, etc.
There is not that much differing sources of damages, perhaps 20 maximum. The biggest work is to have the AI tweak his priorities according to the threat represented by these 20 sources.

* : you can have a kill coming for several sources by the way, the system ought to be refined.

Thats just a remark on top of my head, but if you want to engage into serious ai programming, sites like gameai.com or ai-depot.com are must read. You would invest 50 hours of reading in doing so, but it is well worth the effort.

We all have the tendency to reinvent the wheel...

Wendigo

October 14th, 2003, 02:33 PM

Hopefully I didn't sound too negative. I consider the AI correct for what is to be expected from this kind of game, it has far too many variables to acount for & they interact in far too many different ways.

This is why the claims from the betatesters about a killer AI in Dom II came as a bit of a surprise, but hey, I am glad to hear that it has been tinkered with to appeal to the SP crowd: we can never have too many players & sooner or later those solo players will consider making the jump into MP for the enjoyment of us all.

And I should have included bloodhunts in the strategic section & not the tactical of course.

[ October 14, 2003, 13:34: Message edited by: Wendigo ]

Taqwus

October 14th, 2003, 02:41 PM

Mortifer --

It's not just coding that's hard. In fact, coding may be the easy part... It's the design of the algorithm, including formal description of the problem and how you decide what features et al need to be considered let alone what you do with them. If you can't identify what inputs need to be factored in, you can't even begin to code.

Nerfix

October 14th, 2003, 02:49 PM

Originally posted by Taqwus:
Mortifer --

It's not just coding that's hard. In fact, coding may be the easy part... It's the design of the algorithm, including formal description of the problem and how you decide what features et al need to be considered let alone what you do with them. If you can't identify what inputs need to be factored in, you can't even begin to code.You have personal experience about that?
You sound like you would have...

Pocus

October 14th, 2003, 02:59 PM

Originally posted by Taqwus:
Mortifer --

It's not just coding that's hard. In fact, coding may be the easy part... It's the design of the algorithm, including formal description of the problem and how you decide what features et al need to be considered let alone what you do with them. If you can't identify what inputs need to be factored in, you can't even begin to code.UML courses Taqwus? http://forum.shrapnelgames.com/images/icons/icon10.gif

Wendigo

October 14th, 2003, 03:07 PM

Originally posted by johan osterman:
[QB]The strategic AI can handle more situations in dom 2 than in dom 1, it is slightly better but it isnt Deep Blue. I hardly ever play single player TBS games so I do not know how it holds up against the AI's of other games.
QB]Well, we certainly do have a crowd of tbs game players that can offer opinions on this.

I do not play as many games as I used to in the past (in part due to RL being more demanding, and in part due to your game stubornly staying in my drive), but I can talk about the AI of a few classics.

MoM: Mom AI was very simple, and depended on bonuses mainly to stand vs the player. The tactical AI made a priority to target heroes, mages & missile units, but apart from that it was realtively simple to beat with inferior/specialized troops. The strategic AI just hyperexpanded relaying on its bonuses.

Kohan: Pausable RTS, so it basically plays as TBS.
The tactical AI retreats its units to heal them when badly damaged, but that's about it. When the AI wants a city it just sends anything it has vs it. AI Units are not tailor-designed to the opponent even on the highest lv as far as I can see, despite the design being far far simpler than Dominions. The strategic AI is correct.

Total War: A very annoying bonus of Shogun:TW was that the AI could 'see' your strategic move, and act upon its knowledge...this was extremely fustrating, you could have 2 defended provinces, empty one moving the army elsewhere & the AI would walk a few peasants from a nearby province & conquer it.
The tactical AI is extremely simple: just rush forward in offence (even walking into a killfield of arquebuses) without flanking maneuvers. In defense it sometimes stays put until targetted.
It's worth noting that missile light cavalry really performs as such in this game, with skirmish orders & keeping distance from the opponent.

All the above were notable games, yet none of them had a particularly challenging AI.

I wont comment on crap like Legion, Lords of Magic or similar, suffice to say that those barely survived a week in my HD.

Maelstorm

October 14th, 2003, 03:39 PM

Reading Johan Osterman's reply.....I had a feeling that the Dominions 2. AI will suck balls once again.

On the old shrapnel's Dom 2. site there was a sentence that Doms 2. will have a very good AI.
On the new site this info cannot be found.

I think that Shrapnel realized that the AI will suck, so they removed that sentence, because lying is not wise for a publisher. http://forum.shrapnelgames.com/images/icons/shock.gif

I tell you, that one of the most imprtant thing is the AI in a strategy game, if the game is not a MMORPG. You should think about this.

Truper

October 14th, 2003, 03:51 PM

Hey Johan,

How about a new spell in Maelstorm's (sic) honor?
We could call it Legions of the Clueless...

Zerger

October 14th, 2003, 04:04 PM

johan osterman

October 14th, 2003, 04:09 PM

Originally posted by Maelstorm:
Reading Johan Osterman's reply.....I had a feeling that the Dominions 2. AI will suck balls once again.

On the old shrapnel's Dom 2. site there was a sentence that Doms 2. will have a very good AI.
On the new site this info cannot be found.

I think that Shrapnel realized that the AI will suck, so they removed that sentence, because lying is not wise for a publisher. http://forum.shrapnelgames.com/images/icons/shock.gif

I tell you, that one of the most imprtant thing is the AI in a strategy game, if the game is not a MMORPG. You should think about this.Your colorful desciptions aside I wonder what paragon of AI programming you are comparing the dom 1 AI with? As for as what Shrapnel wrote about the AI I guess it is what they honestly felt. The only TBS I have played single player lately is the TW games, and as far as I am concerned they were not very impessive, and both the total war games have a lot less troops, special abilities and other factors to take into account than dominions. I have never played a strategy game of any kind except chess and checkers where the AI can holds it own on even footing with a competent human. It would also be both more helpful and interesting to hear you voice your specific complaints against the AI rather than the crap you are currently posting.

As for your statement that the AI is most important fact in a strategy game unless it is a MMORPG, dominions is designed from the bottom up to be a multi player game, primarily for network or PBEM play, but I do think the single player experience holds its own as well.

[ October 14, 2003, 15:38: Message edited by: johan osterman ]

Nerfix

October 14th, 2003, 04:09 PM

licker

October 14th, 2003, 04:44 PM

Yes yes, we all know that Dom is designed first and formost from a MP point of view. We also understand the complexities involved in creating a 'suitable' AI (suitable is in the eye of the beholder of course) for a game with as many complex systems as Dom has.

However, one thing that Maelstrom said has at least some merit, the majority of play on Dom will be in SP mode, to ignore that market (not that Illwinter is ignoring it) would be sheer folly. If there are asperations for great sales of Dom2 then the SP game needs to be reinforced as being terrific. Look for all any of us knows it is, and hopefully once the community gets their hands on the game there will be alot of new ideas and discussions about the AI.

I will remain hopeful that Illwinter remains as faithful to their fan base as they've been with Dom1, and when and where they can they will make improvements to the AI to further improve the SP experience.

A Last resort is to move to heavilly scripted maps to give a challenge to SPers, however, thats not as good of a result as it is to be able to make continued improvements to the Dom AI.

Saber Cherry

October 14th, 2003, 05:52 PM

Sorry folks, I got a copy of DOM2.exe from a betatester, ran a disassembler on it, and managed to recreate the logical class structure. The results look bad. I'll post the source code of the AI class here:

/*Dominions 2 AI code*/

Class Dom2AI{

//primary AI command
int suck(MonkeyBalls *MB){

//Remember to fill this in by patch 1.3

return 2;
}
}

Well, there you have it. I don't want to draw any premature conclusions, but the only Dominions 2 AI function seems to be... shorter than I had hoped. Oh well...

-Cherry

P.S. I think the DOMI AI is pretty good. And for those of you who are extra sensitive today, the above is intended to be humorous http://forum.shrapnelgames.com/images/icons/tongue.gif

[ October 14, 2003, 17:17: Message edited by: Saber Cherry ]

Nerfix

October 14th, 2003, 06:08 PM

ROTFLMAO!

I didn't know that function "suck monkey balls" does miracles...

Ok, but we still don't have a real answer about the targeting efficency...

Zerger

October 14th, 2003, 06:11 PM

Originally posted by Saber Cherry:
Sorry folks, I got a copy of DOM2.exe from a betatester, ran a disassembler on it, and managed to recreate the logical class structure. The results look bad. I'll post the source code of the AI class here:

/*Dominions 2 AI code*/

Class Dom2AI{

//primary AI command
int suck(MonkeyBalls *MB){

//Remember to fill this in by patch 1.3

return 2;
}
}

Well, there you have it. I don't want to draw any premature conclusions, but the only Dominions 2 AI function seems to be... shorter than I had hoped. Oh well...

-CherryYou should stop posting your uber BS. You are making no sense usually. http://forum.shrapnelgames.com/images/icons/icon8.gif

I have a question: What is this histeria about the Dominions 2. AI? We dont even played with the game! http://forum.shrapnelgames.com/images/icons/icon8.gif

[ October 14, 2003, 17:12: Message edited by: Zerger ]

October 14th, 2003, 06:27 PM

We can say this or that about TW AI, but as far as the AIs go, it is good, maybe not at its brightest as the strategical one, but the tactical one is. Of course you can beat it, but you can fully control your own units there, so it's not really comparable to the game where you cannot do that. Even with tactical control, if you mod the game so that the strategical AI does not build droves of peasants, but rather some quality troops instead, it puts up a very decent challenge on hard and expert. Try autoresolving in MTW every time, and you'll see how you'll fare.

And why would you leave an undefended province when attacking anyway? http://forum.shrapnelgames.com/images/icons/icon12.gif

Nerfix

October 14th, 2003, 06:39 PM

Zerger:
That is humor.
Now it has been proved that internet humor NEEDS the http://forum.shrapnelgames.com/images/icons/tongue.gif or http://forum.shrapnelgames.com/images/icons/icon10.gif smiley to be understood as humor or a joke. Sarcasm needs the [/sarcasm] tag to be understood as sarcasm.

Note:Even if this post has the http://forum.shrapnelgames.com/images/icons/icon10.gif and http://forum.shrapnelgames.com/images/icons/tongue.gif smileys and [/sarcasm] tag, this isn't humor, sarcasm, or a joke.

[ October 14, 2003, 17:39: Message edited by: Nerfix ]

Wendigo

October 14th, 2003, 06:57 PM

Originally posted by HJ:
[QB]We can say this or that about TW AI, but as far as the AIs go, it is good, maybe not at its brightest as the strategical one, but the tactical one is.
I disagree, the AI will charge with its heavy cavalry towards your pikemen formation, while heavy punished by missile fire & while exposing its flank to your own heavy cavalry...that doesn't strike me as bright. More often than not the opposing commander will also be leading that Heavy cavalry unit instead of staying in the back.

Of course you can beat it, but you can fully control your own units there, so it's not really comparable to the game where you cannot do that.
I consider Dom's scripting powerful enough to not really miss the full control of other games, you basically replace battlefield maneuvering with mage power & special units. If any, with more variables to account for I consider Dom AI more difficult to program, with TW it's just a matter of 'do not fight X with Y in terrain Z' (and it doesn't even do that).

Try autoresolving in MTW every time, and you'll see how you'll fare.
But then, if you autoresolve you have no tactical AI at all.

And why would you leave an undefended province when attacking anyway? http://forum.shrapnelgames.com/images/icons/icon12.gif Maybe you did not understand my (hypothetic) example: the turns are supposed to be simultaneous but they are not, when playing its turn in the strategic map in Shogun:TW the AI does so with the knowledge of the orders you just issued, so it has full knowledge of where your armies are moving.

Don't get me wrong, I liked the TW games & consider them revolutionary, but that was certainly not because of the AI, but because of the great atmosphear, the powerful tactical engine that allows for all those battlefield maneuvers with huge armies, the historical settings...

[ October 14, 2003, 17:59: Message edited by: Wendigo ]

Taqwus

October 14th, 2003, 07:00 PM

Heh. Yeah, I've done a fair bit of coding, and wandered into the machine-learning material a bit. Haven't done much AI-ish stuff since TA'ing a course for it (heh; the students got one eccentric final that semester, as my odd sense of humor affected much of it...).
Strong AI doesn't exist yet. What one can get now is some degree of function-fitting; but that presumes you know the structure of your inputs and outputs. The much-hyped neural network, for instance, needs to be given a set of inputs, and needs to be told what output to learn. Genetic programming methods need a vocabulary of operators, plus inputs to operate on them. Pick an input set that doesn't work, or don't include the flexibility necessary to fit your output, and it'll fail; have too much, and perhaps your search will trend towards getting stuck in meaningless local minima. It's not a coincidence that many games prefer to have fixed scenarios with pre-written scripts for AIs that also get vastly superior starting configurations... and most games have far fewer tactical choices than Dominions.

October 14th, 2003, 09:16 PM

Originally posted by Wendigo:
I disagree, the AI will charge with its heavy cavalry towards your pikemen formation, while heavy punished by missile fire & while exposing its flank to your own heavy cavalry...that doesn't strike me as bright. More often than not the opposing commander will also be leading that Heavy cavalry unit instead of staying in the back. I said "as far as the AIs go", not when compared to Napoleon or Alexander the Great. Very few games have anything comparable to TW AI. The AI is exploitable, for sure, but it can perform really well. If you see another game with the AI that can perform flanking and combined arms attacks, let me know.

I consider Dom's scripting powerful enough to not really miss the full control of other games, you basically replace battlefield maneuvering with mage power & special units. If any, with more variables to account for I consider Dom AI more difficult to program, with TW it's just a matter of 'do not fight X with Y in terrain Z' (and it doesn't even do that). And all the calculations are done in real time. (I'm not a RTS fan, I'm just saying why it isn't comparable). And what other variables might those be, btw? I mean, since there is no terrain, no formations, no charge bonuses, no accurate targeting? As I said, if given tactical control over the battles, then we would be able to somewhat compare the two. This way, it's like letting AI control your units in TW. I don't know about you, but I would be pulling my (remaining) hair out. And I do that in Dominions as well.

But then, if you autoresolve you have no tactical AI at all. Just comparing the numbers, right? Nothing you can do about it? You just see the end result? True

Maybe you did not understand my (hypothetic) example: the turns are supposed to be simultaneous but they are not, when playing its turn in the strategic map in Shogun:TW the AI does so with the knowledge of the orders you just issued, so it has full knowledge of where your armies are moving.

Don't get me wrong, I liked the TW games & consider them revolutionary, but that was certainly not because of the AI, but because of the great atmosphear, the powerful tactical engine that allows for all those battlefield maneuvers with huge armies, the historical settings...They are not supposed to be simultaneous, and they are not. You take this into acount when making moves that it's IGO-UGO (sort of), and not simultaneous.

I also liked them because of all those things, but I also find them to put up a decent challenge. More so than vast majority of other games. Since you're talking about Shogun, may I suggest trying out MTW if you can find it on sale (I guess it's really cheap now) if you haven't already? Although it lost a good deal of atmosphere that STW had, some other things are greatly improved nevertheless.

Maelstorm

October 14th, 2003, 10:26 PM

Originally posted by Taqwus:
Heh. Yeah, I've done a fair bit of coding, and wandered into the machine-learning material a bit. Haven't done much AI-ish stuff since TA'ing a course for it (heh; the students got one eccentric final that semester, as my odd sense of humor affected much of it...).
Strong AI doesn't exist yet. What one can get now is some degree of function-fitting; but that presumes you know the structure of your inputs and outputs. The much-hyped neural network, for instance, needs to be given a set of inputs, and needs to be told what output to learn. Genetic programming methods need a vocabulary of operators, plus inputs to operate on them. Pick an input set that doesn't work, or don't include the flexibility necessary to fit your output, and it'll fail; have too much, and perhaps your search will trend towards getting stuck in meaningless local minima. It's not a coincidence that many games prefer to have fixed scenarios with pre-written scripts for AIs that also get vastly superior starting configurations... and most games have far fewer tactical choices than Dominions.You are a fool. You have no idea about scripting.

Psitticine

October 14th, 2003, 10:47 PM

Maelstrom, that post is seriously over the line. This is not the place for personal attacks. Please keep things civil from here on out.

Endoperez

October 14th, 2003, 10:49 PM

Originally posted by Maelstorm:
<blockquote>quote:<hr />Originally posted by Taqwus:
Strong AI doesn't exist yet. What one can get now is some degree of function-fitting; but that presumes you know the structure of your inputs and outputs. The much-hyped neural network, for instance, needs to be given a set of inputs, and needs to be told what output to learn. Genetic programming methods need a vocabulary of operators, plus inputs to operate on them. Pick an input set that doesn't work, or don't include the flexibility necessary to fit your output, and it'll fail; have too much, and perhaps your search will trend towards getting stuck in meaningless local minima. It's not a coincidence that many games prefer to have fixed scenarios with pre-written scripts for AIs that also get vastly superior starting configurations... and most games have far fewer tactical choices than Dominions.You are a fool. You have no idea about scripting.<hr /></blockquote>He didn't say that much about scripting... In fact, because his Posts are much more imformative than yours, I think that he knows quite a bit more about scripting than you. I am not commenting on that 'Evolving AI' -stuff, mainly because I am not sure did I understand it. But if you meant THAT as a scripting he has no idea about, I quess that you are a total fool. I would like to say troll, but I'm afraid they would come to eat me if they heard me say so...

[ October 14, 2003, 21:52: Message edited by: Endoperez ]

johan osterman

October 14th, 2003, 10:57 PM

The only acceptable way of calling another man fool is in the context of a Mr.T reference. For example: I pity the fool that thinks the dominions AI can be improved.

Jasper

October 14th, 2003, 11:16 PM

Originally posted by Maelstorm:
<blockquote>quote:<hr />Reasoned post by Taqwus snipped.
You are a fool. You have no idea about scripting.[/qb]<hr /></blockquote>Actually he gave a concise and accurate report of the current state of AI, and clearly has some experience with it. On the other hand, you have giving nothing to support your purile view.

[ October 14, 2003, 22:18: Message edited by: Jasper ]

Aristoteles

October 14th, 2003, 11:33 PM

Originally posted by johan osterman:
For example: I pity the fool that thinks the dominions AI can be improved.Huh??? http://forum.shrapnelgames.com/images/icons/confused.gif http://forum.shrapnelgames.com/images/icons/shock.gif

Psitticine

October 15th, 2003, 12:33 AM

Originally posted by Aristoteles:
<blockquote>quote:<hr />Originally posted by johan osterman:
For example: I pity the fool that thinks the dominions AI can be improved.Huh??? http://forum.shrapnelgames.com/images/icons/confused.gif http://forum.shrapnelgames.com/images/icons/shock.gif <hr /></blockquote>That is, if I'm not mistaken, a rather amusing Mr. T reference. http://forum.shrapnelgames.com/images/icons/icon7.gif

Potatoman

October 15th, 2003, 12:55 AM

For the T ignorant among you:
Mr. T vs Everything (http://www.mrtvseverything.com)

Enough of this Jibba Jabba! All you forum foo's are crazier than Murdoch!

Mortifer

October 15th, 2003, 01:19 AM

Originally posted by Wendigo:
Indeed, to follow up on what Pocus & others point out:

The Dom I AI cannot stand vs a vet player either strategically nor tactically:

Strategically:
-It doesn't build the right unit-mix to face its opponent (with lack of priest power when facing Ermor being the most notable failure).
-It disregards supply & usually fields starving hordes.
-It does not patrol, nor builds local defense so it is totally open to sneak attacks.
-Cannot equip supercombatants & mages in a competitive way.

Tactically:
-It just bunches its troops in a big mass, disregarding friendly damage due to auras & such.
-Has no scripts for specialized troops like fliers, cavalry or missile troops.
-Does not protect its mages & commanders.
-Cannot deal with ward+battlefield spell combos.
-Cannot deal with supercombatants.
-Does not bloodhunt.

I do think the tactical spell AI is competent though as I have said before, but a vet will likely script the 5 initial spells for maximum efficiency anyway.

Still, this is no different from any other 4X game in pc gaming industry, the more complex the game (and Dom is more complex than any of them), the harder it is for the AI to stand vs a competent player.

But that's why we have MP in Dom anyway, and seeing as I got 2 full years of enjoyable gaming from Dom I will gladly support IW by purchasing Dom II. I have great hopes in this game.Nice list about the AI weaknesses and I agree with those.
Lot of players are playing singleplayer mainly, so the AI must be upgraded. If the devs know that what was wrong with the AI, they can upgrade it. This list is valid, so these things should be fixed/updated.

I tell you something. If the mod tools will be out, you will be able to tweak the AI. How? Simply disable the tricky spells, what the AI cannot use properly. Than the players cannot trick the AI that much.

Anyways these issues with the AI should be fixed and than the AI will kick some ***. I am totally sure that the AI can be tweaked like that.
Just check the list and tweak/update the necessary parts of the AI. http://forum.shrapnelgames.com/images/icons/icon6.gif

LordArioch

October 15th, 2003, 02:34 AM

Wow...a thread with programming humor from saber cherry and a mister T reference!

Honestly I think the Dominions 1 AI was pretty good and difficult for anyone who wasn't a master player. The main things I noticed human players have over it are advantages due to things dominions 2 is fixing...namely elemental abuse and patrolling.

Tatical AI was good too...it's just everyone notices it's failings because it's controlling their units as well as the enemies. http://forum.shrapnelgames.com/images/icons/icon7.gif Although an improvement in not shooting the one militia in the middle of your army would be nice...but it sounds like that's been worked on.

Taqwus

October 15th, 2003, 03:43 AM

*shrug*

And as for pre-game scripting, that mostly works when the situation tends to be fairly predictable.

For instance, RTSes generally have far more limited possibilities. "Tech trees" or their equivalents tend to be small, unit choices can be constrained by lack of resources or even artificial limits (e.g. "you're not allowed to build that building this level"), and so forth. In addition, many of the popular ones such as the *craft series start the player with a very, very small force and the AI with a significant base already built up -- and having a pre-built base limits what the AI needs to consider. Then, it may have scenario-specific instructions, e.g. in a "survive for XX minutes" scenario the AI may be told to attack in waves of certain sizes or at a given time.

Taken to the extreme, a strategy game gets turned into a puzzle game; the AI isn't so much factor as the decisions that the level designer made before the game was ever started.

Dominions is far, far more varied. The number of units is huge; combinations of units can get pretty strange courtesy of the independents and the charming/enslaving spells; and the magical spells and items can significantly alter things. Even if one had decent strategic scripts written beforehand, tactics and events may quickly render them meaningless... because you can't prepare optimally for everything. One might face an early war with cheap units; one might have to deal with hordes of mindless undead; one might face assassination, or magical assaults, or disease warfare. Perhaps an enemy has a heavily-decked out combat leader; perhaps he has a rainbow mage. Maybe he's going for weak hordes, maybe stronger elite units. Perhaps he's bringing bows, or perhaps he's invoking storms again to limit bows. Maybe an army's planning to siege you out; maybe that castle the AI wants to siege has vast numbers of ghouls in it so sieging isn't too practical. A human player brings out the Ark and blinds half your army; how does that change things? Or he's got an immortal commander casting Summon Lammashastas, or summoning other nasties and then magically leaving the battle?

Is it the mage that's the threat, or would it be easier to take out the communicants? Or is one of the mage's constructs or some tough combatant a bigger threat? Send units to fight the toughie with the damage shield and wraith sword, or merely try to hold him off and send the bulk against the rest of the enemy? Heck, even deciding whether to burn gems can be tricky, when you're attacking an enemy province without a lab so you might be caught short in a counterattack.

You've got fliers, and the enemy has a strong flier. Try to ground everyone? How to decide? Ditto for bowmen, et al. Super combatant versus super? Are those militia advancing numerous enough to merit attention, or no? The enemy's using mindless units; fight them, or find a way to kill the leaders?

It's an enormously complicated game, and it doesn't have the advantage of drastically constraining the problem space. In addition, hand-eye coordination doesn't matter, so the game can't rely on old stand-byes like insanely good speed-of-light reflexes ala AI Paladins in WC2 healing each other constantly during battle. And learning approaches will be hard, too; even saying something like "learn from what just happened" is difficult, because it needs to grok "why". And that why may be pretty subtle, or go back a considerable number of turns, or involve a diverse set of factors ranging from research to greater gem supply to even dumb luck -- e.g. getting lucky and killing an enemy supercombatant when it botches its MR check, or simply getting outlandish results from open-ended dice. Factor in hidden information and the large number of players involved, and it's a bit surprising that it can do much at all. There's so much stuff that can happen that planning can't be easy, nor would learning.

DominionsFan

October 15th, 2003, 09:40 AM

Taqwus, why the heck are you writing stories? LOL. http://forum.shrapnelgames.com/images/icons/icon10.gif

Can't you complain in 1 sentence?!
So in your opinion the Dominions I. AI was good enough?
Please reply with a yes/no. http://forum.shrapnelgames.com/images/smilies/rolleyes.gif

[ October 15, 2003, 08:40: Message edited by: DominionsFAN ]

Wendigo

October 15th, 2003, 12:37 PM

Originally posted by HJ:
I said "as far as the AIs go", not when compared to Napoleon or Alexander the Great. Very few games have anything comparable to TW AI. The AI is exploitable, for sure, but it can perform really well. If you see another game with the AI that can perform flanking and combined arms attacks, let me know.
This is strange, it sounds as if we were talking of different games. I have never seen the TW AI perform a flanking maneuver (although it indeed answers your own flanking atempts by rearanging its army formation). In my experience (in the highest lv in STW, and one of the higher if not the highest in MTW) the AI will just charge forward its melee units to engage the nearest enemy unit & fire with skirmish or hold orders with its missile units. That can hardly be considered 'combined arms attacks', when it basically ignores its own rock-scissor-paper rules in its offensive.

Frankly, it's anything but bright. If it tried to at least engage your cavalry with its spears, maneuver for a flank charge, try to gain the higher ground...but it doesn't do anything like this. Its only notable doings are the skirmish script for missile units and the army formations that at least keep some order before they break & the mounted troops rush forward leaving the infantry behind.

And all the calculations are done in real time. (I'm not a RTS fan, I'm just saying why it isn't comparable). And what other variables might those be, btw? I mean, since there is no terrain, no formations, no charge bonuses, no accurate targeting?.
Some other poster expanded on this, but basically in my little knowledge of AI scripting I find it easier to write some glorified IF-THEN conditionals to account for:
a couple dozen units x 3 different facings x 3 or so different terrains x higher/lower ground x a handful of different formations in order to decide whether to charge, fall back or maneuver for a better postion. Maybe add a couple more conditionals for morale & experience.

In Dominions however said conditionals would have to acount for _many hundred units_ ^ modified by many hundred spells (note that multiple spells can affect the same unit, thus we have an exponitial increase in posibilities here)^ magic items x morale, experience, afflictions, HoF bonuses, dominion bonuses, starvation.... see the difference?

Even with TW being RT handling a few thousand triggers (or maybe only a few hundred, as units can be grouped into similar types that would act the same 90% of the time) should be doable for any modern computer.

They are not supposed to be simultaneous, and they are not. You take this into acount when making moves that it's IGO-UGO (sort of), and not simultaneous.
Just curious, is this a guess or a deduction you made from MP experience? And what game does it refers to? I ask so because this issue raised many complaints with STW, but I do not recall the same feeling from MTW, so I have to wonder if it was changed.

I also liked them because of all those things, but I also find them to put up a decent challenge. More so than vast majority of other games. Since you're talking about Shogun, may I suggest trying out MTW if you can find it on sale
Thks for the sugestion, already got it. While I enjoyed both installments I guess I must differ regarding the challenge, for the reasons stated above.

[ October 15, 2003, 11:39: Message edited by: Wendigo ]

DominionsFan

October 15th, 2003, 02:05 PM

Well, I think it is pointless to post about the AI right now, let us wait for the demo first.
We can complain than, if we want. http://forum.shrapnelgames.com/images/smilies/rolleyes.gif

PDF

October 15th, 2003, 02:09 PM

Originally posted by Maelstorm:
<blockquote>quote:<hr />Originally posted by Taqwus:
Heh. Yeah, I've done a fair bit of coding, and wandered into the machine-learning material a bit. Haven't done much AI-ish stuff since TA'ing a course for it (heh; the students got one eccentric final that semester, as my odd sense of humor affected much of it...).
Strong AI doesn't exist yet. What one can get now is some degree of function-fitting; but that presumes you know the structure of your inputs and outputs. The much-hyped neural network, for instance, needs to be given a set of inputs, and needs to be told what output to learn. Genetic programming methods need a vocabulary of operators, plus inputs to operate on them. Pick an input set that doesn't work, or don't include the flexibility necessary to fit your output, and it'll fail; have too much, and perhaps your search will trend towards getting stuck in meaningless local minima. It's not a coincidence that many games prefer to have fixed scenarios with pre-written scripts for AIs that also get vastly superior starting configurations... and most games have far fewer tactical choices than Dominions.You are a fool. You have no idea about scripting.<hr /></blockquote>Hey Grand Master Maelstrom, please enlighten us with your brightetest ideas about AI and scripting, rather ! http://forum.shrapnelgames.com/images/icons/tongue.gif
http://forum.shrapnelgames.com/images/smilies/rolleyes.gif http://forum.shrapnelgames.com/images/smilies/rolleyes.gif

DominionsFan

October 15th, 2003, 02:12 PM

Originally posted by PDF:
Hey Grand Master Maelstromlol. http://forum.shrapnelgames.com/images/icons/icon10.gif

Taqwus

October 15th, 2003, 04:26 PM

DominionsFan --

Heh. Well, I suppose that depends.

For instance, going back to the original thread target, it was possible to use weak cheap units as "chaff" to distract its missile firers. Not being a beta-tester, I don't know whether this has been improved. Mind you, the missile-firers may have a different interest than their general -- what if the "chaff" isn't the main threat in the battle, but it threatens the missile-firers themselves?

Magical targeting is trickier, for balance reasons. Consider the "save-or-die" spells, such as Disintegrate. Imagine Illwinter generated an excellent evaluation function -- or even let players script their own -- that compared enemy units based on proximity, whether or not they could fly given weather conditions, their hp, att, prot etc. Suppose it could identify, perfectly, the optimal threat. Would it be reasonable to allow these death spells to always target the nastiest threat in range, or would that lead to magic-dominated gunfights (mages on both sides slaying each other, then the surviving mages pick off supercombatants, then killing off the remaining commanders...) ?

There can be /too/ good targeting, I'd say; both from a balance perspective, and also from a "realism" perspective e.g. could a mage really, during a melee involving perhaps hundreds or more units, pick out the main threats so accurately? Outrageous cheese -- truly silly uses of chaff, say -- shouldn't be possible, but there aren't many units that can justify supreme targeting logic either. Say, flying units might have an edge on this since they could see more of the battle. If Illwinter wanted to implement "does the commander notice this" sort of logic, they could probably also factor in unit experience.

Richard

October 16th, 2003, 01:50 AM

October 16th, 2003, 06:56 PM

Originally posted by Wendigo:
[QB] This is strange, it sounds as if we were talking of different games. I have never seen the TW AI perform a flanking maneuver (although it indeed answers your own flanking atempts by rearanging its army formation). In my experience (in the highest lv in STW, and one of the higher if not the highest in MTW) the AI will just charge forward its melee units to engage the nearest enemy unit & fire with skirmish or hold orders with its missile units. That can hardly be considered 'combined arms attacks', when it basically ignores its own rock-scissor-paper rules in its offensive.

Frankly, it's anything but bright. If it tried to at least engage your cavalry with its spears, maneuver for a flank charge, try to gain the higher ground...but it doesn't do anything like this. Its only notable doings are the skirmish script for missile units and the army formations that at least keep some order before they break & the mounted troops rush forward leaving the infantry behind. Well, it would seem as if we do talk about different games. http://forum.shrapnelgames.com/images/icons/confused.gif

Some other poster expanded on this, but basically in my little knowledge of AI scripting I find it easier to write some glorified IF-THEN conditionals to account for:
a couple dozen units x 3 different facings x 3 or so different terrains x higher/lower ground x a handful of different formations in order to decide whether to charge, fall back or maneuver for a better postion. Maybe add a couple more conditionals for morale & experience.

In Dominions however said conditionals would have to acount for _many hundred units_ ^ modified by many hundred spells (note that multiple spells can affect the same unit, thus we have an exponitial increase in posibilities here)^ magic items x morale, experience, afflictions, HoF bonuses, dominion bonuses, starvation.... see the difference?

Even with TW being RT handling a few thousand triggers (or maybe only a few hundred, as units can be grouped into similar types that would act the same 90% of the time) should be doable for any modern computer. The TW system is actually quite complex when you get to know it, especially the morale system. It superficially seems as if there are only a few variables, but this is not the case.

Just curious, is this a guess or a deduction you made from MP experience? And what game does it refers to? I ask so because this issue raised many complaints with STW, but I do not recall the same feeling from MTW, so I have to wonder if it was changed. I don't play MP, so my judgements are based solely on SP (why else would we be talking about the AI)?

Thks for the sugestion, already got it. While I enjoyed both installments I guess I must differ regarding the challenge, for the reasons stated above.Things have changed for the better quite a bit ever since MTW 1.0, and the AI is quite capable for an AI now. Doms AI never gave me the same run for the money, on the other hand, even when I was playing for the first time. So yes, I guess we differ on this, probably because of the different time spent on playing the games, and hence the ability to get a good idea on the AI in the first place.

[ October 16, 2003, 17:58: Message edited by: HJ ]

Wendigo

October 16th, 2003, 07:20 PM

Originally posted by HJ:
Things have changed for the better quite a bit ever since MTW 1.0, and the AI is quite capable for an AI now.
I will have to check on whether my MTW copy is upgraded to this patch or not then, and give it a shot if not.

Maybe our particular points of view are coloured by me having spent far more time with STW, and you having spent far more time with MTW 1.0.

I cannot for example forget that in STW you could basically win without taking a province: The enemy Daymos (sp?) would charge at you singlehandely ahead of their army, resulting in their death, the routing of their army afterwards (that more often than not outnumebred yours 3to1), and the destruction of their factions because of lack of heirs...this seemed sooo moronic.