AI in Heroes of Might and Magic 5 - Nival principles

[MistWeaver]

Logistics has no prerequisites.

No,skills dont have prerequests,but feats do.Rarelly youll see an AI with a feat.

Logistics basic feats (navigation, pathfinding, scouting) has no prerequests too. They can be taken anytime after youll have logistics itself.

[DaemianLucifer]

Thats a prerequest,isnt it?You need to have logistics to take any of them.

[MistWeaver]

Thats a prerequest,isnt it?You need to have logistics to take any of them.

Wait a minute, you said:

The decision is harder because HIII had no prerequests to think about.Its a small change,yes,but apparently big one for the AI to swallow.

So why its harder ? H5 AI must not think about this prerequests.
It just have to deside to take Logistics itself or not, like in H3.

You see, AI will not think this way "Do I need Pathfinding in this game ? If so I must take Logistics first"
Instead it probably will "Do I need Logistics and its stuff in this game ? If so Ill take it once Ill have a chance"
And feats comes after that on its own.

[Gaidal Cain]

But seriously,if I knew where the values for the AI are stashed Id easily make it do some smarter things.

p#-data.pak\GameMechanics\RPGstats\DefaulStats.xdb

Only stuff like how much it prioritizes towns and stuff though. No skills/abilities.

[Jolly Joker]

H 1 clearly has the toughest AI (cheats like hell, though), simply because it is the easiest game. It's even easy to program competent spell-casting, because there's only room for one spell and each successor will dispell a predecessor, so if you cast make sure your spell has effect immediately. No waiting. Not many adventure objects, not much town building. In combination with the cheats, no prob.
With H II, basically the hero skills are added - plus a more complex spell system and town building process with the upgrades. The AI is still good, but not that good - you'll much too often take a town with a dozen Titans in it, for example.
H III AI already sucks. Combat AI is a lot worse than their predecessors, mainly due to 3 facts: 1) inability to utilize wait command; 2) inability to pick the good skills and develop competetive heroes. 3) AI attacks weakest unit most of the time/too easy to lure AI into doing that. The adventure AI "looks right", but it still doesn't know what to do. It has a lot of resources as well.
H IV had no AI; the reason for this is mainly that the battle AI couldn't cope with what is necessary. It can't exploit spells to beat vastly superior numbers - which is a must on expert and impossible. So the adventure AI is rather helpless most of the time.
H V has the best battle AI by far if you take the complexity of battles into consideration. Sure, you sometimes see Horned Demons exploding without anyone near - but strangely enough, when that happens those Demons wouldn't have done any damage in that battle anyway. Another reason is that there are no bad skills, so an enemy hero level 17 will as a rule be tough, has decent skills and will use them. You can have a fun time just picking Multiplayer/Hot Seat/Duel Mode, pick two heroes and leave one of them the AI to play. The same i true for Quick Combat: It's decent, very decent and you CAN actually play with it.
Now, the adventure AI. On hard and heroic the AI gets a resource advantage that is pretty normal. When you play heroic I don't think you can argue against the AI starting with the Easy resources. If that would be all the AI gets it would still be more than enough to field more in terms of builds and creatures up to a certain point. It will try to win the game in a time where it has the material. If that doesn't work, the AI may turtle in it's town from a certain point, at least it looks that way in some or even many cases. This is the smart decision, and it is the decision that will make it the most difficult to conquer the AI: attacking the town. The humans play the same way, or don't they? You keep in the vicinity of your town to be able to defend it against surprise attacks and to be able to field the weight of the walls and the towers to defend against an enemy being superior.
All in all, for a game like this the AI is rather decent. I'd rate it on par with H II - considering the fact that H V is a bit more complex.
Sure you can always make something better, but I think in H V, if you play a simple MP map you will have an easy play on easy, a "normal" (easy for non-novices) game on normal, a very interesting, albeit not too tough game on hard and a seriously hard game on heroic. Moreover, I think that the starting amounts in H V are ALOT better chosen than all those in the previous games, because the fact that you start with SOMETHING doesn't make a map depend on what the map maker placed where in the vicinity of your town. You have enough to never have to reload necause you missed the dearly needed chest by going into the wrong direction and so on.
Now, all that doesn't mean the AI is perfect or something, that would be a ridiculous claim. But I think, they simply made a lot of things very right.

[Pitsu]

H IV had no AI; the reason for this is mainly that the battle AI couldn't cope with what is necessary. It can't exploit spells to beat vastly superior numbers - which is a must on expert and impossible. So the adventure AI is rather helpless most of the time.
H V has the best battle AI by far if you take the complexity of battles into consideration.

Have to disagree here. I think it was Gus who wrote the tactical H4 AI and commented it on the old Round Table and he was pretty proud of it. In fact The H4 combat AI is indeed OK except siege and handling some last-minute changes. It was adventure AI that suffered most due to premature release. Remember that AI fights against neutral most likely use only a quick combat check that has little to do with actual AI combat capability. And comparing the complexity of battles: see the combat grid and possible combinations of unit placements, amount of spells, number of units and their specials (in H4 heroes are also units on battlefield). How can you say that HV has the best combat AI considering the complexity?

[DaemianLucifer]

You see, AI will not think this way "Do I need Pathfinding in this game ? If so I must take Logistics first"
Instead it probably will "Do I need Logistics and its stuff in this game ? If so Ill take it once Ill have a chance"
And feats comes after that on its own.

And you saw it with feats,when?Not counting the premade ones,of course.

Oh,and HIV combat AI is nice in open combat.And in equi it is even better(excelent that is).

[Jolly Joker]

H IV had no AI; the reason for this is mainly that the battle AI couldn't cope with what is necessary. It can't exploit spells to beat vastly superior numbers - which is a must on expert and impossible. So the adventure AI is rather helpless most of the time.
H V has the best battle AI by far if you take the complexity of battles into consideration.

Have to disagree here. I think it was Gus who wrote the tactical H4 AI and commented it on the old Round Table and he was pretty proud of it. In fact The H4 combat AI is indeed OK except siege and handling some last-minute changes. It was adventure AI that suffered most due to premature release. Remember that AI fights against neutral most likely use only a quick combat check that has little to do with actual AI combat capability. And comparing the complexity of battles: see the combat grid and possible combinations of unit placements, amount of spells, number of units and their specials (in H4 heroes are also units on battlefield). How can you say that HV has the best combat AI considering the complexity?

About H IV, we all know that quick combat was the death of the AI (and compare that with H V which is lightyears ahead). However, the tactical AI wasn't that good either. The main flaw is that it targets illusions, summons and raises instead of the real thing or even the heroes most of the time or at least much too often. We don't even want to start talking about the use (or better not-use) of potions, spell-casting and so on. H IV has a potentially very complex battle interaction of creatures and spells and the AI is not up to it. Another point is, that the AI isn't going to get the chance to battle on equal terms often in AI (we are talking about simple MP maps here). The only point I agree with is, that considering the complexity of the game the battle AI (meaning the AI you face when fighting an AI player army) was clearly something you could live with.
About how I can say that H V clearly has the best battle AI considering the complexity, I think this is rather obvious. Battlefield size as such is no measure of complexity. Instead it reduces complexity from a certain point. I think, in H V , in a real battle the odds favor acting (and risking getting damaged) instead of waiting in the majority of cases. It's the other way round in H III and H IV (where your main goal is finding ways so that your "real" units don't suffer damage). The game does this by making the battlefield smaller in terms of the unit speed/battlefield size ratio (except in Sieges) and giving a lot of units rather interesting abilities with probability triggers. You may get overrun: Nightmares attack: fear -> no retal -> good moral plus high init -> early next turn -> aura of fear leading to moral failures of adjacent units and suddenly the waiter is in trouble (I hope you know what I mean).
Basically I think that the battling as such in many respects is a lot more "AI-friendly" than in previous versions which leads to desirable results.

I mean, you can easily check that: Simply make two identical armies with identical heroes and so on and fight it out in II, III, IV and V. I think, that V will have very good results.

[Jolly Joker]

.And in equi it is even better(excelent that is).

Equi doesn't count. You should know that.

[Humakt]

H IV had no AI; the reason for this is mainly that the battle AI couldn't cope with what is necessary. It can't exploit spells to beat vastly superior numbers - which is a must on expert and impossible. So the adventure AI is rather helpless most of the time.

Try out Duel Arena (very old map which actually shows, but it is ideal for combat AI testing purposes anyway) and see how many bonuses you can give the AI only players before you are unable to beat them in combat. You'll see combat AI aren't clearly so bad as you think.

And really problem with HIV was the other way around, the adventure map AI and the skill selection (maybe this was fixed with Equilibris) were bluntly said bad. Also I'm pretty sure AI battles neutrals with Quick Combat and they didn't tone it properly for expert and impossible setting.
Which is the reason why AI's heroes' skills were bad and branched instead of good and focused and it lacked almost always in troops as well, on normal maps.

What you forgot to also mention is ridiculous spell casting of Heroes 3 AI (inability to cast Berserk, preference to cast Protection from Fire type spells).

[DaemianLucifer]

Equi doesn't count. You should know that.

But it does,because the equi team didnt change the AI almost at all.If they could,they would.They only opened doors for it,and it took the step on its own,so it shows how good it actually is.Its just hidden.

[Jolly Joker]

It does not, because you don't know what doors can be opened by a dedicated team in 3 years work in H V. If you really allow fan-made modules into the equation you cannot compare things at all, so it doesn't make sense to do it.

[DaemianLucifer]

Thats always a posibility,and I always said that the best thing about HV is its easy modability(compared to previous sequels,that is,not to other games),and I honestly do hope that something as good as equi can be made out of it.

[Jolly Joker]

H IV had no AI; the reason for this is mainly that the battle AI couldn't cope with what is necessary. It can't exploit spells to beat vastly superior numbers - which is a must on expert and impossible. So the adventure AI is rather helpless most of the time.

Try out Duel Arena (very old map which actually shows, but it is ideal for combat AI testing purposes anyway) and see how many bonuses you can give the AI only players before you are unable to beat them in combat. You'll see combat AI aren't clearly so bad as you think.

And really problem with HIV was the other way around, the adventure map AI and the skill selection (maybe this was fixed with Equilibris) were bluntly said bad. Also I'm pretty sure AI battles neutrals with Quick Combat and they didn't tone it properly for expert and impossible setting.
Which is the reason why AI's heroes' skills were bad and branched instead of good and focused and it lacked almost always in troops as well, on normal maps.

What you forgot to also mention is ridiculous spell casting of Heroes 3 AI (inability to cast Berserk, preference to cast Protection from Fire type spells).

Well, I don't think that the battle AI in H IV is so bad. I think, however, that the complexity of battles in H IV is highly overrated. The only complex thing there is actually the hero/units ratio or, in other words, the force you field into the battle (i.e. how many heroes and what kind of heroes). After that (once there ARE forces locked in combat) I find the battles not too complex, the big battlefield notwithstanding: Kill the heroes. So the inability of the battle AI starts already with their heroes not starting a battle with a PoI downed.
Anyway, I don't think, that battle AI is the point here. Fact is, the AI in H 5 is a very good opponent - or does someone think different?

[MistWeaver]

With H II, basically the hero skills are added - plus a more complex spell system and town building process with the upgrades. The AI is still good, but not that good - you'll much too often take a town with a dozen Titans in it, for example.

And why dozen Titans in captured town is bad ? May be because H2 AI does not cheat for money & res ? Or may be when you play you always have enough gold & gems to by them (5000 + 2gems each. most expensive unit in H2) ? Of course youll never see that in H5 with its nearly unlimited gold.

H III AI already sucks. Combat AI is a lot worse than their predecessors, mainly due to 3 facts: 1) inability to utilize wait command; 2) inability to pick the good skills and develop competetive heroes. 3) AI attacks weakest unit most of the time/too easy to lure AI into doing that. The adventure AI "looks right", but it still doesn't know what to do. It has a lot of resources as well.

It seems as if you like to give loud phrases that are opposite to common opinion. You are saying H III AI already sucks but arguments stops on its battle part. While the major any Heroes game AI part is adventure part. If its dead, nice battles will not help.
And guess what - H3 adventure AI is simply the best in the series.
As for

It has a lot of resources as well

Did you test it personaly ? And how ?
I did. Just build the map with the AI on isle blocked with large monster stack. Also I put a daily script that removes exactly the sum that town gives.
And you know - after AI run out of its start money he stoped buying & building.

H IV had no AI; the reason for this is mainly that the battle AI couldn't cope with what is necessary. It can't exploit spells to beat vastly superior numbers - which is a must on expert and impossible. So the adventure AI is rather helpless most of the time.

Now that the most ridiculous part. Did you ever watched a single game played by H4 AI ? Its about random moves.
And the battle H4 AI (except for siege) - is realy good. And dont forget that H4 has most complicated battle gameplay. Heroes, potions, scores of spells, line of shoot, line of cast. A lot of creature abilities. And AI could handle all this pretty good. So it is adventure AI that ruined all.

With you saing absolutly opposite to facts things - I just dont know what to think. Or you are doing this in purpose, or didnt play H4 much.

H V has the best battle AI by far if you take the complexity of battles into consideration.

If take into consideration that complexity of battles here is much lower than in H4 - Its not the best.[/quote]

Sure, you sometimes see Horned Demons exploding without anyone near - but strangely enough, when that happens those Demons wouldn't have done any damage in that battle anyway.

Yes, and lets not forget about casting armagedon while fighting secondary level 1 heroes with bunch of pixies.
But I agree, its more like a bugs.

Now, the adventure AI. On hard and heroic the AI gets a resource advantage that is pretty normal.

It is not getting advantage. It becomes nearly independent of what is called economics. Considering that H5 is the most "expensive" game in series - that is not acceptably. But even this does not help it.

When you play heroic I don't think you can argue against the AI starting with the Easy resources.

Of course not. But I can argue 10000 gold/day on NORMAL.

All in all, for a game like this the AI is rather decent. I'd rate it on par with H II - considering the fact that H V is a bit more complex.

I belive conscience didnt let you put H3 instead H2
Ok lets see .. yes I agree that they can seem equal. But where H5 takes with packs of archangels and thousands of traiend archers, H2 AI just doing good on its own.

Yes, and one more thing about H5 AI - it almost never loses troops on neutrals and other AI players. Its because nival did "special" combat emulation for it. In other words AI does not use same Quick Combat that we do. It uses special one with which he CAN NOT lose fight vs other AI controled creatures/heroes and suffers almost no loses.

Moreover, I think that the starting amounts in H V are ALOT better chosen than all those in the previous games, because the fact that you start with SOMETHING doesn't make a map depend on what the map maker placed where in the vicinity of your town. You have enough to never have to reload necause you missed the dearly needed chest by going into the wrong direction and so on.

As I said H5 - is the most "expensive" game. So doing starting amounts in H5 similar to those in the previous games would be ridiculous.

[DaemianLucifer]

It seems as if you like to give loud phrases that are opposite to common opinion. You are saying H III AI already sucks but arguments stops on its battle part. While the major any Heroes game AI part is adventure part. If its dead, nice battles will not help.
And guess what - H3 adventure AI is simply the best in the series.

Actually HII is better because its much more agresive.

Yes, and one more thing about H5 AI - it almost never loses troops on neutrals and other AI players. Its because nival did "special" combat emulation for it. In other words AI does not use same Quick Combat that we do. It uses special one with which he CAN NOT lose fight vs other AI controled creatures/heroes and suffers almost no loses.

Thats not true.Turn on quick combat and youll get the same advantage.It managed sometimes to lose not a single creature in battles I couldnt win without losses even in a dream.

[Jolly Joker]

@Mist Weaver
The simple thing is that the AI got worse with each Heroes game, with Heroes III already having an AI that was more on the really bad side, while H 2 was more on the good side (and by the way, being able to conquer a town with a dozen Titans in it does simply show wrong priorities and no planning abilities).
You can tell whatever you want, the AI in H 3 sucks. The point is not having an AI that behaves like a human but fails short because it just is no human. The point is to have an opponent.
You have an opponent in H I and H II. You don't have one in H III and H IV. In H III it LOOKS like you had one (but superior playing beats the AI, haha), but in reality, seeing what the building costs, seeing the advantages the AI has (troop amount of starting heroes, amount of starting resources, added creature growth, complete view of the whole map) the result is pathetic: while all those cheats lead to the AI building something up what looks like an opponent, the actual conclusion is always a downer because the AI develops inapt heroes (main reason for battles being unsatisfactory). The costs of H V are not that much different from those in H III, at least not money-wise, by the way.
You have no opponent in H IV as well, mainly due to the inability of the Quick Combat AI - the opposition cannot develop. If you want to, you can think that H IV is oh so complex, but for me it isn't. It LOOKS complex on the surface, but the core is surprisingly simple. The complex part is the massively upped spell part since an army may be able to cast a dozen times within one round, however, as I said, the difficult part is in reality to "compile" the correct force, not the battle with that force.
However, to avoid battling on secondary fields, ok, let's give H IV a decent battle AI. So what? There is still no opponent on any simple MP map you try alone.
That leaves H V and there we have an opponent again which is why I rate it as on par with H II. There is an opponent that will deliver a game. If you want to have an opponent who isn't cheating as well, look elsewhere.
I don't care about the AI (for god's sake, a freaking machine; so you take it personal when you are cheated by a machine?) cheating like hell. All I care for is an opponent.

[MistWeaver]

Yes, and one more thing about H5 AI - it almost never loses troops on neutrals and other AI players. Its because nival did "special" combat emulation for it. In other words AI does not use same Quick Combat that we do. It uses special one with which he CAN NOT lose fight vs other AI controled creatures/heroes and suffers almost no loses.

Thats not true.Turn on quick combat and youll get the same advantage.It managed sometimes to lose not a single creature in battles I couldnt win without losses even in a dream.

That true DL. First - try to fight equal army and youll see losses. Second - Ive read talk with nival on offical forum where people asked to remove insant and cheatable AI combat resolver with the normal "quick combat". Nival representative said that they would not do it.

[Alamar]

@Mist Weaver
That leaves H V and there we have an opponent again which is why I rate it as on par with H II.

Of course the fact that you have an opponnent isn't because of anything "good" that the H5 AI has done. H5 [IMHO] doesn't do any better job of picking which skills to learn than any other iteration of HoMM. The fact is though that you have fewer "cruddy" skills / abilities compared to other iterations.

I'd still like to see the way that AI heroes pick skills / abilities get totally revamped even if you have to go to the point of developing multiple scripts for each AI hero so it doesn't have to actually make choices.

------------------------------------------------------------------------------

@MistWeaver: As far as the AI having it's own "quick combat" option I don't really mind that so much as long as Nival doesn't get TOOOOO carried away with it.

------------------------------------------------------------------------------

@JJ: If the AI doesn't keep up the illusion that it's playing by the same rules that we are I do think that's a big problem in playing a satisfying game against the AI. Trading off some competitiveness to preserve that illusion would make for a more enjoyable game for many people IMHO.

As far as "nobody can argue that the AI should start with the EASY difficulty resource amount on HEROIC difficulty games" I would argue that it would be far better to just give the AI small daily bonuses instead of giving it a huge lump sum bonus at the start of the game. This one action would help minimize how blatant the resource cheats of the AI are.

[Jolly Joker]

It starts only with Normal resources, by the way.
As I opined in another post: since the AI is no human, I don't expect it to play like a human. IN fact it should not even play like one, because it simply cannot. If I want to play another human I play another human. The AI is some kind of sparring partner and it's aim is not to create the illusion of a human playing on the other side. The AI NEVER plays under the same rules we play - or do we see the whole map, for example?