Question about AI Quality

[Gaidal Cain]

Hmm. Maybe for shorter games (final level around 15). If it the game goes on long enough for most skills to max out, the warlock will need it to be able to make use of IM and its abilities though- I'm not really sure if it combines with Fire wall, and even if it does, it's nowhere near as effective as Meteor shower or Implosion. Summoning magic is better than Light and Dark, but it' still nowhere near as dependant on Spell Power as destructive.

And this doesn't affect the main point: the fact that you'll occasionally (more than 19% of the time, going by the percentages, assuming four skills have to be choosen and that the fifth is one of the 10%-chance skills) end up with Dark or Light magic, or not end up with either Destruction or Summoning (harder to make a quick and good estimate, but somewhere between 10 and 18%, so say 15% of the time). Neither of these usually make for an effective warlock on a standard map, yet one of these would happen about 1/3 of the times. Can you honestly still say that letting the "percentages take care of it" would make for a good hero build?

Edit: more precise calculations

[Jolly Joker]

Are you sure about the percentages?
I mean, if you'd simply take a completely random pick of the usually four offers, even if Light or Dark would be offered there was only a 25% chance it would be picked. Since there is only a 2% chance for light and dark to get offered, this would reduce things massively.

[Gaidal Cain]

Hmm. I might have made a few mistakes, but the magnitudes should be right. Suppose the hero starts with a 10% skill (not very unreasonable), and the standard IM. Then the chance of the AI picking either dark or light is 0.04/(1-2*0.1)=0.05. The chance of not getting either skill for the levels can then be taken to be 95% for all levels (simplification here), and thus the total chance of getting them is 1- (0.95)^4 which is 19%. The chance of not getting Dark or summoning is similarly 69%, and 0.69^4 is about 22% (so in reality, the heroes would something out of either set more often than that, due to the number of choices left being reduced).

[Alamar]

The main thing is this here:

All you need to make gameplay experience similar is:
1. Have the heroes that the AI builds basically look like the sort of heroes a human might build [even if the builds have to be scripted] under similar circumstances.
2. Have the army sizes of the AI look roughly [maybe a little larger] than what a human would build over that time.
3. Have the AI on a time schedule that is not too dissimilar from a human.
4. Have the AI use some basic goals that humans might share.
Note you could "cheat like the dikens" where the AI doesn't have to do ANYTHING in a manner that a human would do it and still end up with a situation that would make the gameplay experience reasonably similar.

With this kind of priorities I don't see a challenge anywhere, at least not a heroic challenge. To makeit a heroic challenge you would have to assume that the AI does everything BETTER than a human - a LOT better. Because you can safely say that the battle AI will never be able to be as good as a human. Heroic should be a challenge for veteran players as well, otherwise there is no long time motivation. There is nothing wrong with a non-veteran player being beaten by the AI on heroic. There is not even something wrong with a veteran player being beaten by the AI on heroic.
So it's simply not enough to do things roughly the way a human would do - which is not possible anyway considering the wide variety of possible victory conditions and tactics.
Strictly spoken, a heroic challenge on a map where the initial position is equal for all (a typicl MP map) means that the AI players must be stronger than the human player and I don't think this is possible with your basic concept.

IMHO having AI factions with slightly more powerful heroes than the player with slightly more powerful armies can still provide a challenge on heroic difficulty.

Having AI heroes with mass spells could improve their Light / Dark spell effectiveness by 5x. Empowered Lucky Meteor Showers [with slippers] or Armageddon could dramatically weaken a human's army.

Depending on how good or bad the battle AI is it would be nice if a human player would routinely lost 1/2 to 3/4 of their army to beat the AI.

If beating the AI wipes out 3/4 of your army then you'll be in TERRIBLE trouble when the next fully powered AI happens to start making a B-Line to your territory.

Of course anything that is done that provides a challenge must be balanced so that the end result is fun.

[Jolly Joker]

@ GC
Yes, I checked. it If the first 4 picks are always the new offered skill, chances for either one of the Light or Dark Magic skills to appear are between 19% and 26% depending on what skills are actually picked as 3rd, 4th and 5th.
I think we can safely agree that the main focus point on hero development is the hero interaction in terms of skills, spells in the guild, and abilities: you don't want to see a Warlock with Destructive Magic and Master of Ice who does neither have Ice Bolt nor Circle of Winter (and Fireball and Lightning Bolt instead), but I would definitely have no problem at all with a Warlock who had Basic Dark Magic, Master of Mind and Slow and Confusion to cast (since you won't have a fully developed hero all the time a Basic non-fitting magic skill would be a good pick for Dark and Light especially when you have the two spells of an ability in the guild, even though Power of Speed makes Light Magic less desirable for a Warlock).
I wouldn't mind the following cheat for the AI: since spells are fixed as soon as a map is generated, the AI should be able to "see" the spells its guild will produce. That makes it possible to see "UNdesirable" magic. "UNdesirable magic" would be defined as follows: the guild does include less than THREE spells of that skilll AND not TWO spells to be combined under ONE ability.
So a Warlock could only pick Dark Magic if level 1-3 of the guild would include a DM spell each or one of the 3 two-spell Master of combinations.
Another condition on ability picking would be of course that a hero could only pick a magic skill base ability if the mage guild OR spell book would contain a spell of that ability.
I think that could work.

[Jolly Joker]

@Alamar
Well, what you propose is basically the old king of the hill difficulty - all AI players ALLIED against the human, which would be another step up the difficulty.
But that isn't happening. On any normal mp map - say with 4 players -, when one of the AI players comes a-calling to the human, the others get active as well: in that moment there is an undefended town (that of the attacking player). So mp players are battling among each others as well which makes things easier once again and which must be possible as well. You need a sensible difficulty setting for those as well.

[Alamar]

@JJ: Ah you remembered the old "king of the hill" difficulty setting I was thinking that doing something similar for "heroic" may not be so bad ....

A lot of that is, of course, map dependant. If there is one human and 5 AI factions a King-of-the-Hill-like approach COULD be quite challenging.

However for a 1-on-1 [or similar map where ganging up isn't feasible] then my system for heroic breaks down to simply "a little harder than hard difficulty" because of larger AI bonuses. .....

As far as Spell-Tower-lookahead cheats I think that would be a very important and "good" kind of cheat. It would result in stronger heroes theoretically AND it's the kind of cheat that is pretty subtle so most people hopefully wouldn't care. FYI: Spell Shrine lookahead might not be so bad either ....

One thing that we're not addressing that should possibly be looked at is what happens when the map starts with each faction at an advanced level. Part of my system does include economic bonuses to keep the AI ahead of humans but that system is only "really" effective when those extra resources could allow the AI to build faster than the human [or recruit more effectively]. Any ideas for what should be done then or should be fall back on the good old "rely on the mapmaker" for help??

EDIT: I would prefer that additional creatures not be awarded as a function of the game [mapmaker only power].

My AI [like HoMM5] would have its own set of combat estimators that the AI could use against neutrals. Basically its goal would be to give results that are similar to a human but perhaps just a little better [so the AI will have a little larger army than the human under normal circumstances]. Therefore this sort of system wouldn't help much when each faction already has completed castle(s).

Any ideas?

[Jolly Joker]

Difficult to answer this, Alamar. (had to take some time to come up with some ideas.)
No matter the way you look at it, a human can accomplish everything the AI can "legally accomplish", and in battle, if we don't want no cheats - and I don't - the human will be superior.
So in the end the only way to make things difficult here is to make sure that the AI has a bigger army AT ANY TIME OR additional primary skill points for the main hero.
One very subtle way to get that last point in MIGHT be adding an effect bonus to artifacts (+50% on hard, +100% on heroic).
I'm no friend of extra creatures, somehow. Don't really know why - additional creatures really make sense.

[Khelavaster]

As far as Spell-Tower-lookahead cheats I think that would be a very important and "good" kind of cheat. It would result in stronger heroes theoretically AND it's the kind of cheat that is pretty subtle so most people hopefully wouldn't care.

I think you've hit the nail - the idea is that if there is to be any cheating, it should be sufficiently minor and subtle to maintain the suspension of disbelief, so to speak.

It's too bad I can't seem to find it anywhere now, but I distinctly remember reading an interview with Gus Smedstad (guy in charge of adventure AI in HOMM3) some years ago and he said that the AI only cheated in knowing where were your heroes at all times, thus enabling enemy heroes to remain just outside your scouting radius and surprising you now and then (plus extra starting resources and gold in higher levels).

HOMMV's cheating is of the worst kind because of its cheapness. A trained monkey with free troops, gold and levels will have a good chance to win against a human, in fact it just has to remain still and it becomes an obstacle in itself- another neutral stack. Not unsurprisingly, the travesty that is the campaigns in this new installment has you battling hordes of neutrals and garrisons only to reach the big static foozle at the end... or escaping from someone, or some other scripted silliness.

HOMMV adds some minimal AI to the process, but not enough to make the computer to play by the rules or act remotely like a human (picking up unguarded resources for example). Far inferior than its predecessors (skipping IV), this AI seems focused on creating artificial challenges by pulling bunnies from hats while forgetting what a strategy game is all about.

Khel.

[Alamar]

@JJ: See below at 2nd part of the post

As far as Spell-Tower-lookahead cheats I think that would be a very important and "good" kind of cheat. It would result in stronger heroes theoretically AND it's the kind of cheat that is pretty subtle so most people hopefully wouldn't care.

I think you've hit the nail - the idea is that if there is to be any cheating, it should be sufficiently minor and subtle to maintain the suspension of disbelief, so to speak.

I agree that it would be nearly perfect if the only kind of cheats that the AI needed were of the variety that were either so minor as to never offend or so subtle that it would "never" be caught.

I'm not a game designer but I suspect that to be competitive on nearly all maps on nearly all difficulties in nearly all circumstances the AI would have to cheat in overt ways. Cheating in these overt ways may cause a conflict between goals of competiveness & getting mad at AI cheating ...

The more I discuss things the more I'm convinced that players should be able to configure AI performance, cheats, bonuses, etc. via some set of sliders. Then people that want the utmost in challenge can have it while others that want to play a straight-up game can.

------------------- Avoiding double posts -----------------------------------

Difficult to answer this, Alamar. (had to take some time to come up with some ideas.)
No matter the way you look at it, a human can accomplish everything the AI can "legally accomplish", and in battle, if we don't want no cheats - and I don't - the human will be superior.
So in the end the only way to make things difficult here is to make sure that the AI has a bigger army AT ANY TIME OR additional primary skill points for the main hero.
One very subtle way to get that last point in MIGHT be adding an effect bonus to artifacts (+50% on hard, +100% on heroic).
I'm no friend of extra creatures, somehow. Don't really know why - additional creatures really make sense.

I think that these "high level" maps where the map maker makes it balanced for a human without bothering to worry about the AI is the bane of AI developers. IMHO it would be nice for mapmakers to take care of this one but as a designer you at least have to try ....

A map like this is going to come down to 2 things basically ... hero strength and army strength.

If you're going to allow cheats then consider:

1. Allow the AI to pick the best skills/perks available [or at least the best it can find]. If this means that on day 1 the AI has an ultimate ability then that might not be soooooo bad. Obviously play testing should be done to make sure that it passes the "fun" test.

2. While I wouldn't enable this as a default I'd have a slider that you could use IF you wanted the AI producing more creatures than humans.

3. Maybe allowing the AI +X% to level [yes I said level not XP] would be something to look at.

4. Catchall ... consider anything as long as the player can control it.

IDEA: I may have a "lot" of sliders and configs controlling AI design. Maybe it's not a bad idea to allow players to save templates out for how they want the AI configured. This way once they have the sliders the way they want them they just choose their template from a drop-down list. This would save lots of time frutzing with sliders.

[Khelavaster]

I'm not a game designer but I suspect that to be competitive on nearly all maps on nearly all difficulties in nearly all circumstances the AI would have to cheat in overt ways. Cheating in these overt ways may cause a conflict between goals of competiveness & getting mad at AI cheating ...

When playing HOMM3 cooperative with a couple of friends against the computer, we used to play on a high difficultty and ramped up the number of opponents to face a good challenge.
So I'm not against voluntarily stacking the deck for the computer if you feel you've reached that level.

The essential thing is, even if we won most of the time, the process to get there was immensely satisfactory. Why? The actions-> results chain followed logically. That is, if we took a mine from the computer, that action hurt its economy.

Not so with shoddy HOMMV. This is basic stuff kids, you don't mess with the foundational rules. That's what we call bad AI.

Khel.

[Alamar]

I'm not a game designer but I suspect that to be competitive on nearly all maps on nearly all difficulties in nearly all circumstances the AI would have to cheat in overt ways. Cheating in these overt ways may cause a conflict between goals of competiveness & getting mad at AI cheating ...

When playing HOMM3 cooperative with a couple of friends against the computer, we used to play on a high difficultty and ramped up the number of opponents to face a good challenge.
So I'm not against voluntarily stacking the deck for the computer if you feel you've reached that level.

The essential thing is, even if we won most of the time, the process to get there was immensely satisfactory. Why? The actions-> results chain followed logically. That is, if we took a mine from the computer, that action hurt its economy.

Not so with shoddy HOMMV. This is basic stuff kids, you don't mess with the foundational rules. That's what we call bad AI.

Khel.

IMHO no static / unconfigurable AI is going to please "the vast majority" of HoMM5 fans [at least hardcore].

Giving map makers the tools they need to easily help the AI AND giving advanced players ways to configure the AI behaviour before the game starts [sliders or similar] is the only way that I see to please "the vast majority" [of hardcore fans].

If anyone else has a better / more realistic idea then please let me know what it is ....

[Khelavaster]

IMHO no static / unconfigurable AI is going to please "the vast majority" of HoMM5 fans [at least hardcore].

Giving map makers the tools they need to easily help the AI AND giving advanced players ways to configure the AI behaviour before the game starts [sliders or similar] is the only way that I see to please "the vast majority" [of hardcore fans].

No, of course that would be peachy. The best solution (IF, of course, the AI slider allowed for a challenging non-cheating AI).

But this is Nival we're talking about. They couldn't even provide basic options that nowadays are commonplace. Lets say, er, how about disabling tree or object animations to compensate in some way for the suckiness of their sorry engine.

Beautiful concept, but somehow I don't see AI sliders happening anytime soon...

Khel.

[Alamar]

Realistically AI sliders don't stand a chance until HoMM6. Of course something might be implemented in an expansion but it would take Nival / UBI being willing to overhaul the AI and do some additional GUI screens.

For an expansion I'm not sure that they'd be willing to do that .... or anyone else for that matter.

[aulfgar]

Realistically AI sliders don't stand a chance until HoMM6. Of course something might be implemented in an expansion but it would take Nival / UBI being willing to overhaul the AI and do some additional GUI screens.

For an expansion I'm not sure that they'd be willing to do that .... or anyone else for that matter.

I really hope hope that Nival could give us a working AI soon, as the current one does dumb things in the adventure map and in battle. I don't necessarily mind cheating AI, its just overt cheating that bothers me.

Is there a reason the AI likes to go through the border guards to come after my mines... when he hasn't even snagged his own? There needs to be a priority for close and usefull and not so much in the "ooo lookie an undefended town 4 weeks away" vein.