H5 AI harder than H3 AI?

[Slepytchee]

I myself dont think the AI would need more help when being played against...

[zero cool]

1. Allow the AI to have perfect knowledge of the map.

I didn't mean to imply that the H5 does or doesn't have perfect map knowledge. That's just one item that I think would be "required" to help even the odds against humans that can reload if they mess up.

Well, I can see how this would work for single player but doesn't it make it unfair when playing bots in multi-player?

Perhaps I should have asked this first: Can you save a multi-player game in HOF or TotE? I only have the original.

[Asheera]

Of course, you can save a multi-player game in all versions of Heroes 5. However, when loading it you need all the players to join, obviously (and they must have the same name from the save to be recognizable)

[zero cool]

Last MP I played, I couldn't find out where to save it and had to quit. I must not have looked hard enough.

I'll have another look, thanks.

[Asheera]

Only the first player can save, and only when it's his/her turn.

[Rife]

Why has no one tried building staged AI? Three would suffice: Build/Defend, Explore/Expand, and War. Five, if you count Battle AIs: Vs Human and Vs Creep.

These are basically the stages humans use, and doing it this way would seem more satisfying to play against (not to mention make things easier for scripting maps). You could switch stages according to resource levels, time limits, or when enemies pass into certain territories.

Certainly it would seem easier to create multiple AIs with specific, focused objectives rather than a single one that tries to do everything (and by that usually failing at everything).

If they could combine the Battle AI of H4 with the adventure map AI of H3 it would be the best Heroes game ever. *sigh*

[Asheera]

Five, if you count Battle AIs: Vs Human and Vs Creep.

The AI in H5 cheats against neutrals anyway. I've seen it take 50 ArchAngels week 3 with Inferno.

Either it cheats or it is extremely clever and knows tricks to 'abuse' the neutrals... very unlikely lol

[Rife]

Which is one the the reasons I liked H4 battle AI so much. The autocalc system was the best of the series, only human players with excellent battle tactics or powerful spells could survive a battle with fewer losses than the autocalc predicted. So the computers never had to cheat.

[Akul]

I think the only hoMM game where AI never cheats is H4. It ain't even bad. Unfortunately, it is unable to recognise the value of artifacts and is often too cowardly. However, if you give it a good army, reveal the entire map to him (he doesn't have FoW so he doesn't need scouts but is still unable to see unexplored land) and though scripts make enemies look much weaker, it can be pretty decent (that, of course, means that it is good in only fan-made maps).

[Kristo]

Why has no one tried building staged AI? Three would suffice: Build/Defend, Explore/Expand, and War. Five, if you count Battle AIs: Vs Human and Vs Creep.

These are basically the stages humans use, and doing it this way would seem more satisfying to play against (not to mention make things easier for scripting maps). You could switch stages according to resource levels, time limits, or when enemies pass into certain territories.

Certainly it would seem easier to create multiple AIs with specific, focused objectives rather than a single one that tries to do everything (and by that usually failing at everything).

You say that as though it were easy. Who's to say that's not how the H5 AI is currently written? The real problem IMHO is that most, if not all, TBS games are ultimately about controlling land, not objects such as mines and castles. You can teach a computer to control objects, but it's very difficult to get it to effectively define and defend the territory it owns. I think the human brain has a huge advantage here. If you look at a Heroes game in progress and ask a human player to define what territory he controls (and I mean defensible ground, not just sphere of influence), he'll make you a rough sketch of the ground he can defend. We do this mostly by feel, a concept that doesn't translate easily to 1's and 0's. I've not yet met a computer AI that could do it.

The result is that the human player wins all fair fights. The computer wins only if the human makes serious errors or if it has advantages in information or material. Now, in order for computer games to sell, the human player has to win most of the time anyway. The trick is to make the computer opponent fun to play against. Here's where the H5 AI has failed miserably. It is clearly playing by a different set of rules (or is brain dead when the rules are fair). That's not fun.

In summary (and to actually answer the thread's question), the H3 AI is better because it provides a more enjoyable experience, despite the fact that the H5 AI wins more games.

If they could combine the Battle AI of H4 with the adventure map AI of H3 it would be the best Heroes game ever. *sigh*

I think the best overall AI player is from H2. My impression is that the computer says, "I know I can't beat you, so I'm going to be a constant thorn in your side until you win."

[Asheera]

The real problem IMHO is that most, if not all, TBS games are ultimately about controlling land, not objects such as mines and castles. You can teach a computer to control objects, but it's very difficult to get it to effectively define and defend the territory it owns. I think the human brain has a huge advantage here. If you look at a Heroes game in progress and ask a human player to define what territory he controls (and I mean defensible ground, not just sphere of influence), he'll make you a rough sketch of the ground he can defend. We do this mostly by feel, a concept that doesn't translate easily to 1's and 0's. I've not yet met a computer AI that could do it.

That's why I always had this idea to make a game AI that could use the help the map-maker gives it according to the specific map. Place in a map helpful waypoints and other zones for the AI to understand better and control its territories.

For example, I always hated in games like Starcraft where the AI places defensive structures (like Photon Cannons, etc). With some 'defensive' positions written by the map maker it would build them where it should.

Sure that would be more of a pain for map-makers, and that's the downside of this idea unfortunately.

[Rife]

Indeed, in fact Warcraft 3 comes to mind; you could set rectangular zones for pretty much anything and assign simple or complex AIs for units or computer players in those zones. Granted it's not a turn based strategy game, but that just demonstrates it can be done.

[Pitsu]

Wow, is that really Rife here? Who resurrected him?

I think the best overall AI player is from H2. My impression is that the computer says, "I know I can't beat you, so I'm going to be a constant thorn in your side until you win."

While some would start talking how much simple and bla blah H2 was, i think a problem is that h2 maps were more open. In H3 many maps are divided into zones and by guarding a few passages one can control entire zone territory. In H1 and H2 it was more easy to find a way to opponents backyard. It makes the AI more challenging.

In real time strategies AI has a fair chance against human due to being faster in computation and ability to govern every unit and action nearly simultaneously. Time cannot be a factor in turn based games. Controlling many armies, towns, strategic keypoints, however can still be applied to some extent. With a single main hero, small number of towns and few conflict zones, human easily controls the situation, even if the user interface lacks some info. Force human to control more objects and events in parallel, and soon the comfort in "controlling the game" starts to shake.

[Rife]

Yeah, I'm back. Picked up a copy of Heroes 5 some time back cheap and got hooked again. Thinking of taking up mapmaking again as well, but I must admit the H5 editor looks intimidating. I'm hoping I can mitigate the the AI's "RUSH! RUSH! RUSH!" mentality with scripting, as I've found it to be mostly unfun.

[Emperor_]

It's alwasy fun that the AI know when you target their town and start heading for it, since every time you do it, they start heading home to that home-town , on every homm game.

But yes, the AI is tuned,tweaked and more polished then the Homm3 AI, tbh, the easiest AI on the whole serie is in Homm IV, where it was no problem for a summoner to beat the starting mobs (that are in high stacks) in Impossible at level 1, aswell if it was a hero 5 levels above you

[Banedon]

In real time strategies AI has a fair chance against human due to being faster in computation and ability to govern every unit and action nearly simultaneously. Time cannot be a factor in turn based games. Controlling many armies, towns, strategic keypoints, however can still be applied to some extent. With a single main hero, small number of towns and few conflict zones, human easily controls the situation, even if the user interface lacks some info. Force human to control more objects and events in parallel, and soon the comfort in "controlling the game" starts to shake.

Never played it, but apparently Galactic Civilizations II (a turn-based game ala Civilization IV and Heroes) had some really well-written AI, capable of holding their own and even outplaying human players. I don't see any reason why the H5 AI can't do the same, I mean sometimes I stall a vastly superior AI hero for weeks simply by moving my own hero back and forth, and then overwhelming him after gathering enough troops, what gives?

The H5 AI offers no challenge once you survive his initial onslaught - but to be fair, neither does the H3 AI.

[Asheera]

The H5 AI offers no challenge once you survive his initial onslaught

True. But the initial attack is pretty hard to counter especially on Heroic, because the AI cheats when fighting neutrals.

Still I don't really like early fights like that, and afterward it is no challenge at all so you could just give up the game.

[gaspi2]

IMO heroes 5 has the worst AI from all series.
About games with good AI...I think AI in Alpha Centauri is almost unbeatable at the highest difficulty. And it doesn't cheat at all since you can activate editor during play and see what's he doing.
Alpha centauri gives a great challange if you choose right difficulty, that's how it should be IMO, AI cheating on all difficulties cuz it can't beat human is ridikulous, I don't mind cheating on higher ones since AI can never match human opponents on it's own(until it's perfect).

[Kristo]

Never played it, but apparently Galactic Civilizations II (a turn-based game ala Civilization IV and Heroes) had some really well-written AI, capable of holding their own and even outplaying human players. I don't see any reason why the H5 AI can't do the same

Let me know when Nival's CEO can write the AI himself. That's what it takes.

[Little Raven]

GalCiv 2 (one of my favorite games of all time) does indeed have excellent AI, but there are a couple of things to keep in mind:

• Brad Wardell, the CEO, literally spent months and months tweaking and tuning it, because it was near and dear to his heart. Stardock is not like most game companies.
• It still cheats, depending on the difficulty level. At normal levels, it plays fair, but at normal levels most competent players will win, with perhaps a bit of struggle. Good players will walk all over the AI at normal. Expert players routinely beat the AI even when it's given god-like powerups unless they get a particularly bad start or something.

Programming AI is very, very hard. Computers are still very limited at dealing with complex, dynamic situations.

I'm coming to the HOMM series very late, with V being my first game. I'm not disappointed in the AI...it's not great, but it's not completely brain dead. It punishes me if I mess up too badly, which is about as much as you can reasonably ask for.