Will the ToE AI be less dumb??

[okrane]

@Kristo

I am certain that it's not just a copy paste thing, but I find it not that hard to use the same ideas like in that one. I mean, basically the game is the same. The game play is mainly the same, and the adventure map decisions are not that different.

What I liked about H3 AI is that is never gave up. You would end up finding yourself besieged for months in a row by strong enemy heroes that came, attacked, retreated, recruit new army and attack again. You were forced to train multiple heroes just to withstand the attacks from all sides.
Furthermore the AI fought to the last breath. You would have found it always trying to capture your town and attack you in some manner. Really rarely did I find an AI opponent in H3 that would just sit in its town.

In H5 the things are much different. The AI trains only one hero, which he does not retreat in case of a lost battle and it gives up on attacking you in the middle of month 2 usually.

I don't buy this, "let the player win" thing simply because of the fact that if the AI turtles in its towns in the mid of the second month, it will take you at least a couple of weeks to fully eliminate him. This coupled with the long turns the AI makes(just by sitting in its town) brings a long time spent doing nothing. I mean it's simply a wasted time of my life, that hour or so, when I just go between its towns in order to conquer them all. It's by no means fun, since I know how it will turn up, it's simply a chore. And if you suggest quitting before that, I'll say that this will diminish the satisfaction of winning. So, I'll have to ask... does the fact that the AI lets the player win bring more fun to the game, or not... My answer is no.

I agree to the AI cheating. In terms of money at least. I do not agree with the fact that it gets double growth. I found myself once attacked by an AI who had like 15 level 7 creatures in week 3 (it had 2 towns). Now that pissed me off. I agree to it upgrading really fast all its town, also to not losing creatures when fighting neutrals, but no extra creatures please.

[Apocalypse]

Actually I modded the game and play with double growth AI on Heroic and a 1.5 growth AI on Hard (the Heroic one is tough)

2) The AI can develop a map-specific strategy only as far as it is programmed to do so; does someone really think the AI programmers would be able to foresee all possible map constellations that may occur on any map.

They could add some events and special places on the map. That's what I hate in most strategy games: the AI should have specific strategies for each map, like Mobile Forces has (a FPS game based on the UT engine). For example, about Warcraft 3 (some yucky things):
- sometimes the AI blocks his units completely in his own base with his buildings.
- the defensive towers are never placed on strategic places (you should have some 'defensive' positions for each player on every map to make the AI more 'intelligent')
- it always goes for the closest gold mine (to expand), though sometimes the closest is actually the 3rd supposed one (on some maps) and is guarded heavily. So, you should have some 'second and third Gold Mines' positions and that kinda' stuff.


I'm not saying you should all be stressed to make a good AI on your map using those special positions; they could make the game so that without them (special positions and stuff) the AI would play how it currently plays (dumb)

[Kristo]

In H5 the things are much different. The AI trains only one hero, which he does not retreat in case of a lost battle and it gives up on attacking you in the middle of month 2 usually.

Yeah that's inexcusable.

I don't buy this, "let the player win" thing simply because of the fact that if the AI turtles in its towns in the mid of the second month, it will take you at least a couple of weeks to fully eliminate him.

A good AI allows the human player to win, but not before making him feel like he's earned it. The turtling behavior you describe indicates a lack of effort on the developers' part.

This coupled with the long turns the AI makes(just by sitting in its town) brings a long time spent doing nothing. I mean it's simply a wasted time of my life, that hour or so, when I just go between its towns in order to conquer them all. It's by no means fun, since I know how it will turn up, it's simply a chore.

It's pretty easy to have the AI do the majority of its thinking while you are taking your turn. In terms of modern CPU cycles, a few seconds is an eternity. The computer can get a lot done in the time it takes me to select another hero to move around, or decide on which building to buy. By the time I click the end turn button (or even as I move the mouse toward it), the computer should have a pretty good idea of what it's going to do.

That all goes out the window of course if all the CPU cycles are used up rendering the scenery.

[ThunderTitan]

@ TT
Sure, I see it before me, players losing each map, and being happy about it: "That's what I call an AI, I never won a game against it, but I learned so much for playing online. A pity I don't like playing online due to having to wait on your opponents."

Against THEM, as i said 4 allied, which are huge odds. If it can't pose a challenge with such overwhelming advantages what's the point? Might as well just play against the neutrals.

A good AI allows the human player to win,

Which makes playing against it pointless since there's only one possible outcome. A good AI will be surpassed by the human, not lose to him no matter what.

[Elvin]

It never was a point in making the super duper wow AI that will pwn humans in any given game. Just a better AI that will be more of a challenge should you opt for from the settings. The current one is not fulfilling.

[Alamar]

It never was a point in making the super duper wow AI that will pwn humans in any given game. Just a better AI that will be more of a challenge should you opt for from the settings. The current one is not fulfilling.

Heck all that I'd like to see is an AI that is "better" in all phases of the game than any of the other HoMM AIs. While that's not much of a goal at least it is something that's theoretically obtainable.

Please note: It is not easy to create a "good" AI. I would think that to design a good AI you would practically have to design the AI THEN design the game features around things the AI can do fairly well [and consider scrapping things that the AI just can't do well]. Even then you'll only end up with a so-so AI.

Also note: I'm all for letting the AI cheat. Allowing the users to configure the type of cheats the AI gets would seem to be a reasonable idea to me. Also different kinds of cheats [less obvious] and restructuring cheats may also be a good idea.

[PhoenixReborn]

I like Gaidal Cain's idea of having the bonus apply to the actual resources so that it's still possible to starve the a.i.

But I'll stick to my idea that the A.I. is as good as Nival is going to make it, or it would already be better.

[Alamar]

I like Gaidal Cain's idea of having the bonus apply to the actual resources so that it's still possible to starve the a.i.

But I'll stick to my idea that the A.I. is as good as Nival is going to make it, or it would already be better.

I tend to agree. H6 is our best hope of getting a new HoMM game with a better AI.

[ThunderTitan]

Also note: I'm all for letting the AI cheat.

For me it depends on difficulty level, but a customizable one would be great too.

And the problem is that they never even tried making a so-so AI, because it was their opinion that after a few weeks it should just let you win.

[Jolly Joker]

I've given the matter some thought again, but for me the situation is as follows:

H 5 is giving me a tough game with tough battles, and that's true on every map. If I had to compare it with a predecessor I'd say it's like in H 2 - even though the battles tend to be better in H 5 ON AVERAGE. In this respect it is MUCH better than H 3 and infinitely better than H 4.

However, come to think of it that's not due to the AI, mostly. The battle AI is pretty good considering that you can wait and defend and so on, but I think that the smallish battlefield works in favor of the AI because most of the time it is right to attack all-out (and the AI blunders when it would be tactically sound to play a waiting game).
Secondly, skills and abilities will be effective no matter what, so basically a hero can take any skill and profit, another thing that works in favor of the AI (as opposed to H 3)
Third, and you won't like this. While the turtling tendencies of the AI may be, well, not spectacular, it basically guarantees a tough final battle - which is a lot better a game ending as beating the AI peacemeal, conquering empty towns on day 1 and so on.
Realistically spoken, this is the best tactic the AI can make use of: try to expand early because it has the guns AND the numbers here, and if you happen to get caught on the wrong foot the AI will steamroller you. Game over. If you survive - or the border monsters are too strong -, turtling in a town with 90% or so of your army will not win the AI the game, however, it won't lose it either: you'll still have to beat what may be a superior force.

The bottom line here is - for me - that I find it a challenge to actually win a nice MP map on heroic. It's not easy. Most of the time it's fun. The AI could be better, definitely. However, it could be worse as well. The AI can't take the place of a real human opponent and it makes no sense even trying it. Obviously they are still working to get the quirks out of the AI, but they won't change basic things, and comparing the Heroes games so far I think overall they accomplished the job.

[Apocalypse]

H 5 is giving me a tough game with tough battles, and that's true on every map

I gave double growth to the heroic AI and only then can I say that H5 is giving me a really tough game.

The main problem with the AI for me is that I ALMOST ALWAYS kill his level 13 hero in week 3-4, and then, even though with double growth he has double army, the level of his most powerful hero is around 2-3 which makes me win the game... of course, it's a tough fight, but without double growth he would have the same army but with a level 2 hero vs mine of level 18 or so... (not to mention it's weaker in the battle ofc)

And I found another stupid thing about the AI: sometimes, it's a smartass, meaning that when I'm close to its castle (I want to attack), it goes out of the castle with its hero and ATTACKS me! Even though I'll WIN the battle.

[ThunderTitan]

"it basically guarantees a tough final battle

It guarantees that i have to go back and wait a few weeks to amass more creatures for the final battle. And that no matter what i'll still win, even if a few weeks later.

[Alamar]

I'm of the opinion that turtling up is less of an issue of AI quality as it is a gameplay issue. I would argue that if the AI's goal is to either win [or at least lose in a manner that would make the winner pay dearly in time or creatures] then turtling is often the right decision.

This is esp. true if the AI knows it can't beat you in a fair fight but it thinks that it can weaken you enough so that its allies can then begin taking the fight back to you.

From a gameplay standpoint though it's bothersome that to "win the map" I have to spend 4-6 weeks doing nothing but mop-up. At a certain point it would be nice if the AI just offered to surrender to avoid the annoyance to the player.

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

Before too many people mention AI performance in RTS games I think that its obvious where an RTS AI could easily find performance gains compared to a TBS AI.

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

The only terrible thing about this situation [as TT pointed out earlier] is that nobody at UBI / Nival even cares at all about improving AI quality [even cosmetically]. I think that the proof of that is there have been many AI improvement threads with easy-to-implement-cheats that would improve the AI and/or gameplay but I don't recall seeing those in any of the patch updates. [And as far as I know no expansions]

[ThunderTitan]

Turtling isn't a winning strategy unless resources are limited and the enemy wastes his on your defenses, which doesn't apply to HoMM.

[Alamar]

Turtling isn't a winning strategy unless resources are limited and the enemy wastes his on your defenses, which doesn't apply to HoMM.

The resources that the enemy makes you waste are "creatures in your army" which is a limited resource.

In a free-for-all turtling is never a winning strategy but it might be something that a human would do out of spite. I.E. you may beat me but I'll make it hurt you enought that someone else MIGHT be able to beat you.

In a case where the AIs are basically allied against you turtling could become a winning tactic in the right circumstances.

However, as I was trying to point out, it is often bad for game play to do this as it will bore the crap out of the humans.

[Jolly Joker]

It's just a question of VCs. "Destroy all opponents" is a bad VC for an all-human game, so why should it work against the AI? The "interesting" (MP) maps are those with other VCs anyway. With a nice little "flag all this and that" or "Conquer town X" turtling is a losing tactic indeed and here the AI should obviously NOT turtle (and lacks if it does).

[Orfinn]

Sry for interupting this hot debate about artficial intelligence. But what exactly is turtling?

[Jolly Joker]

Barricading yourself in your town with your main hero, buy out the creatures each week and wait for the opponent to attack you in your town. It's a denying strategy more than a winnig strategy either in the hopes to stalemate or even win a hero battle you'd otherwise lose.
The main thing is that it works only when there is only the "destroy all opponents" VC, which is impractical in my experience in an online or hotseat game against human opponent(s).

Bottom line is that the AI is showing its quality only on maps with other VSs because passive play and turtling won't help winning, while on the other hand other VSs COULD be fulfilled cheating (you can't really control whether the AI really HAD the force to take that town or really DID visit enough monoliths to dig out the tear, do you) or never at all due to the AI not reacting and simply turtling.

[ThunderTitan]

The resources that the enemy makes you waste are "creatures in your army" which is a limited resource.

In a case where the AIs are basically allied against you turtling could become a winning tactic in the right circumstances.

Wrong. you don't need to attack him until his ally is taken care of. Just take his mines, that way he won't be able to buy enough creatures unless he's cheating. And even if he can, your control of adv map creature dwellings and high income means you'll have more eventually.

If he only turtles and doesn't attack you can leave him last.

[Caradoc]

I'm glad to see this topic come up again. For reasons that have already been hashed out, I think it's clear that the AI will never be a challenging pseudo-human player. The human brain is too fast, too adaptable for a static AI to keep up. However, an AI that was able to learn might have a chance. Since we are unlikely to see a neural net or some other heuristic engine running well on a PC that already has a lot to do, it is necessary go to the next best thing -- an AI that can be programmed by the map maker to pursue strategies suited to the map. To an extent, this is already possible, although it more could be done if the mapmaker had access to the priorities assigned to various objectives. Absent that, Heroes IV mapmakers were able to cue the AI by varying force and resource levels --i.e. giving the AI a stack of Black Dragons to make it aggressive and then taking them away at the start of the combat. Unfortunately, this is not easy in Heroes V. However, it is possible to design maps around the proclivities of the AI. This requires the mapmaker to have a good understanding of the AI's behavior. It is not enough to simply have the AI replace a human's starting position.