AI in Heroes of Might and Magic 5 - Nival principles

[Caradoc]

I'd say Nival has this exactly right. This is entertainment and the player/customer should get to win most of the time. A scoring system like we had in Heroes III would be another step in this direction. Compensating for the weakness of the AI is practically inevitable and resource bonuses are the way to do it, since that's part of the handicapping system for different play levels.

As to the question of whether an AI can be superior to a human expert, I would point doubters to the record of Deep Blue. However, that represented millions in research and specialized hardware. We could have an expert AI for Heroes, but the game would be mighty pricey.

[maltz]

They can simply try their best to put a very smart AI, and make it available only through the highest difficulty.

Putting a medicore AI in the excuse of "letting players to win" sounds pretty lame to me.

[Sir_Toejam]

This is a pretty polemic poll with pretty polemic posts. You should face some facts.

the statement: "You should face some facts", is in fact, a polemic statement. Thus, it seems you wish to contribute in kind?

There is no way you can make an AI that can beat a human playing under the same condition than a human IN A GAME LIKE HEROES.

ahhh, I do remember those who also said: "There is no way you can create a computer program that could beat a good human player at chess."

so is what you say a fact, JJ? or a mere assumption on your part?

are the variables controlling battles and resources in H5 predictable?

seems to me they are.

if so, it really is only a matter of programming skill and time, rather than a de-facto impossibility as a task.

nice polemic, though.

[Kristo]

I'd say Nival has this exactly right.

Then why is there still only one "yes" vote?

As to the question of whether an AI can be superior to a human expert, I would point doubters to the record of Deep Blue. However, that represented millions in research and specialized hardware. We could have an expert AI for Heroes, but the game would be mighty pricey.

Deep Blue plays mostly by brute force combined with "books" for opening and endgame. It's raw processing power, not skill, that beat Kasparov. Unfortunately for us, a game like Heroes has too many variables to be effectively searched. The branching factor is too high.

[Sir_Toejam]

The branching factor is too high.

did you mean that in absolute, or relative terms?

it's hardly too high to build an AI to begin with now, is it.

seeds produce predictible results, these results can be used to build a much better AI than currently exists, it's just work.

...and i think Kasparov would disagree he was beat by superior computing power.

EDIT:

of course, you could always listen to what the man himself says about it:

Yes, Deep Blue was 100 times faster, but so what? Sheer power means little in chess because it is a mathematically near-infinite game. The only way to measure the strength of a chess-playing computer is to analyze its moves. While putting Deep Blue's six games to the test with current top programs--Deep Junior and Deep Fritz--we discovered that they consistently play better than Deep Blue. The only exception is when Deep Blue showed a stroke of genius in one game (when I suspected certain interference).

http://www.opinionjournal.com/extra/?id=110003081

[fly away]

Nival are just a bunch of stupid idiotic russian morons...

What does it have to do with them being Russian? It's very impolite to say things like this regardless of what they do.

[ThunderTitan]

As to the question of whether an AI can be superior to a human expert, I would point doubters to the record of Deep Blue.

Which cheated too. Here: http://en.wikipedia.org/wiki/Deep_blue

It only won 3.5 to 2.5 the second time, and IBM didn't want a rematch. Still, it was a pretty good AI.

[DaemianLucifer]

Also I would like to hear JJ coments on that, 'cause I never saw him not agreeing with anything that nival does.

I did a few times

This is such a big pile of crap

Major AI principles :
1) Surrender to player gracefully.
Wisdom that learned nival's programmers from designers sounds like this:

The main task for AI is not to win, but to surrender (lose) gracefully.

Soooo untrue!The main task of the AI is to win,or if it cannot,then at least provide a good fight until the end.

Main purpose of the game - to entertain human, please him. And the human likes to win, he likes to feel himself more superior, even in comparison with electronic blockhead. Therefore AI, that works in game industry as well, must show some resistance in the begining, and then - lose.

Well duh!But beating a cripple is no fun unless you are some kind of sick sadist.

2) If not caught - not a thief (trans. note: its a russian proverb)

That is true,but this one has been caught long ago.

Should AI cheat ? - ideally no. But even creating AI that not wins, just resists - is a very complicated task. Therefore it happens that one needs to use roundabout roads to make AI more tougher

There is nothing bad in it. Its unnecessary to create a worthy opponent. Making an illusion of one - is quite enough. And if you cheat while not being red-handed - you are riding high.

No,one doesnt need to use a roundabout to make AI tougher if they make it tough enough in the first place.Thats just lazy.

Oh,yes,guess what I voted.

[DaemianLucifer]

I think it is possible to make a quality AI for a game like Heroes. It just takes a lot more time than Ubisoft is willing to let Nival spend on it. Look at Galactic Civilizations 1 and 2. GalCiv is at least on par with Heroes in terms of complexity, but GalCiv's AI is leaps and bounds ahead of the AI in Heroes 5. Their AI is constantly being upgraded by the developers. As players post exploits and tactics on the game forum, the developers upgrade the AI to react.

Dont waste your breath.Hell just go on to say that galciv is a simple game and heroes is the ubercomplex god of a game.There is no point in arguing JJ on many points,and this is one of them.

[Sir_Toejam]

OT:

I hear GalCiv being bounced around quite often on these boards.

Never played it.

Is it really that good?

[DaemianLucifer]

If you like civilization type of games,it is.The only thing it lacks is the tactical combat,but it does perfect without it.As for its AI,it indeed is the best there is.

[fanta]

Hi boys & girls,

I've played Heroes 1 to 5, skipping Heroes 4 expansion packs. I must say that Heroes 5 AI is pathetic. You just have to play a few times and you get the hang of it. There's no desire to improve your style to beat the AI. Because you already beat it too quickly, too often, too boring-ly.

I love to play HOT-SEAT if the game doesn't take 4 minutes to end each turn. I love to play multiplayer, but, hey, notice how jerky/rough it is? Can't believe this when there's more communication going on in War3 / other good multiplayer games, yet those games are smoother than Turn-based H5.

It is Nival's responsibility for AI and the game's overall performance. But it is Ubi's responsibility to solidify the game concept requirements, i.e. getting it right, and not change them so much that the programmers can't catch up.

In Software development (of course including game AI), Senior Project Managers (UBI staff) are responsible to give clear guidelines and test scenarios. No matter how good and talented programmers are, the whole project will fail when the PM (proj. mgr.) keep changing the project's direction. Or changing the requirements. Or thinking that AI is not that important and leave it to near the date of release (earlier this year).

Put yourself into Ubi's and Nival's shoes and you know why this mess began...

I was thinking that DT or TT complained too much. That's quite true, but this AI mess is making me agree with some of their posts and that Ubi really stinks.

[Arqane]

While writing a good AI is difficult, it's not impossible. The reason why AI is still relatively horrible is that most of the resources spent on video games are spent on other things.

Games should be the front-runner that increases AI technology, since so many games use some form of AI. However, 90% of the resources in games for the past 10 years have gone to graphics (engine and artistry). You notice how so many other things in games tend to be horrible? That's because there's limited time, and graphics take up so much. You'll also notice that games with simpler graphics have much better AI, game-play, and more because you can devote resources to them.

Human AI really isn't THAT hard to program, and I'm doing a project on it now myself. It still takes up a good bit of processing power to have a really good AI, but again, if you don't devote 90% of your game to graphics processing, you've got some space. Personally, I have had enough of the eye candy, and want the focus to go back to the actual games and AI.

[zarakand]

Wow this is sad to hear. This game is horrible to play against the computer, even on heroic. It really poses little to no challenge, and that's a shame. The most fun games I've played are where I've won by just a thread or lost by just a thread. Heck, even when I get trounced I don't mind. It's challenging.

The point of the AI is not to loose gracefully, but to defeat me. On the lower levels you can compensate this by giving the human bonuses, not the AI.

What a shame this game is only playable multiplayer. I hope Nival gets dumped if they make a part 6.

[okrane]

Nival are just a bunch of stupid idiotic russian morons...

What does it have to do with them being Russian? It's very impolite to say things like this regardless of what they do.

didn't mean it as an insult... even if it was surrounded by them...sorry if it appeared that way... I'll edit that right away so there aren't any misunderstandings...

[DaemianLucifer]

Everyone knows that we east euorpians are lazy and evil krimals,its nothing new

[Sir_Toejam]

Wow this is sad to hear. This game is horrible to play against the computer, even on heroic. It really poses little to no challenge, and that's a shame. The most fun games I've played are where I've won by just a thread or lost by just a thread. Heck, even when I get trounced I don't mind. It's challenging.

The point of the AI is not to loose gracefully, but to defeat me. On the lower levels you can compensate this by giving the human bonuses, not the AI.

What a shame this game is only playable multiplayer. I hope Nival gets dumped if they make a part 6.

just like with H4, it will be the custom mapmakers that create fun scenarios to play that take into account exactly how the AI works.

If you wrote off h4 because of poor (heh, non existent!) AI, you would have missed some really great user made maps and campaigns.

I still remember that "Wind of Thorns" (IIRC) campaign for h4 - that was simply the best strategy campaign I've ever played.

[Killroyan]

Well I do have to agree that the AI will have a very hard time ever to beat a human in this game simply because the programming would be so complex to calculate every odd/spell/damage dealt. But some things should have been made better already like the autocombat that still does stupid things like rushing units in 1 square before the entire army and then gets trounced on. This would already improve the AI a lot.

And I also think that an AI should be made developped to win against a human. Give us the challenge. Making the AI to loose is like an insult to all players because we can't figure it out apparantly how to win ourselves.

About the cheating, I agree that it is necessary to give the AI an edge against us. We are at a distinctive advantage already. What the best way to cheat is is a good question. Seeing the whole map and starting with more resources are good cheats. Giving the AI 10.000 gold every day is a bit strange.

All in all they should improve the AI. Have heroes out there that can attack your castle if you leave it alone long enough. This will add the suspense again and keep you on your toes. Town portalling back in at just the right moment is fun. Getting a secondary hero that can defend is also good, etc......

[Alamar]

I don't necessarily think that the statements made by the AI developer are that outrageous ... meaning that I'm sure this concept is embraced by a lot of companies.

My guess is that the clear majority of sales of HoMM5 went to fans that wouldn't be considered to be "hardcore". Perhaps a majority might even be just casual game players .... I wouldn't be surprised if these casual game players DID want an AI that appeared to put up a struggle but then just lost gracefully in the end.

As for the "it's only cheating if you're caught" aspect I actually strongly agree with this approach [theoretically]. I personally believe that it would be quite difficult [impossible with budget, time, and computing constraints?] to make a non-cheating AI that would truly be competitive for the hardcore players. Given that I would just design an AI that would be competitive for a casual player [which is hard enough by itself!] and then let it cheat in hopefully subtle ways so as to not spoil the ILLUSION that you're playing against someone that ISN'T CHEATING.

The bad things about the H5 AI [IMHO] are:

1. The devs shot for the ABSOLUTE MINIMUM goals that a developer should use. IMHO [as a programmer] if your goal is to make something that is just barely good enough then the end result will almost NEVER be good enough.

2. The devs failed miserably at both stated goals. [lose gracefully & don't get caught cheating]

The AI in no way loses gracefully. If you struggle and can get past the initial salvos that the AI throws at you then it often just cowers in its castle and waits to die. This does not fulfill the LOSE GRACEFULLY goal that they were shooting for.

Second the "it's only cheating if you get caught" principle is shot in the head because the cheats in H5 are in no way subtle. The bonuses are blatant and it's clear that the AI is not even playing anything resembling the game that you are.

3. My personal belief that the AI should appear to be a human player that is perhaps simply a lot LUCKIER than you are, or perhaps better, is totally shot. The illusion that you're fighting something that is playing by the rules is non-existant ----- When you play the AI it's not even like you're playing HoMM any more.

[DaemianLucifer]

My guess is that the clear majority of sales of HoMM5 went to fans that wouldn't be considered to be "hardcore". Perhaps a majority might even be just casual game players .... I wouldn't be surprised if these casual game players DID want an AI that appeared to put up a struggle but then just lost gracefully in the end.

Thats what lower difficulties are for.Normal however should never aim at loosing,and harder should always aim for a win,no matter the means.