Question about AI Quality
- Jolly Joker
- Round Table Hero
- Posts: 3316
- Joined: 06 Jan 2006
DL, you misunderstand intelligence. Artificial INTELLIGENCE would have to incorporate the ability to LEARN. If that isn't there - and it isn't - the AI is AT BEST comparable to some dog doing some tricks the master taught him. Sure, if you see a trick at first it can look amazing. But it can never do something else than what the master taught.
I don't say that this might change some time in the future. But you won't see it in Heroes V, that's for sure.
I don't say that this might change some time in the future. But you won't see it in Heroes V, that's for sure.
- DaemianLucifer
- Round Table Hero
- Posts: 11282
- Joined: 06 Jan 2006
- Location: City 17
But not because its imposible,but because no one bothered doing it.At least no one of the developers,that is.There is a fan made mod for civ that makes the AI actually learn what you are using against it.Jolly Joker wrote: I don't say that this might change some time in the future. But you won't see it in Heroes V, that's for sure.
- Jolly Joker
- Round Table Hero
- Posts: 3316
- Joined: 06 Jan 2006
- DaemianLucifer
- Round Table Hero
- Posts: 11282
- Joined: 06 Jan 2006
- Location: City 17
- MistWeaver
- Wraith
- Posts: 1277
- Joined: 28 Feb 2006
- Location: Citadel of Frosts
Besides money cheats, lets not forget about thing that AI has special quickbattle that prevent it from taking losses in small & moderate fights (while taking extra small losses in major fights).
AI can never lose to neutrals. And total army of two starting heroes can beat all neutrals around on 1st week.
BUT at the same time AI can attack human and LOOSE quickbattle. How sweet.
And keep in mind extra luck bonuses for AI. (and may be some more that im not aware so far)
AI can never lose to neutrals. And total army of two starting heroes can beat all neutrals around on 1st week.
BUT at the same time AI can attack human and LOOSE quickbattle. How sweet.
And keep in mind extra luck bonuses for AI. (and may be some more that im not aware so far)
- Jolly Joker
- Round Table Hero
- Posts: 3316
- Joined: 06 Jan 2006
- DaemianLucifer
- Round Table Hero
- Posts: 11282
- Joined: 06 Jan 2006
- Location: City 17
If you use a strategy it learns from that strategy and comes up with a way to counter it. For instance if you 'zerg' (rush) it, after a time it learns this and the next time you try to zerg it, it has a defense ready that crushes you like a bug. If you always sit back and build, it learns from that and rushes you so you don't have time to (giving H5 examples of similar things). If you rely on one type of unit, it builds the counter to that unit quickly.
It would be like if the h5 AI could see into your skill wheel. If you always took a set skill list, it would take a skill list that countered yours effectively. If you built your town in a certain manor, it would build a town that could effectively counter that build. Things like that. That technology does exsist, to make an AI learn from the human opponit. Take the chess computer, it learns from it's opponits to better itself. It is not unbeatable, but you can't play it the same twice or you will loose.
It would be like if the h5 AI could see into your skill wheel. If you always took a set skill list, it would take a skill list that countered yours effectively. If you built your town in a certain manor, it would build a town that could effectively counter that build. Things like that. That technology does exsist, to make an AI learn from the human opponit. Take the chess computer, it learns from it's opponits to better itself. It is not unbeatable, but you can't play it the same twice or you will loose.
Warning, may cause confusion, blindness, raising of eybrows, and insanity.
- Jolly Joker
- Round Table Hero
- Posts: 3316
- Joined: 06 Jan 2006
I agree that this is the right way to go at it, IF the AI can be made to use the learned things well which is of course the next problem. Prerequisite is, however, the ability to notice every step of the player (at least in Heroes) which would actually a very high level of cheating.
To solve this you'd need a record and analyze function in the game, a rather ambitioned thing to have.
To solve this you'd need a record and analyze function in the game, a rather ambitioned thing to have.
- DaemianLucifer
- Round Table Hero
- Posts: 11282
- Joined: 06 Jan 2006
- Location: City 17
Yes, the skill wheel thing was only an example, no way should the computer be able to peek there. But noticing if you are a builder, rusher, explorer, or whatnot is well within current technology. In fact it has been arround for a long time. The military use simulations to train their soldiers for years, and those similations learned from the encounters in order to make those encounters harder. By years, I mean a lot of years. Now even games use this to an advantage. Not many, to be honest, but more then a few. This is how some games increase your skills the more you use them (the computer remembers how you use them, and increases the strength accordingly). It is possible, but maybe a bit cost intensive for now. I don't know.
Warning, may cause confusion, blindness, raising of eybrows, and insanity.
- DaemianLucifer
- Round Table Hero
- Posts: 11282
- Joined: 06 Jan 2006
- Location: City 17
The only thing the AI cant do now is compare parts of strategies and combine them into something new like the human can.So if the AI sees you are using one strategy,it will use a counter strategy it learned.But if you change your strategy in the middle,youll confuse it and end up victorious.Inovation on the go is the only thing that separates the human from the machine.
After reading over the thread it seems like a lot of counter-arguements to some proposals are along the lines of "boy that's cheating a lot" or "that's a cheat HoMM has never done".
As I've said many times I think that the AI will need to cheat in order to stay competitive with a human player.
However certain cheats are much more subtle than other cheats. [and therefore more acceptable to me and probably other players]
For example at the end of week 5 it should be "impossible" for a human to have a fully built castle AND have bought out all creatures [fully upgraded]. This sort of a cheat is a slap in the face.
On the other hand if we ran into an AI hero that was nearly perfectly constructed skill & perk wise then I have no problems with that because [on the face of it] it looks like something a human playing by the rules could have pulled off.
-------------------------------------------------------
On the topic of "look ahead" cheats I don't have too much of a problem with those sorts of things as long they are subtle.
For example I have little problem with the AI running its combat estimators [with exact creature numbers] to figure out how many losses it would take and if it is worth it to fight creatures. I do almost the same thing when I'm playing except I just don't have exact info ....
I also don't have too much of a problem with the AI looking ahead and knowing [before time] what spells will be in its spell tower so it could build heroes more appropriately.
On the other hand I do have a problem with the AI knowing what creatures will and won't join their army and the formula it uses for that because the AI isn't very subtle in its use .... Seeing the AI avoid several stacks of creatures, mines, etc. making a bee-line for some other creatures just to get a join and see it start heading back to your castle to attack you is at best annoying.
As I've said many times I think that the AI will need to cheat in order to stay competitive with a human player.
However certain cheats are much more subtle than other cheats. [and therefore more acceptable to me and probably other players]
For example at the end of week 5 it should be "impossible" for a human to have a fully built castle AND have bought out all creatures [fully upgraded]. This sort of a cheat is a slap in the face.
On the other hand if we ran into an AI hero that was nearly perfectly constructed skill & perk wise then I have no problems with that because [on the face of it] it looks like something a human playing by the rules could have pulled off.
-------------------------------------------------------
On the topic of "look ahead" cheats I don't have too much of a problem with those sorts of things as long they are subtle.
For example I have little problem with the AI running its combat estimators [with exact creature numbers] to figure out how many losses it would take and if it is worth it to fight creatures. I do almost the same thing when I'm playing except I just don't have exact info ....
I also don't have too much of a problem with the AI looking ahead and knowing [before time] what spells will be in its spell tower so it could build heroes more appropriately.
On the other hand I do have a problem with the AI knowing what creatures will and won't join their army and the formula it uses for that because the AI isn't very subtle in its use .... Seeing the AI avoid several stacks of creatures, mines, etc. making a bee-line for some other creatures just to get a join and see it start heading back to your castle to attack you is at best annoying.
From my perspective, it's not about how hard the game is or whether the AI starts with more or less resources. It's about how interesting the game is. The problem with an AI cheating and playing like the one in Heroes V is that it lacks a lot of the aspects of playing the game and therefore makes less of an actual opponent. I need an AI to do all the things a player would do. If its play is simple and is hardly affected by what I do, I'm not much better off than I am fighting neutral creatures. That's why I like the AI to get advantage by getting more mines, capturing more towns/castles, pulling off some smart moves. If I leave a heavy stack of neutral creatures to guard one of my entrances, I don't like to AI to be able to take it without losses using a small army. If I leave one of my castles to fend for itself because I know I can outproduce the AI, I don't want th AI to be able to somehow get creatures more easily. If I capture all castles quickly, the AI should be desperate to catch up. The AI and I are supposed to be playing the same game for me to relate to what it's doing.
Hell, I always make the map myself and decide at what terms the different AI players and I start. So, whever advantage either side gets doesn't really matter. I don't care if the AI defeats me. I want it to be responsive, interactive and balanced. People talk as if the ones praising the Heroes 3 AI have "collective memory loss", but I don't think anyone really remembers liking something they didn't. Technically, I'm sure some people would be blinded about the AI capabilities. No one likes Heroes IV og V AI, but Heroes III AI always kept things interesting and did a lot of moving around on the adventure map and kept pressuring. That way you always felt that you were exploring and chancing into the unknown. No matter what flaws it might have had it was able to make you feel like you had an opponent and whatever action you made played into the entire action of the game.
Hell, I always make the map myself and decide at what terms the different AI players and I start. So, whever advantage either side gets doesn't really matter. I don't care if the AI defeats me. I want it to be responsive, interactive and balanced. People talk as if the ones praising the Heroes 3 AI have "collective memory loss", but I don't think anyone really remembers liking something they didn't. Technically, I'm sure some people would be blinded about the AI capabilities. No one likes Heroes IV og V AI, but Heroes III AI always kept things interesting and did a lot of moving around on the adventure map and kept pressuring. That way you always felt that you were exploring and chancing into the unknown. No matter what flaws it might have had it was able to make you feel like you had an opponent and whatever action you made played into the entire action of the game.
- winterfate
- Round Table Hero
- Posts: 6191
- Joined: 26 Nov 2006
- Location: Puerto Rico
Asjo: My thoughts exactly . IMO, Heroes 3's AI was, is (and until Nival fixes 5's AI), and will be the strongest AI in the series.
The Round Table's birthday list!
Proud creator of Caladont 2.0!
You need to take the pain, learn from it and get back on that bike... - stefan
Sometimes the hearts most troubled make the sweetest melodies... - winterfate
Proud creator of Caladont 2.0!
You need to take the pain, learn from it and get back on that bike... - stefan
Sometimes the hearts most troubled make the sweetest melodies... - winterfate
- Jolly Joker
- Round Table Hero
- Posts: 3316
- Joined: 06 Jan 2006
Can someone please tell me where the H 3 AI was strong?
1) Hero developing: Zilch. The AI favors all the crappy skills the players avoid like Scouting, Eagly Eye and so on. You could say the AI made it an art to create the worst heroes possible.
2) Battle AI. Bad. No grasp of the waiting command. Silly preferences. Easily tricked into attacking single level one creatures when it could have wreaked so much havoc. Extremely exploitable. Because of 1 Heroes have no mass spells. Spells actually cast are a joke.
3) Sieges. Incapable, as attacker AND as defender. Give the AI a defending hero with Tactics and have a good laugh before the siege starts. No concept of moat damage. No concept of even using the walls.
4) Adventure AI. Easily exploitable and tricked. No concept of the fact that visiting and garrisoning forces are being unified when the town is attacked.
So what was good?
1) Hero developing: Zilch. The AI favors all the crappy skills the players avoid like Scouting, Eagly Eye and so on. You could say the AI made it an art to create the worst heroes possible.
2) Battle AI. Bad. No grasp of the waiting command. Silly preferences. Easily tricked into attacking single level one creatures when it could have wreaked so much havoc. Extremely exploitable. Because of 1 Heroes have no mass spells. Spells actually cast are a joke.
3) Sieges. Incapable, as attacker AND as defender. Give the AI a defending hero with Tactics and have a good laugh before the siege starts. No concept of moat damage. No concept of even using the walls.
4) Adventure AI. Easily exploitable and tricked. No concept of the fact that visiting and garrisoning forces are being unified when the town is attacked.
So what was good?
- DaemianLucifer
- Round Table Hero
- Posts: 11282
- Joined: 06 Jan 2006
- Location: City 17
Who is online
Users browsing this forum: Google [Bot] and 9 guests