Playing on hexes: Notes from Battle for Wesnoth

[Ahriman]

2) To me there is a world of difference between the computer/player having a handicap in the for of a bonus/penalty that the other player does not have, and the AI cheating by simply ignoring some of game rules. (Like an AI playing with the complete map revealed.

I don't see a significant difference here.

 

[civ_king]

Civ IV is an infinitely (well more than a million times harder) more difficult game to make a brilliant AI for

 

[Ahriman]

Civ IV is an infinitely (well more than a million times harder) more difficult game to make a brilliant AI for

As an overall AI? Yes. As a tactical combat AI? Not really.

 

[Mercade]

Regardless of the quality of AI and how difficult it is to make a good one, I must say that I did enjoy the perspective that Battle for Wesnoth gave me on what a game on hex tiles looks like and how that affects game play in combination with the 1upt rule. I'm looking forward to the civ5 experience.

 

[Truronian]

I imagine defensive lines are something that will be easily programmable for the AI... defensive warfare will probably function quite well. Patching holes in lines and offensive warfare we'll have to wait and see about.

Wesnoth is a good game, but the units all move a lot further than those in CiV will. Many of the bad AI traits are a result of this.

 

[civ_king]

I'm pretty sure civ V will have a bigger developing budget than Battle for Wesnoth

 

[The_J]

A bigger budget than 0$? Yes, probably.


Oh, we'll have to introduce the line of doom and the circle of doom in Civ5 .

 

[Trias]

No, this isn't true at all, and there is no indication the AI in civ 5 will be close to this ideal. In civ4, some clevers modders made the AI perform multiple difficulty levels higher, by changing exactly one thing - part of the horribly broken diplomacy system (that the AI also "cheats" at). Making them ignore previous hard-coded preferences to like each other and behave at Pleased or Friendly or whatever.

Did that really make the AI play better in ALL situations. It clearly, made the AI play better on average, mostly by being less predictable to the Human player. It is very much possible that an island of isolated AIs would fare better without this modification. Note that I wasn't saying that it is impossible to improve the civ AI. This is clearly possible, as is proven by the Better AI mod, which currently outperforms the BTS AI by about one difficulty level. My point was that it is nearly impossible, to make multiple levels of difficulty in the AI, because the effect of a change in the AI on the difficulty is hard to gauge and seldom consistent along all possible game situations. (For example, a change that improves the AIs skill at early land wars, may decrease its effectiveness at late game naval wars.)

 

[Ahriman]

I think there are several points to be made here.

1. Tactical combat AI is different from the overall strategic AI. Tactical AI is much more linear; there *is* such a thing as a better tactical AI that is "better" in nearly all circumstances. Partly you can achieve this by broadening the set of variables that the AI looks at when moving its units.

2. Yes different AI leaders have different personalities; one might favor horse hordes, another might favor a big navy. Some of these might be more effective than others. But that alone isn't reason to remove this differentiation, because its *fun*.

3. I agree that in general (other than 2) what should be done is to make the tactical and strategic AI as good as possible, and then differentiate difficulty level using AI bonuses/player bonuses. The AI should always play to the best of its ability (exception, maybe the very lowest 1-2 difficulty levels, for the utter newbs); its no fun playing against an AI that deliberately acts stupidly because of a lower difficulty level.

4. Barbarians are an exception. AI civ factions should have good sensible AI. Its fine for barbarians to be overly aggressive and just charge at you and try to kill your stuff and pillage and burn and not preserve their units. Barbarians aren't supposed to win, unless you're really weak.

 

[Trias]

I think there are several points to be made here.

1. Tactical combat AI is different from the overall strategic AI. Tactical AI is much more linear; there *is* such a thing as a better tactical AI that is "better" in nearly all circumstances. Partly you can achieve this by broadening the set of variables that the AI looks at when moving its units.

Agree.

2. Yes different AI leaders have different personalities; one might favor horse hordes, another might favor a big navy. Some of these might be more effective than others. But that alone isn't reason to remove this differentiation, because its *fun*.

Agree.

3. I agree that in general (other than 2) what should be done is to make the tactical and strategic AI as good as possible, and then differentiate difficulty level using AI bonuses/player bonuses. The AI should always play to the best of its ability (exception, maybe the very lowest 1-2 difficulty levels, for the utter newbs); its no fun playing against an AI that deliberately acts stupidly because of a lower difficulty level.

Agree to some extent. Playing against a set of cutthroat AIs that will do anything to win is not that fun. The AI trying to win at any cost would, for example, lead to all AIs dog piling a player (Human or AI) that is about to win. This is OK for military victories, but completely removes the point of peaceful victory conditions.

On the other hand, the AIs basically ignoring the fact that a player is about to win cultural victory, as they do now in civ4 also isn't much fun. Some middle ground has to be found here. (For example, Better AI, increases the chance of an AI going to war with a player near to a cultural or space race victory, but still lets this depend on the player relations. AIs that are at friendly, still will not declare war for example.)

4. Barbarians are an exception. AI civ factions should have good sensible AI. Its fine for barbarians to be overly aggressive and just charge at you and try to kill your stuff and pillage and burn and not preserve their units. Barbarians aren't supposed to win, unless you're really weak.

Agree. Samething goes for city states.

 

[Ahriman]

Agree to some extent. Playing against a set of cutthroat AIs that will do anything to win is not that fun. The AI trying to win at any cost would, for example, lead to all AIs dog piling a player (Human or AI) that is about to win.

I should make clear; I'm talking about the economic and military AI. Not the diplomatic AI.
As has been discussed to death in previous threads, a diplomatic AI is designed to make the AI fun to play against, and to make it feel like a real country, not a deathmatch bot.
Real countries don't backstab their allies because the ally is about to "win".

increases the chance of an AI going to war with a player near to a cultural or space race victory, but still lets this depend on the player relations. AIs that are at friendly, still will not declare war for example

Agreed.

Samething goes for city states

Well, unclear. I think city states are supposed to really last. They're supposed to be more like mini-civs than like barbarians; they're there to be allies or bullied around diplomatically and to provide interesting friction with other civs, they're not there to be military speedbumps who are easily slaughtered.

 

[nomooon]

Player handicaps are not cheats. Playing with a handicap is a time honoured tradition for matching opponents of different skill. It is common in games ranging from Golf to Go. It is not like anybody would call playing on settler level cheating. (OK, maybe some people would)

I always find it surprising that find how little the AI in civ actually really cheat. There are some minor thing where the AI routines take short cuts. (For example, the AI can "see" blue circle in FoW) But most of those just level out the (experienced) player's advantage of understanding of how the iner mechanics of the game work. (For example an experienced player can guess the presence of resources near his capital from the absence of forests.)

This is an incorrect comparison. I know very little about Golf, but for Go, player handicap is about giving an upperhand to enemy when starting out. During the course of the game both sides obey same set of rules and there is no player handicap there! If an AI gets large chunk of free gold whenever its treasury runs low, then no matter how I blockade/pillage its economy, it will never go bankrupt! This would be different if the AI simply starts with 20k more gold but can still go bankrupt as time goes on.

 

[Earthling]

Did that really make the AI play better in ALL situations.

Close to it, yes. At least, if there was a human civilization playing in the game, the AI would always fare better with those changes.

Also, I disagree with that one viewpoint on what constitutes AI "cheating." Really nothing listed is different from other things listed, and the AI cheats just about as far as it possibly can. Seeing further into the fog of war is no different from free units or cheaper technology or whatever - they are all equivalent bonuses/cheats. Sure, the AI can't reload the game to an earlier state or something but just about every pertinent gameplay mechanic it can "cheat" on it does, there's not use getting angry about semantics. I think most people saying they want to see such bonuses reduced and just overall better AI in return are justified.

 

[EmpireOfCats]

After playing some more BfW, I must say there is one thing that annoys me no end and is done a lot better in Civ: Probability-driven battles. The first time four level two units don't even hit a lowly zombie, you shrug it off. The second time, you get angry. The third time, you reload the turn. Yes, the FAQ talks about the math, but this is simply not my idea of fun, and takes away a lot of the enjoyment of setting up ambushes and such. Civ does this right, and (please pretty please) should continue to to it right in Civ V.

(Oh, and when I'm rich, I'll hire somebody just to port a version of BfW to non-probability combat. What does Bill Gates do with all that money, anyway? Charity, yes, but what about the important things? Jeez.)

 

[Ahriman]

Probability-driven battles

???
Both Civ and BfW have probability-driven battles. And the probabilities are displayed in-game (though BfW displays the distribution of outcomes, which you only get in Civ through BUG).

There is a chance that your strength 8 unit will lose to their strength 6 unit.
There is a chance that your melee unit attacking their dark initiate won't do any damage.

[Also note in BfW; the probability of hitting doesn't really depend on your units; it depends only on the terrain. If the enemy is on 70% defensive terrain, then there is a ~6% chance that 8 successive attacks will miss. Ie more than 1/20th of the time.]

 

[Trias]

If an AI gets large chunk of free gold whenever its treasury runs low, then no matter how I blockade/pillage its economy, it will never go bankrupt!

Saves and/or reference to the source code or it did not happen.

Close to it, yes. At least, if there was a human civilization playing in the game, the AI would always fare better with those changes.

Also, I disagree with that one viewpoint on what constitutes AI "cheating." Really nothing listed is different from other things listed, and the AI cheats just about as far as it possibly can. Seeing further into the fog of war is no different from free units or cheaper technology or whatever - they are all equivalent bonuses/cheats. Sure, the AI can't reload the game to an earlier state or something but just about every pertinent gameplay mechanic it can "cheat" on it does, there's not use getting angry about semantics. I think most people saying they want to see such bonuses reduced and just overall better AI in return are justified.

So, in your opinion playing on settler difficulty level is cheating? Because that is effectively what the AI does at a higher difficulty. (With the remark that the difficult at which the AI is playing at Deity, would two or three levels below settler.)

There are some situations in which the AI actual does things that should be considered cheating. One is the increased sight range in fog of war. (Although to some extent this is meant to simulate the AI player making an educated guess on your units locations based on information from previous turns. Actually having the AI make an educated guess would be dreadfully slow.)

Another has to do with the AI seeing all "blue circles", even those in areas he hasn't scouted. This may actually be just a bug.

There are the only cases where I know for sure that the AI cheats, in the sense that he does something that he should not be able to do based on the game rules. There are probably a few more like this. I agree that in an ideal situation they should be removed. Although, there is always a trade off in performance. I'd rather have reasonable turn time and AI code that takes a few shortcuts, then have really long turns.

 

[EmpireOfCats]

???
Both Civ and BfW have probability-driven battles.

I probably phrased that wrong. What Civ IV doesn't do (anymore) is have, say, a tank get killed by an archer. In BfW, I'm seeing the equivalent -- seeing it a lot. For whatever reason, it seems a lot more random, and frustrating, in BfW.

 

[Ellestar]

Whereas with 1upt, a dumb AI strategy (eg like that in BfW, where they basically just attack-attack-attack) is massively outperformed by a smart human player. The difference is larger.

It's was an intentional decision to make AI behave that way in BfW. AI calculates vulnerability of it's position and you can mod AI behaviour to make it play safer.

 

[Ahriman]

What Civ IV doesn't do (anymore) is have, say, a tank get killed by an archer

Its possible. Its incredibly unlikely, but it has a positive finite probability.

In BfW, I'm seeing the equivalent -- seeing it a lot. For whatever reason, it seems a lot more random, and frustrating, in BfW.

If you're seeing it a lot... then its not the equivalent.
There is no equivalent in BfW of tank vs archer. They're all using the same level of technology. the game very clearly displays what the probability of outcomes are. It doesn't matter how "tough" your unit is, that effects the amount of damage it does per hit but not the probability of hitting.

I think if it "seems more random" relative to your expectations, then your expectations aren't correctly based on the (displayed in-game) probabilities.

Fwiw, I *never* see a single level 2 steelclad get killed by a single level 1 skeleton, on equal terrain.
If the skeleton is on 60% defensive terrain while the steelclad is on 20% terrain, well thats a different story.

It's was an intentional decision to make AI behave that way in BfW

Agreed, because the base AI isn't designed to try to put you on even footing; its designed for campaigns/scenarios where the AI is given high starting gold or a big army or particular scripts.
Which is different from the "competing among equals" of normal Civ games.

I haven't tried modding BfW at all though, so I haven't looked at any of the details.

 

[Frasco]

Real countries don't backstab their allies because the ally is about to "win".

This could be addressed by making AI's favor team-victories. If that AI knows a player can win and is friendly to him, it might as well try to win with the player instead of winning against the player, wouldn't it?