recalling units changes scenario difficulty?

[julie]

I'm playing Heir to the Throne (medium difficulty) and am on the "Isle of the Damned" alternative scenario. On my first attempt I came very close to killing both enemy leaders; so close that I figured it was worth trying it again from the beginning, with a few strategy tweaks. On my second attempt I then discovered that I could recall mermen from my previous scenario, which of course I did, as they're loyal.

I found the scenario much more difficult the second time around; more enemy units, targeting mine better. So I'm having to try it a third time and am convinced something has changed to make this harder.

My question is: was recalling the mermen the thing that did it? I know nothing about game coding but my fuzzy logic says that the computer thinks I'm now a "stronger enemy" because I was able to recall a level 2 and assortment of loyal units... If that's the case I'd like to know so I can go back to playing how I did the first time, not recalling anything.

Thanks!
Julie

[Dunno]

As far as I remember, recalling does not affect the gameplay in any way. Maybe RNG simply pranked you

[julie]

Well, my observation was more that the enemy leaders are creating more units, which are doing a better job at targeting my units (rather than being easily distracted by a merman offshore, say). I wondered if it was just perception, but something more than RNG has to account for sailing through (in comparison) the first attempt, and being bombarded north and south in the second and third. Hmmm! ^^

J

[beetlenaut]

For recruiting, the enemy AI doesn't look at the level of units you use, but it does look at the type. If you use a lot of one type, the computer will try to choose units that are good against that type. If those units happen to be cheaper, the AI will end up with a couple more units, but it doesn't receive more gold based on what you do. (Also, you would have had fewer units the second time because recalling is more expensive than recruiting.)

When it's time to fight your units, the AI does look at their level. The offshore merfolk would have been poor targets because they had more hitpoints than the land units. That's why the computer wasn't distracted by them--it went for the easy pickings.

[tr0ll]

related question: does the AI know what you are recruiting if the fog is on?

[Iris]

The AI currently ignores fog and shroud, yes.

[nuorc]

The AI currently ignores fog and shroud, yes.

Why?

And I think it's an unfair advantage for the AI.

[ancestral]

The AI currently ignores fog and shroud, yes.

Why? And I think it's an unfair advantage for the AI.

I would have to agree… in most RTS games, only the highest difficulty cheats. Why should Wesnoth cheat?

[Crendgrim]

Please, go ahead and write an AI which plays reasonable intelligent and does not ignore fog and shroud.
I think that is the big problem: The current Wesnoth AI is just not able to know which units to recruit and which not if it doesn't know what you recruited. This especially applies to non-mainline eras where it would have to recruit heuristically and try to make the best out of it.
The same problem occurs when looking at maps: An experienced player knows where his enemies' units are most likely to appear first – but how is the AI supposed to know this for all kinds of different maps, maybe even with custom terrain or events?
Having an AI which doesn't cheat would make the game vs. the computer way too easy, not to mention that it would destroy all singleplayer balance. However, having an AI that doesn't cheat in the first place is more or less impossible. Just imagine an AI on Aethermaw and think about how it should play it.

[Jozrael]

In an ideal world, the AI would not cheat. The AI we have now needs to cheat to maintain its current level of effectiveness (which already needs lots of help from other large advantages). AI coding is a very difficult and complex subject, and Wesnoth could always use more people expert in that discipline. If you know someone who could help create an AI where this was unnecessary, by all means refer them here!

[Daravel]

If you know someone who could help create an AI where this was unnecessary, by all means refer them here!

Wouldn't that be you? I haven't tried out the new and improved bots yet though.

[Atz]

Why [does the AI ignore fog and shroud]? And I think it's an unfair advantage for the AI.

I would have to agree… in most RTS games, only the highest difficulty cheats. Why should Wesnoth cheat?

Untrue. The AI in most RTS games will be given free resources on any difficulty above normal, and in some cases on every difficulty above the easiest. Regardless of difficulty, in the vast majority of games it will also know the position of all units on the map at all times, regardless of fog of war or any other form of concealment, though it may not be allowed to attack them as a measure to obscure its omniscience.

The reason AIs normally see everything is that it's extremely to program an AI that can account for unknown data and (in an RTS, at least) do so fast enough to keep up with the game. A human can guess what forces their opponent has and what they are doing by estimating their income, comparing it with the number of units they do see and where they are, what sections of the map they don't have sight on, and so on. However, this is extremely complex and creating an AI that can do something which humans learn to do intuitively through experience is very difficult.

The same problem occurs when looking at maps: An experienced player knows where his enemies' units are most likely to appear first – but how is the AI supposed to know this for all kinds of different maps, maybe even with custom terrain or events?

Modern strategy games already have a solution to that - when you make the map you just put tags on it which tell the AI where the enemy units will likely appear, where good spots to attack and defend are, and so on. In other words, you make humans do the thinking for the AI in advance. Unfortunately this can also lead to exploits because it tends to make the AI very predictable. In C&C Generals, for instance, the AI would always attack along certain routes which were marked for it in the map editor, which could easily be exploited by showering them with artillery and/or building a million turrets in front of them.

[ancestral]

Am I in the minority who think the Wesnoth AI is plenty or even overly effective on lower difficulties in scenarios?

[artisticdude]

Am I in the minority who think the Wesnoth AI is plenty or even overly effective on lower difficulties in scenarios?

No. Lower difficulties are just about all I can handle.

...And even then, I still manage to lose more often than I probably should...

[nuorc]

Probably there's an more appropriate thread for this, but anyway...

I like BfW and since the ai is an integral part of it, I'm thankful for having it and the people that created it. Personally, I don't know shtuff about coding/ai. Having said that...

1. The ai often appears to fight suicidally, which is fine if your slaughtering your enemy, but it sucks if you're having a needed ally doing mass-suicide. Why can it choose a favorable terrain if available, but often fails at withdrawal or concentrating the fire to kill a unit instead of spreading damage, making it ineffective?

2. How does the ai determine if the leader attacks? Sometimes the leader comes forward boldly, but sometimes it refuses to deal damage even if there's no chance for retaliation damage. An annoying example is a retreating ai-ally with powerful ranged attack that only does one useless step but doesn't kill one of those powerful melee-units cornering it...

3.

(an AI) would have to recruit heuristically and try to make the best out of it.

Isn't that what humans do until they see the enemy? What's the problem with that? Can't the ai be told to recruit scout, offense/defense, melee/ranged? And recruit more informed after seeing the enemy?

Just asking.