AI quality (or lack of it)

[timek28]

HI guys!

As much as I like playing PC, the late war scenarios are over and over convincing me that AI is just sub par in certain segments. I understand as a programmer that it is hard to write quality AI, but I'm not sure if what I want to suggest is impossible to implement.

I think that brute force method that AI uses is really simple, and probably not too hard to implement. Basically AI throws tanks after planes and in range artillery first even in close terrain (unless he is certain that he will get horrendous losses). He ignores potential for rough defense, and excludes support unless it is in firing range of the target. In this way AI has to throw much more resources than he actually needs, he looses tanks foolishly and completely underutilized artillery which is the last unit to come on the battlefield (while it should be other way around).

This also produces crazy results such that IS2 of 14 SP gets blown apart to 5-6 SP from 2 star infantry in close rough defense. While it is realistic, it is also hard to believe that any tank would run repeatedly into close terrain without any support.

While I like this as it is predictable and makes huge assaults survivable with not too many losses, it is just boring sometimes. Of course if AI did use artillery as it should, things would be harder for player but that could have been leveraged somehow, or not leveraged at all.

Be it as it be what are your comments about this issue?

I know it will probably not be fixed as it is probably too much work, and 1.21 patch is already coming out...

[Horst]

Sometimes, and only sometimes, you notice in each of the AI's turn that it already has a kind of pre-artillery phase when it fires first before anything else. This pre-phase progress bar completes very quickly that you often don't notice it. Why this phase doesn't happen all the time and makes artillery often fire more or less at last, can only Rudankort explain (or not). I like to know at least too.

[timek28]

Yes pre-artillery phase actually can be seen almost in any turn, if AI has lots of units. But rarely any artillery is in range to fire. This is especially visible in mass attack scenarios that Russians employ late in the war. Simply no Katyushas or SUs are close to the front line and they just linger in back with no usage whatsoever. On the other hand if they where near front they would be deadly.

Instead they should get as close to the front as possible just behind the tanks in the first front line and help them out so they are not pummeled in suicide attacks. So basically the logic would be that artillery has priority place near the front line in their firing range (2 or 3). This would be different from current tactics of units just massing one after another and getting destroyed or surrendered in waves, while artillery arrives last.

In this way AI would have a artillery line behind their front line and it would simulate the player who (should) never leave his front line without support. This would undoubtedly make game harder, albeit winnable (with maybe slight prestige modifications). This would also make game more fun as player would have to try to flank attacks and destroy artillery from air and behind just as he does when he is the one attacking.

[Horst]

My guess is that it also highly depends on the chance to inflict damage that often artillery won’t fire in its pre-phase. A little more generous waste of ammo couldn’t hurt here, because even a minimum chance of 5% or less often inflicts a lot of damage, especially with overstrengthed artillery. It almost looks like artillery units fear for retaliation or running out of ammo. The possibility to simply suppress enemy units seems to be overlooked here.
What I’ve observed that in the normal phase, the AI still fires too before it can move to a different position. It will fire more often then compared to when it simply stays somewhere.
In my modified game, many artillery pieces can’t move on their own and have weaker HA than in vanilla. Maybe that’s a reason why I often don’t see artillery in the proper firing order. I guess it only fires at last then when allied tanks and other units already did some damage and reduced entrenchment in particular with the new 200% defense bonus vs. ranged combined with +4 defense bonus in close terrain with v1.20.
I really need to add a couple of transports to towed artillery on many GC maps in the future, what is lesser an issue in other player-made campaign mods available here.

It depends on the scenario if artillery should participate in an assault or not. Sometimes, it can happen that the AI behaviour is accidently set to Hold Position or won't be affected by a script command. If you wonder why certain units didn’t attack, then check out with the scenario editor how their behaviour is set manually and/or by script.
Anything with Hold Position will typically sit around all the time, and if it’s even HP (passive) then it won’t even fire when a 1-strength transport unit parks besides them. Many AI infantry defenders are set on HP (passive) in GC scenarios, sleeping on city hexes. HP (Fire) as AI behaviour would be often a better choice in my opinion. The problem of HP (fire) and (passive) is whenever an AI unit is pushed out of a victory hex, it won't be able to occupy it again. This only works for behaviors like Default, Attacker, Defender and Hold Position (Active), however, these could possibly cause some silly move actions again, what a unit shouldn't risk if it is suppose to defend a victory hex.

[timek28]

Hi Horst

I believe you are creating scenarios on your own since you know all the possibilities this well. I cannot say that I understood everything you said, but obviously game mechanics is much more complicated then I previously thought...

You mentioned that some attacks shouldn't involve artillery. Well theoretically yes (clear ground assaults on infantry or soft targets, chasing the enemy etc...), but if I created a perfect armored front defense, AI should do a much better job of trying to penetrate it using artillery barrages first, instead of throwing his tanks into oblivion. I remember that there was a scenario in GC east, where there were about 20 Soviet artillery pieces across the river from my units right from the beginning of scenario. This was a perfect (and realistical) placement. I remember having a hard time trying to neutralize those batteries and trying to stop Soviet advance since pre-artillery phase would fire 20 pieces suppresing MY artillery and flaks, and even tanks. This is how the game should look like.

Mechanics should be also done in this way due the fact that German armor is much better then allied. AC proves that in a VERY obvious way. The idea is not to attack with allied tanks head on, but to try to destroy german AA and planes in order to bomb the tanks from the air and with artillery first, or force them into close terrain. At least that was the only way for me to win AC campaign. AI tries this but in a very weak way. He should try to bomb flak guns with artillery, move his SPAAGs and try to weaken my fighters and then mass attack my fighters with his planes - as he usually has at least 2 times more planes then I do (even if that causes him big losses). Instead he never attacks player planes if they are overpowering, so AI looses air supremacy very quickly, and from there on he is more or less doomed since the only feasable way of breaking steel wall in German front would be repeated bombing of German armor, and only then advancing with tanks.

[the_iron_duke]

I've started playing the single-player campaign again (on max A.I. level) and was just shown a couple of examples of terrible A.I. decision-making:

1) A tank attacking infantry in close terrain (losing 5 and killing 1).

2) A cavalry attacking at the beginning of their turn, going down to 1 strength and then not using their movement phase to retreat to safety.

This is why I find pitting one's wits against a real human opponent in multiplayer a much better test of one's tactical capabilities.

[rezaf]

Well, the later dlcs are for the most part not about testing ones tactical wits anyhow, the only technique used in scenario design is "moar russian units!!!!!!111!".
While this is, of course, historical, it also nicely shows off the fatal flaws in the AI - if it had a shiver of intelligence, the player would not stand a chance in many of those scenarios.

The "best" flaw isn't even the artillery failing to get used in time, imo, it's the reluctance to attack if the odds are bad. In all these east-front defensive scenarios, it's the best tactic to have a handful of 5 star tanks, preferrably tigers, and just park them in the way of the offensive. This will work like a literal dam, only the very best russian tanks will even dare to attack, all the others will pile up in front of the blockade to be taken out one after the other. If the weather permits, I like to take out the AA with long range artillery and tacs and then just let the masses of units be a feast for my airforce.

Mentioning airforce, another fatal AI flaw is how the AI loves to strafe with it's fighters. In face of concentrated Luftwaffe fighter formations, it should try to break air superiority by ganging up on fighters, instead I've sometimes see then even ignore unescorted Ju87s in favor of strafing a piece of artillery or even a Panther.

These scenarios should be hard to play, with the gazillion of russian units in repeated offensives, but instead they are fairly easy - I often managed to defeat all forces on a map with some turns to spare, and I can't consider myself a very good player. The only parts where it's not only hard but impossible to hold the line is the areas where you are given sub-par german or allied (romanian etc.) troops to hold the line somewhere - what you get would struggle to defeat the polish in '39, vs. the russian juggernaut it's a hot-knife-butter-situation. Also at least partially historical, but not terribly fun, until you sent a couple of Panthers over to repeat the stem-the-tide tactic described above.

That said, it's probably for the best, because what I dislike most about those DLCs is how all your victories are in vain anyhow, so fighting even harder to achieve nothing wouldn't be that thrilling after all.

TLDR: The AI is still dumb as a rock, but at least scenario scripting was vastly improved as to better hide this fact. But don't you look closely...
_____
rezaf

[FOARP]

Arguably there isn't really much of an AI in the game - in most of the major co-ordinated maneouvres the computer seems to be responding to scripted triggers, with simple rules controlling how the units respond to enemy forces within a few hexes of them. It's only when the job of the AI is to attack that this becomes glaringly obvious though.

[rezaf]

There's no such thing as an artificial intelligence worth bearing that name.
So, in all games, even those with AIs perceived as pretty good (are there any?), it's all just a matter of scripted reactions.
The scripting in PG and, subsequently, in PC seems to be especially limited, though.
_____
rezaf

[timek28]

@razaf

You are 100% right. DLC is great and it's late war scenarios are glorious and atrocious in the same time. The amount of Soviet units I destroyed by simple dam tactics you described is beyond comprehension. And as you say they just come again and again to be swiped away repeatedly. This gives me as a player sweet feeling but on the other end it just becomes boring after a while and also it becomes a tedious task of repeating the same thing over and over. I also agree on the fighters and planes in general.

Of course if it was any different with this AMOUNT of Soviet units (and their prestige) as you said player would have little chance to win. That is due the fact that developers probably didn't want to change AI from PC version 1, and just realized that the only way that AI has a chance against steel wall of 10 5 star Tiger2s is to throw 5 times more tanks than player has. That is typical brute force algorithm.

Since we recognized the problem, the question is what developers can do about it (if they ever decide fix this thing)? IMO there are several things that can be corrected and that actually lead to this problem:

1) Air supremacy one way or another is the key in the game (as it is in real life). AI fails to recognize this. In most scenarios it has numerical superiority (or even superiority experience wise) but it fails to materialize on that. In 43-44-45 DLC scenarios Soviets (and Allies) have way more planes then Luftwaffe but they get blown away repeatedly. This is mainly due the fact that AI doesn't even try to attack Luftwaffe fighters that are experienced and fail to recognize range of German flaks (flaks are too strong BTW). Even if AI has 10 3 star fighters and player has 2 4 star fighters AI won't attack player's planes at all - trying to at least cripple them (maybe I'm not exact on the parameters but surely they are something like that). AI would probably get battered in the process but eventually it would give a hard blow to the Luftwaffe, one LW could hardly recuperate from (as planes gain experience very slowly). If AI did that that would open up lanes for bombardment of experienced Panzers - the only feasible way of defeating German steel walls.

2) Artillery usage. AI has to PUT ARTILLERY INTO FIRING POSITION FIRST (SP units) then fire and then attack with bombers, tanks, and infantry. The way it works now is that AI fires only artillery in range which makes 90% of it batteries completely unused. I think this is not that hard to implement (just drive SP arty 2-3 hexes from enemy front line, while first line is held by tanks). This could mean a lot of suppressed German tanks and force some surrenders.

3) As you razaf said AI tanks don't attack if the odds are bad (I've seen opposite to be true also esp in forests and other close terrain - not in cities). So AI tanks go from not attacking at all to attacking well entrenched positions and getting blow apart. This is just rubbish. If the point 2 is implemented right AI tanks could attack suppressed targets (just like human does in order to avoid losses), and create one shot kills (or at least not loose units like a fool).

4) AI should remember where ambushes are. If AI unit falls into ambush and gets destroyed there is nothing stopping other AI units from repeating the same stupidity 10 times over.

5) Rivers. AI fails to cross rivers effectively. They just don't know where the gaps and weak points are, and fail to put bridge engineers on those gaps and exploit them. They try to cross in the same German strong points again and again. Final Berlin scenario is maybe the biggest proof of this obnoxious "feature". As well as Bagaration and other scenarios where AI is supposed to cross Don river but fails to do so countless times.

Maybe there are some other points but I forgot them.

So that could be fixed. But if this was fixed then number of AI units and it's prestige should be tweaked not to be 5 times that of player, as in that case player would simply have no chance. Number of units are maybe correct historically right now, but we still need to realize that this is just a game and AI should win by showing some intelligence and not win just by having 10 times more units then player.

Or simply we should play against each other and forget about AI improvements altogether...

[rezaf]

Since we recognized the problem, the question is what developers can do about it (if they ever decide fix this thing)?

At this point, I'm pretty sure the answer to the unspoken question whether the developers will "fix" this is most definately "No".
First and foremost, there's a lot of official content out that - as I remarked above and you acknowledged - if the AI was significantly better, this would cause extreme difficulty spikes and make some scenarios unplayable or at least unwinnable without major redesign. That alone is reason enough not to make fundamental changes, before even thinking about the amount of work involved for a game that probably had the majority of it's copies sold.

About the specific points:

1) Air supremacy

Here I see a big issue: The fun the player can have has to be kept in mind, especially considering the RPG-like "levelling up" of units that a typical core will benefit from.
If the AI would make smart and consistent decisions to break a players air supremacy, it'd be inevitable for the player to lose many of his well-nurtured aces - which would be quite historical, but isn't fun at all.
A balanced approach would be to have something similar to the reform units feature, but much more powerful (you lose too much in the reform feature as it is now). For the airforce, I can see something like that working, but it'd require quite some changes - and it remains to be seen whether losing air superiority would not immediately lose the scenario for a player, which also wouldn't be fun.

2) Artillery usage

Same as above, if the AI was consistent about this, they'd mop the floor with player units, especially INF. You'd constantly lose units, concentrated artillery barrages can be very powerful.
If the unit isn't outright destroyed from shell fire, it'll be supressed, making it easy prey for ground assaults. Also, as you wrote, surrenders.
Once again, historical, but completely un-fun.

3)

Basically answered in the reply to 2). Capable AI introduced in the scenarios as they are would lead to frustration and misery for most players.

4) AI should remember where ambushes are.

Heh, hard to argue about that. The AI DOES remember them in that same turn, but will happily run into the same ambush next turn instead of scouting.

5) Rivers.

Yeah, this one could be fixed and might lead to some more interesting emergent scenarios, but on the other hand, mandatory river crossings for the AI are quite rare, so it wouldn't affect many scenarios.

So that could be fixed. But if this was fixed then number of AI units and it's prestige should be tweaked not to be 5 times that of player, as in that case player would simply have no chance. Number of units are maybe correct historically right now, but we still need to realize that this is just a game and AI should win by showing some intelligence and not win just by having 10 times more units then player.

Well, it was so much easier giving the AI tons of units. Let's face it, the DLC campaigns were an afterthought, and it shows. It's great that PzC had more success than anticipated and the DLCs, AK and AC were released, but it becomes rather clear how scenarios of this scope were not the focus of "core development" regarding AI and features.
Personally, I hope we'll get PzC2 one day - PG also started with humble, simple mechanics and did expansive changes in the sequels, some good, some ... not so good.

Even then, though, we mustn't forget PG (of which PzC is a pretty faithful clone) used to be called "beer & prezels" wargame - it's gameplay is the wargame equivalent to clay pigeon shooting.

Or simply we should play against each other and forget about AI improvements altogether...

With the right people, MP is always the superior experience - though personally, I play SP and MP for different reasons, i.e. I get different things out of it.
_____
rezaf

[Tarrak]

Personally, I hope we'll get PzC2 one day - PG also started with humble, simple mechanics and did expansive changes in the sequels, some good, some ... not so good.

Be careful with your wishes. While i certainly hope for an improved PC2 which can fix some of the current problems and maybe expand it a bit one need to be careful to not end in the same way as PG did: Introduce so many changes and "improvements" to totally change into something totally different and, arguably i admit, worse.

[timek28]

Well OK rezaf you maybe made your point here:

"Even then, though, we mustn't forget PG (of which PzC is a pretty faithful clone) used to be called "beer & prezels" wargame - it's gameplay is the wargame equivalent to clay pigeon shooting."

I play this game for fun as you said and I like the pigeon shooting mechanics that you described (gives a nice feeling). So you argued well that all the changes that I mentioned would probably ruin the fun. But then again I said that it could be balanced out by (way) less AI units, less AI prestige and similar. If AI played nearly close as (good playing) human he wouldn't need more then 10-20% more units then human (as well as prestige), and the game would stay as fun as it is (or even may become more fun)!

This would of course raise a need for complete scenario overhaul and that is probably why it will never happen. Game is now up for modding only it seems. The only reason why I suggested all of this is so the game could be more dynamic, and not scripted and limited as it is. But I believe this is the issue 90% of games experience generally. Some mod creators created some smart scripts that introduce dynamics into game (unexpected counterattacks, rear partisan or paratroop activity etc), but still it does not bring anything new with current AI.

But hell if most of the players like it that way then be it, I won't argue about that

[shawkhan]

I think the game works fine for the Axis. They are supposed to be outnumbered by up to 3 to 1, against tactically less skilled opponents. As the Allies though the game falls flat. Outnumbering Germans Should be the only way to beat them. For people thinking the game is too easy there are scores of ways to balance it. One way is to simply restrain yourself from using the uber units. Playing with all your armor Tigers is ridiculous from an historical viewpoint. When humans play games like chess it is customary for the better player to give a handicap, pawn,rook, or even queen. Shouldn't the poor dumb computer be given the same consideration?

[rezaf]

While i certainly hope for an improved PC2 which can fix some of the current problems and maybe expand it a bit one need to be careful to not end in the same way as PG did: Introduce so many changes and "improvements" to totally change into something totally different and, arguably i admit, worse.

There's always a risk, but I actually loved some mechanics of the sequels, at least from their idea.
For example, I often raved about the force pool mechanic, which works splendidly to prevent a player from having an all-tiger core. I also kinda liked the leader concept as an abstraction for experienced crews. PzC could use something like that. I liked tank overrun. And some other things.
Actually, my main complaints with the PG sequels were the graphics, PG2 had the horribly washed-out map graphics (they could have done so beautiful hand-drawn maps...) and then came the horribly low-poly 3D models - the sprite based approach is better (and much easier to mod).

I play this game for fun as you said and I like the pigeon shooting mechanics that you described

Heh, I think we're basically of the same opinion timek.
I principally like the clay pigeon shooting as well, otherwise I would've moved on long ago.
Doesn't mean there is no room for improvements, though.

I think the game works fine for the Axis. They are supposed to be outnumbered by up to 3 to 1, against tactically less skilled opponents. As the Allies though the game falls flat. Outnumbering Germans Should be the only way to beat them. For people thinking the game is too easy there are scores of ways to balance it. One way is to simply restrain yourself from using the uber units. Playing with all your armor Tigers is ridiculous from an historical viewpoint. When humans play games like chess it is customary for the better player to give a handicap, pawn,rook, or even queen. Shouldn't the poor dumb computer be given the same consideration?

Self-restraint is a difficult concept for many people...
Also, like I wrote upthread, the scenarios are balanced in a way that they can be beaten relatively easy with the best units available, at least partially, but as soon as you water that down, there's only a fine line in which you feel more challenged - beyond, you'll constantly lose stuff and feel like you're being beaten up.
For some folks, this can be fun also, but I don't think this applies to the "typical" PzC player.

Finally, about giving the better player a handicap ... that's why in the game we're usually playing with one chess set vs. three or four chess sets of the opponent. Actually, that's not so bad an analogy, in the late war soviet missions, it really feels like the soviet AI has more queens, but also a ton of peons that it can't get out of the way so the queens can maneuver...
_____
rezaf

[the_iron_duke]

You would have thought that the two examples of poor A.I. behaviour I cited earlier would be easy to fix:

1) Don't let tanks attack infantry in close terrain - or have a percentage scale for probability of attacking where the computer will always attack if the odds are in their favour, usually attack if the odds or even, rarely attack with unfavourable odds and never attack with poor odds.

2) If a unit is low-strength it should retreat and regroup - perhaps have a scale where the lower a unit's strength the greater the chance of retreating to regroup, taking into account the entrenchment value of the unit.

It is interesting to compare Panzer Corps to Slitherine's Field of Glory as they are similar games - think of Field of Glory as Panzer Corps 1000BC-AD1500. The game emphasis of the titles, however, is completely different. With Field of Glory, I think pretty much no-one plays single-player - perhaps their first game and that's about it. It's all about the multiplayer game and I must have played over 500 multiplayer battles against human opponents. They really test and develops one's generalship and there are some players that rival Alexander, Napoleon or Rommel in their ability to win battles as superior tacticians.

Panzer Corps seems to be the opposite - 99% about the single-player game with little multiplayer action going on. The commercial strategy of the games reflect this with Panzer Corps going for single-player campaign expansions and Field of Glory going for multiplayer army book expansions.

My feeling, after many years of playing computer strategy games, is that nothing beats playing a human opponent. Why play against a computer trying to simulate a human when you can play against a human? Computer A.I. for strategy games still has a long way to go come anywhere close to matching an experienced human player (computer chess being the exception to the rule).

The only down-side for multiplayer games is waiting for opponent moves. Some players are able to return several moves a day while others might make one every couple of days. The solution is to play more games! I often had a dozen games on the go at one time.

[timek28]

I think the game works fine for the Axis. They are supposed to be outnumbered by up to 3 to 1, against tactically less skilled opponents.

Well the ratio depends. After 43 I say yes, but before and in 43 I say no. And if lacking tactics means suicidal then Soviet commanders have been illustrated accordingly.

But I doubt that it has been exactly like that - getting smashed again and again (Enemy at the gates is not only valid historical depiction of events).

As the Allies though the game falls flat. Outnumbering Germans Should be the only way to beat them.

This is what I agree on. There is too little core slots for player in AC. Not to mention that half of them I have to use for air units (as in real life), so my ground units are outnumbered and out-qualified. There are some scenarios that allow out of core units to be switched on battlefield and some aux units, but still there is a feeling that most of the time there are more axis units on the field. I cannot believe Germans had that many Panthers late in the war?! This again shows AI incompetence and emphasizes the need of AI to have more units then player in order to match up.

For people thinking the game is too easy there are scores of ways to balance it. One way is to simply restrain yourself from using the uber units. Playing with all your armor Tigers is ridiculous from an historical viewpoint. When humans play games like chess it is customary for the better player to give a handicap, pawn,rook, or even queen. Shouldn't the poor dumb computer be given the same consideration?

This is a good point. However problem here are replacements and prestige. Using units of less quality on countless Soviet tanks can lead to losses which are almost impossible to replace with little prestige late in the war. There is huge quality gap between PzIVj and Tiger2 (or Tiger, or Panther), and PzIVj can match with T34-85, but it is going to have losses in that match. Not to mention KV85 and IS tanks. So going historical could be interesting but IMO with current amount of Soviet units that would lead to lost scenarios pretty often no matter which kind of brilliant tactic player uses.

[bebro]

Re numbers, I found them pretty warped in the DLCs I played (39-41, didn't try later so far). I some scns in Poland 39 or Western Europe 40 it looked like I was outnumbered 3 to 1 or even 4 to 1 already. There might be cases where this happened, but generally it felt wrong when opponents like Poland or Benelux have seemingly never-ending supply of stuff that early in the war.

Thing is, I can understand the "make it challenging" argument - but often it wasn't really that difficult after all. Sure, some scns were more than others, but overall it felt like I was mowing down hordes of units, hitting the XP cap for my guys in no time, and getting DVs mostly. I had more difficulty when playing vanilla the first time, on lower settings (tho - admittedly - might be I just have improved since then).

However, I'd rather prefer ratios that fit somewhat to the historic situation and enemy troops which are harder to kill to make good for that (use entrenchment, heroes, whatever). Vs. the USSR it makes perfectly sense of course to face overwhelming numbers, but not vs. every other country on the planet.

[timek28]

@the_iron_duke

I agree that the best solution is multiplayer. As I said before I will try it out just as I finish AC campaign (and I'm near the end).

I believe multiplayer in PC is not that attractive because of 2 reasons:

1) AI is much easier opponent to play against (and probably brings more "fun" to the game, where player knows what to expect and doesn't get frustrated a lot, or he can reload if he gets pissed).
2) Arrangements have to be made prior to game, small amount of games and players available with potentially large inequality in player skills. Also timetable has to be met between players.

Also what I don't understand is the asynchronous concept of multiplayer. So basically if I play one turn and then if my opponent is not on his computer I have wait until he plays his turn in couple of hours say? Then if I'm not at my computer he should wait again etc... ? Is this how it works if not both of us are ready to spend 2-3 continuous hours playing one game? I guess the point being is for both of us to arrange time when we are on our computers...

[the_iron_duke]

2) Arrangements have to be made prior to game, small amount of games and players available with potentially large inequality in player skills. Also timetable has to be met between players.

It's a Play-By-Email game so one player takes his move and then it's uploaded to the server. Then when the other player logs in, the game is downloaded to his computer and so on, back and forth. There is no need to both be on the computer at the same time, keep the program open or any requirement to co-ordinate anything. As I said, some players can return several moves a day (and I like these players!), while others are very slow and play a turn every two or three days. Probably a turn a day is the average.

1) AI is much easier opponent to play against (and probably brings more "fun" to the game, where player knows what to expect and doesn't get frustrated a lot, or he can reload if he gets pissed).

One should adhere to the virtue of chivalry. In my experience of Field of Glory, 95% of players will play out a game to conclusion, even if they are losing. It kind of tends on what sort of player you are. If you are the type of person that can only enjoy a game when winning and enjoy shooting fish in a barrel then a single-player campaign is probably more suitable. If you want to test and improve yourself as a general, developing and executing military strategy and tactics, then multiplayer is the way to go. You will likely lose more games than you win to begin with but you will improve - when I started playing muliplayer my win percentage was well below 50% but a few hundred games later I would win 80-90%!