Myrran AI Performace

[Nelphine]

It checks every turn (because sometimes units go to places from too far away, and get there long after the situation changes - sometimes even sattacking their allies who had conquered the city the AI was originally targeting, even if the AI originally was going to the city to defend it.) So it wouldn't know its already in route the next turn.

Also, part of why seravy has it set up this way is because for a while all of the AI best units just defended cities. Its offense was a mess.

[Seravy]

This is also why things like changing settler priority/amount has to be so carefully considered. The AI literally has no idea what is going on, so if its building too many settlers, then it might not have enough military and it won't know it.

There is this, but there is more to it. The AI
-Doesn't have a battle plan or overland strategy unlike a human - it isn't casting spells or making units to satisfy specific goals ("I need 7 bowmen to be able to attack that city - wait, the target learned guardian wind so I should switch to magicians now" doesn't exist for the AI.)
-Has to observe a personality and objective, so it has to make more of certain things than others for no reason at all
-Has to be diverse enough that it doesn't have an obvious weak point of "ah, the AI always does this so I can beat it with that"
-And has to be diverse enough to keep the player entertained - summoning a stack of 9 War Bears as first 9 spells and sending them into the player's capital on turn 15 works, but doesn't make a fun game - you either play something that can specifically counter 9 war bears or don't play at all - this is why I generally don't enjoy PVP games much, as a human opponent will go for the best strategy and eventually it's either "I can always counter it with this so I always win - wait, why is no one else playing anymore?" or "I have no counters and will lose anyway so better not to even start the game" instead of having to adapt to unforeseen circumstances in every game.

[Catwalk]

-And has to be diverse enough to keep the player entertained - summoning a stack of 9 War Bears as first 9 spells and sending them into the player's capital on turn 15 works, but doesn't make a fun game - you either play something that can specifically counter 9 war bears or don't play at all - this is why I generally don't enjoy PVP games much, as a human opponent will go for the best strategy and eventually it's either "I can always counter it with this so I always win - wait, why is no one else playing anymore?" or "I have no counters and will lose anyway so better not to even start the game" instead of having to adapt to unforeseen circumstances in every game.

I disagree strongly with this last part. We should always aim to make the AI as effective as possible. It should not treat the human player unfairly (giving preference to the human player as a target over other wizards would be unfair), but it should definitely go for the throat. If you're able to make the AI more lethal (without cheating, ie. treating the human player differently) then you should go for it. If that makes the AI too hard, the bonuses can be lowered. Impossible should be nigh impossible to beat, I think that's what I've seen you write elsewhere as well.

[Seravy]

It checks every turn (because sometimes units go to places from too far away, and get there long after the situation changes - sometimes even sattacking their allies who had conquered the city the AI was originally targeting, even if the AI originally was going to the city to defend it.)  So it wouldn't know its already in route the next turn.

This. I believe the AI learning how to pick a target every turn for its stacks was one of the greatest improvements to it. If nothing else, for the fact that targets move on the map - enemy stacks won't stay in place and might even split up - only cities are stationary but those can change owners or get different garrisons.
Old AI would send a stack towards a player city with 3 spearmen in it, only to get slaughtered by the 5 storm giants that got pushed in there when the player saw the stack coming.

[Catwalk]

It checks every turn (because sometimes units go to places from too far away, and get there long after the situation changes - sometimes even sattacking their allies who had conquered the city the AI was originally targeting, even if the AI originally was going to the city to defend it.)  So it wouldn't know its already in route the next turn.

Also, part of why seravy has it set up this way is because for a while all of the AI best units just defended cities. Its offense was a mess.

That would indeed ruin my proposed solution. I'll just stay away from AI discussions, I really don't know enough about it.

[Seravy]

I disagree strongly with this last part. We should always aim to make the AI as effective as possible.

Fortunately due to the other points it cannot be any more effective anyway - and exactly due to what I described, I don't even think that would be more "effective". As soon as the player learns to counter the AI's strong strategy - if such a thing existed-, the game is easier to beat than if the AI was semi-random. In the above example, the player can summon 3 of their own bears, park them near the AI's capital, and strike a turn before the AI does - since the AI committed all their power on summoning and sending the 9 bears, it has no garrisons and loses - same if the player can somehow fight the 9 bears, for example summons 3 Sprites which the bears cannot attack and get chewed up by lightning.

I think the only part where the AI is actually pulling its punches is using Disjunction - global enchantments would be literally unplayable if it wasn't limited due to the AI casting advantages -which are necessary for all other kinds of spellcasting so this is more of a limit on not using the advantage for this purpose as it would be unfair.
Same for city curses I guess, the AI could win in no time if it spammed the max amount of those every turn.

[Catwalk]

I disagree strongly with this last part. We should always aim to make the AI as effective as possible.

Fortunately due to the other points it cannot be any more effective anyway - and exactly due to what I described, I don't even think that would be more "effective". As soon as the player learns to counter the AI's strong strategy - if such a thing existed-, the game is easier to beat than if the AI was semi-random. In the above example, the player can summon 3 of their own bears, park them near the AI's capital, and strike a turn before the AI does - since the AI committed all their power on summoning and sending the 9 bears, it has no garrisons and loses - same if the player can somehow fight the 9 bears, for example summons 3 Sprites which the bears cannot attack and get chewed up by lightning.

I think the only part where the AI is actually pulling its punches is using Disjunction - global enchantments would be literally unplayable if it wasn't limited due to the AI casting advantages -which are necessary for all other kinds of spellcasting so this is more of a limit on not using the advantage for this purpose as it would be unfair.
Same for city curses I guess, the AI could win in no time if it spammed the max amount of those every turn.

If the main thing you're getting it as that you don't want the AI to be one-trick ponies that try the same "strong" strategies over and over, then I agree completely. Predictability is a big weakness, and the semi-random approach is definitely the way to go.

The only thing I had a beef with was that you were hinting at wanting the AI to pull its punches. If it is necessary to make it do that, the game balance itself is to blame. I haven't played around with city curses and global enchantments enough to see how those work out. Personally I'd have a blast if I saw the AI obliterate all my cities with curses  As long as I still have a fighting chance and I'm not being singled out, of course.

On a side note, would you ever consider making a Hardcore mode where every wizard ganged up on you from the start? Obviously I'd play that at Normal or Hard instead, but it'd feel like a fairer game that way.

[Seravy]

I disagree strongly with this last part. We should always aim to make the AI as effective as possible.

Fortunately due to the other points it cannot be any more effective anyway - and exactly due to what I described, I don't even think that would be more "effective". As soon as the player learns to counter the AI's strong strategy - if such a thing existed-, the game is easier to beat than if the AI was semi-random. In the above example, the player can summon 3 of their own bears, park them near the AI's capital, and strike a turn before the AI does - since the AI committed all their power on summoning and sending the 9 bears, it has no garrisons and loses - same if the player can somehow fight the 9 bears, for example summons 3 Sprites which the bears cannot attack and get chewed up by lightning.

I think the only part where the AI is actually pulling its punches is using Disjunction - global enchantments would be literally unplayable if it wasn't limited due to the AI casting advantages -which are necessary for all other kinds of spellcasting so this is more of a limit on not using the advantage for this purpose as it would be unfair.
Same for city curses I guess, the AI could win in no time if it spammed the max amount of those every turn.

If the main thing you're getting it as that you don't want the AI to be one-trick ponies that try the same "strong" strategies over and over, then I agree completely. Predictability is a big weakness, and the semi-random approach is definitely the way to go.

Yes, that's what I'm trying to say. There are pretty much two possible cases with strong strategies :
-they have a weak point and can be coutnered, in which case we have the one-trick pony scenario
-or they don't in which case the game balance is horribly off, as you describe.
So the AI shouldn't go for it in either case - if the balance is wrong, fixing that is the way to go instead.

On a side note, would you ever consider making a Hardcore mode where every wizard ganged up on you from the start? Obviously I'd play that at Normal or Hard instead, but it'd feel like a fairer game that way.

Interesing idea...amazing idea, actually. The only problem I see is the interface.
This clearly is a setting for the start of the game, and adding new buttons there is anything but easy.
The "settings" screen has some unused checkboxes, but this isn't a mode that should be possible to turn on/off during games - as the actual implementation would be to force all AI into an alliance with each other and war with the player - even if you turn it off, the wars and alliances remain.
Making a patch like the self-play one would be an easier way to implement it, but that's not too user friendly.

... curses have a low casting cost, so the AI would have no problem putting Corruption on every tile you own, draining all your mp with Drain Power, putting Chaos rift/famine/pestilence etc on all your cities, counter every spell with Spell Blast and so on.

[Catwalk]

I'm not sure if anyone but me would bother with it, so something crude and easy would be just fine. And yeah, it'd be a permanent setting for the game.

Because of the gloomy scenario you prescribe, this would indeed be a setting for something like Normal difficulty. And I don't have to go with 4 wizards, I might try it with 2 or 3 instead.

[Tlaloc]

I decided to run some tests with the self-playing patch to see how much the AI expands on Myrror. 

What is this self-playing patch? Sorry if I missed the announcement about that. Is it available for download? How does it work?