abolish general "teleportation"

[mackwolfe]

This is minor, but it is a bit jarring to teleport your general across a continent to another theater of war in 1 day, while a diplomat needs many days to travel.

I say allow generals to be assigned to ANY army , regardless of whether in enemy or friendly territory. But require time to travel similar to diplomats ( straight line calculation ). This is more realistic . For a compact nation a single general would be enough. For a large empire, you will need one for each theatre.

I think this is simpler and I bet it would play better.

 

[Jomini]

This is minor, but it is a bit jarring to teleport your general across a continent to another theater of war in 1 day, while a diplomat needs many days to travel.

I say allow generals to be assigned to ANY army , regardless of whether in enemy or friendly territory. But require time to travel similar to diplomats ( straight line calculation ). This is more realistic . For a compact nation a single general would be enough. For a large empire, you will need one for each theatre.

I think this is simpler and I bet it would play better.

Let's not and say we didn't. Let's imagine that we do this. Is there any hope in hell that the AI won't end up with its generals easily baited into say the New World & India so you can then steam roll everything on the continent with a 5 pip advantage (or if you hard code them to keep a general home, baiting them so you can wipe everything in India for cheap)? I think not. The AI is already so bad at managing leaders (teleporting and all) that it needs a free leader slot to be remotely competitive. And half the time I can still get a lot free stackwipes by baiting the AI around (e.g. lead their doomstack with the God-general across a straight, roll out the navy, and go pound mid sized forces as much as forts allow).

Your diplomats are completely interchangeable, have one-off affects each mission, and are not pathway dependent in their function. Generals are none of those things. The AI needs to predict which general to put into which army (hard to know). It needs to account for the fact that it needs to manage general bonuses for sieging, fighting, movement times, attrition, and mil point cost. It also needs to figure out if army A is going to fight, siege, then march a long ways or if it is going to march a long ways, fight, and then siege. All of these make the calculations much harder and will take up a lot of clock cycles or be braindead stupid.

Most of the gimmicks in EUIV (teleporting generals, colonial range, coring range, etc.) aren't there just because Pdox never thought of how something else might work, but because some process brought them into being. Some of them, like say cores, are inherited things that have been evolving. Others are there because changing them takes a lot of work. For any feature it isn't enough that the human can deal with something more historical and be fine - it is that the AI also has to do it and not be dumb about it. Pattern recognition (which general is best with this army for the next two months) are easy for humans and exceedingly difficult for AIs.

This will be anything but simple and play anything but better.

 

[mackwolfe]

I actually think coding the AI for my scenario is easier.

Right now if you want to teleport a good siege leader deep into enemy territory , you detach a single regiment from your stack , occupy unfortified province next door and bingo, that regiment can now receive a leader. Not sure the AI knows how to do that or can be coded to exploit that.

The change with CS to province occupation makes teleportation abuse so much easier to do than before .

 

[Jomini]

I actually think coding the AI for my scenario is easier.

You are wrong. You are moving from P to NP here. NP problems are bears that suck down cycles. To put that another way, as the number of things the AI needs to consider to predict the value of its actions rises (e.g. number of enemies, terrains, etc.) the computational time required increases faster than polynomial time (e.g. exponential). Just about anything involving pathway runs into the dumb or clock cycling burning problem - pathing is trivial for humans and dastardly hard for AIs. Do not ever make a new pathing problem for trivial gains.

Right now if you want to teleport a good siege leader deep into enemy territory , you detach a single regiment from your stack , occupy unfortified province next door and bingo, that regiment can now receive a leader. Not sure the AI knows how to do that or can be coded to exploit that.

This is literally a 10 line or fewer thing to code for the AI:
If you have siege general you want to move in then
{check the regiment won't get slaughtered (check enemy armies in all "visible" provinces if they can intercept)
detach regiment
move regiment
assign general
move general & regiment back

Now compare that to your world. In your world the AI has to do all of that and also check contingencies that arise from shipping off a general with perhaps a year round trip turnaround. Moving generals to India isn't just a matter of getting a general where you can teleport safely, it is also wondering if Brazil will need the general or if Holland will. Likewise, sending a siege general to India means that you need to consider the utility of having a fighting general there if the siege might be contested and every other possible use for the fighting general. My setup literally requires checking maybe 5 adjacent provinces. Yours requires checking every province in the game that the AI owns. Every plausible outcome of every plausible permutation of human/AI behavior.

The change with CS to province occupation makes teleportation abuse so much easier to do than before .

Bullocks. CS makes it harder to abuse this. Think about it. Before the AI did not get a chance to update its generals until it sieged something and it needed to use hunter-killer armies deep in your territory. You could easily bait the AI all the way to Eastern Siberia.

Now if you let the AI dwell in any province outside of a fort for a month, they can reset their generals.

Making it easier to move generals, makes the AI better. The humans will always see the patterns and know "this army will need this type (or any) general, that army will need something else". We don't have to search a lot of stuff because we are extremely skilled at high level pattern recognition. AIs are complete and utter crap at it. We can dismiss 99% of possible states as not likely to be useful, the AI needs reams of heuristics to do that.

Making the mechanisms more difficult handicaps the AI - always.

 

[Suzaku]

They should vastly expand the pool of generals to the point where you're not loosing sword mana when you start in later dates.

 

[Fluffy_Fishy]

This is minor, but it is a bit jarring to teleport your general across a continent to another theater of war in 1 day, while a diplomat needs many days to travel.

I say allow generals to be assigned to ANY army , regardless of whether in enemy or friendly territory. But require time to travel similar to diplomats ( straight line calculation ). This is more realistic . For a compact nation a single general would be enough. For a large empire, you will need one for each theatre.

I think this is simpler and I bet it would play better.

I think I would prefer to see a calculation between diplomats and colonists. Although it would make it quite difficult to do the seven cities event chain, where its not uncommon for conquistadors to die.

 

[ShadowCammy]

I like this idea. It is strange to have a general suddenly teleport and command an army halfway across the world.

And for the AI, maybe Paradox could actually redo the AI to where it's not stupid. Take data from how real players play, and apply it to AI, somehow.

 

[Jomini]

I like this idea. It is strange to have a general suddenly teleport and command an army halfway across the world.

And for the AI, maybe Paradox could actually redo the AI to where it's not stupid. Take data from how real players play, and apply it to AI, somehow.

If Pdox can do that in any seriously decent way they should likely quit programming games and go into drone warfare. Being able to mimic human pattern recognition for strategic purposes is a billion dollar market.

 

[ShadowCammy]

If Pdox can do that in any seriously decent way they should likely quit programming games and go into drone warfare. Being able to mimic human pattern recognition for strategic purposes is a billion dollar market.

I mean taking tactics from human players that are commonly used and allowing the AI to use them. Smarter AI. More difficult AI.

 

[IIWW]

I mean taking tactics from human players that are commonly used and allowing the AI to use them. Smarter AI. More difficult AI.

Tacktics human player commonly use are based on AI. If PDS changed the AI, human tactics will change within a week tops.

Basically it's like taking Friedrich the Great tactics, giving it to a child, saying "just do what it says" and sending him against Friedrich the Great. The same principle.

 

[Jomini]

I mean taking tactics from human players that are commonly used and allowing the AI to use them. Smarter AI. More difficult AI.

The problem is that there are two approaches here, you can either have the AI try to learn (maybe using some sort of machine learning algorithm) or you can hard code via human effort.

So say you go for broke and try to have the AI just look over what people do and then try to emulate. Well how does it break down what it sees? When I split off an army and move back a province am I baiting the AI, limiting my attrition, or swapping a general for a faster siege? Let's say you give the AI enough framework to know I'm going for the siege. How will it know when to go for the fast siege (e.g. start the ticker now) or to go for a long war to burn dip points and spark revolts?

The other option is to hard code. You have a human look at things and try to implement useful decision algorithms to specify when tactics are useful and when they are useless. This is horridly time consuming and bug prone. If you make the rules too strict you end up with stuff like back when you could completely bottle up the AI with scorched earth. If you make them too loose then the AI will try things inappropriately (e.g. the beta development issues).

In either event, all the human strategies have been honed for a given environment and as noted above, the moment we end up with a new rule set governing AI behavior (either machine learned or coded), then human adaptation will be right there again with new tricks. Generating something that gets around the adaptation ... well like I said the US military will pay more than all the Pdox fans in the world.

This is a massive amount of coding for a trivial issue. Rather than putting all this coding into a new and mostly useless flavor feature, we should dump anything remotely like this into more basic AI functions at which the AI is currently weak e.g.:
1. Knowing when it has effectively lost a war (a very hard thing to measure) and should cut its losses.
2. Knowing how to manage amphib operations.
3. Knowing how to measure threats in the long term and friends in the long term.

At heart this is a cosmetic issue that requires a lot of resources to fix. Leave it for later.

 

[BrokenSky]

As discussed the last time this came up, if you did anything like this, you'd need more free leaders.

 

[mackwolfe]

As discussed the last time this came up, if you did anything like this, you'd need more free leaders.

Probably so. I think Kingdom rank would need a +1 leader, and empire another +1 leader . But not more. A single leader should cover an entire theater easily. It only takes 2 or 3 days to move 2 to 3 provinces using "diplomat distance " travel time.