There are three sizes of provinces on the map
- small
- medium
- large
About those three Johan said
After that Johan talked about terrain
...
If I was more vain than I actually am, I'd delete my post. o
There are three sizes of provinces on the map
- small
- medium
- large
About those three Johan said
After that Johan talked about terrain
You can more or less test every possibility in chess without too much work.If a chess computer in the 1980's was nearly as the best human player, I see no reason why the AI in HOI 3 can't be at least adequate? The AI in itself has advantages that the human player doesn't; the AI lacks (or should lack) fog of war giving it complete awareness of the current situation. Additionally, the AI can perform caculations whereas the human can only make educated guesses. And as a result of knowing the exact situation and having the ability to perform battlefield calculations perfectly, the AI should have the advantage of being to plan a perfect campaign. That means anytime the AI is on the strategic offenive it can calculate the best path to encircle the maximum amount of enemies taking into account the exact time it would take the human to counterattack, retreat, etc. The AI has the ability to outplay the human player but the right now the system is garbage.
If a chess computer in the 1980's was nearly as the best human player, I see no reason why the AI in HOI 3 can't be at least adequate? The AI in itself has advantages that the human player doesn't; the AI lacks (or should lack) fog of war giving it complete awareness of the current situation. Additionally, the AI can perform caculations whereas the human can only make educated guesses. And as a result of knowing the exact situation and having the ability to perform battlefield calculations perfectly, the AI should have the advantage of being to plan a perfect campaign. That means anytime the AI is on the strategic offenive it can calculate the best path to encircle the maximum amount of enemies taking into account the exact time it would take the human to counterattack, retreat, etc. The AI has the ability to outplay the human player but the right now the system is garbage. All that is needed to fix it though is tact programmig and trial and error.
If a chess computer in the 1980's was nearly as the best human player, I see no reason why the AI in HOI 3 can't be at least adequate?.
If a chess computer in the 1980's was nearly as the best human player, I see no reason why the AI in HOI 3 can't be at least adequate? The AI in itself has advantages that the human player doesn't; the AI lacks (or should lack) fog of war giving it complete awareness of the current situation. Additionally, the AI can perform caculations whereas the human can only make educated guesses. And as a result of knowing the exact situation and having the ability to perform battlefield calculations perfectly, the AI should have the advantage of being to plan a perfect campaign. That means anytime the AI is on the strategic offenive it can calculate the best path to encircle the maximum amount of enemies taking into account the exact time it would take the human to counterattack, retreat, etc. The AI has the ability to outplay the human player but the right now the system is garbage. All that is needed to fix it though is tact programmig and trial and error.
The fact that human's beat chess playing computer all the time... well enough said.
The computer can fight battles but it has a hard problem sticking to a plan, it is forever evaluating your moves and countering them.
You retreat it attacks. You fall back it follows. You slam the door and its army is encircled. Did the AI do anything wrong? No.
Would a human player have seen the bigger picture, maybe? So getting the AI to do the right thing is a lot harder than you think.
Chess is a very straightforward game to program as there exists large bodies of manuals on what counter to make, there is no terrain, and a strickly limited board. Where a computer beats a human is that it has the ability to evaluate moves in huge numbers of steps ahead. I used to play chess daily and even after years of doing so seeing the board more than 3-4 steps ahead took a "good" day. Also playing against a computer is very "dry" as it takes out the human element.
You example is a bad one because yes the AI goofed as no a human would not have missed that as the strength on your flanks is a particular concern for anyone with a lick of sense. The proper evaluation of enemy strength and intention is paramount of planning. I've called a halt to offensive operations for exactly this reason. I could go forward more but I would be exposing my flanks and leaving myself dangling in the wind which is not something that makes sense to do (normally, as always, particular circumstances could change that).
The task faced by the AI programers is brutally difficult as there are, unlike in chess, very few hard and fast rules. But the largest problem I see with the AI is it does not make long term plans, and it tends to react too much sending units scurrying back and forth endlessly. Watching a unit you have lent to the AI sometime its behavior often comes close to nonsenical. I have seen air units shipped all over the world and rarely even begin to understand why they are there and when they are there its rare they actually seem to do something.
How to program in a long term planning loop is hard to say. I would hate to try and do it (and yes I do program). Just putting down why and how I plan an assault I see a large number of IF-THEN-ELSE conditions showing up. Then once the attack begins and the situation becomes fluid this is when you need quick reactions to developing conditions. Plus the proper utilization of reserves and the reformation of a reserve, stop lines etc. All of this is a complex interaction of short term goals within an overall framwork of a long term objective. But first and formost you need that long term objective as "being only reactive" means basically you have lost in some situations. Reactiing to opportunities is different then always "responding" to the enemy...I hope this makes sense.
As I have a goal with specific objectives the AI is hard pressed to counter this in general. The same it true in Chess, if I have a plan and you don't you may complicate my life but its unlikely in the extreme that you are going to defeat me. As the AI lacks a plan its only chance at sucess largely lies in overwhelming numbers and when that fails it has nothing to fall back on. This isn't to say that even the current AI can't give you a "hard time" I can think of several examples from my last game but for the most part the lack of an overall operational plan is crippling.
As a concrete example from my last game "Overlord" in fall of 44 had the following objectives: Land at Caen, drive through Paris to the Vichy Border as a primary attack goal to isolate the western battlefield from supply, mop up the divisions trapped. Stage 2 establish a secure boundry using the minimum number of required provinces (somewhat reactive). Changes to the original plan were the re-use of the invasion Corps to secure a province by naval invasion after Caen (coast hopping) and that they were used to secure the coastal flank, plus occupation of Djion due to response to enemy moves. And lets not talk about all the naval movements required to assemble the invasion force, use of naval gunfire, plus airpower used to both prepare the battlespace and during the combat itself. My mind boggles at trying to write a conditional sequence that could replicate that.
You can only program something like this if you have first a routine that establishes theatre level objectives plus evaluates them for plausability since piling up troops for an invasion makes no sense when you have no transports for example.
Another concrete example. Japan controls India China and Persia. Where/how to proceed?
Option 1: continue drive to home islands, by island hoping and closing in until bases which allow strat bombing to commence.
Option 2: land in occupied India and cut the japanese forces in half making use of the better terrain to exploit an armoured/mechanized force advantage.
Option 3: land in annexed Persia to allow for a US controlled province to ensure supply without requiring British involvement.
Option 4: declare war on Iraq and attack overland through Iraq into Persia ensuring a land route to the powerful commonwealth land forces.
Option 5+: ???
Which is best? Where to assemble the troops given a particular option? How many divisons are needed? etc etc etc
This is a horse of a different colour to programing a chess game AI.
Exactlly, Corpse AI takes care of 2-3 provinces to attack, Army Group AI cares about 2-3Regions, and HQ AI cares about 2-3fronts. Brilliant thinking.I think I mentioned it in this thread before- with the command chain, you could presumably break down all AI planning (the military party of it, at least), into trivial "tasks" carried out each level of command, in turn, starting with the Theater AI and moving down, relying on crude min/max problems to pick the optimal offensive route. The advantage of this is that the programmer does not have to concern himself with how the AI "wages the war," but with how a Corps chooses which province out of 2-3 to attack. It effectively scales the problem down.
Actually if you read my thread properly you will realise I was saying exactly what you are saying. Just using a lot less words.
To sum up what I think. "It is easy to programme a computer to play chess, very hard to programme a computer to beat a human in a war game!"
Using your examples further reinforces this arguement
As I said, the reason the AI does that in HoI2 is because it is awfully fuzzy and does not really have any way to reasonably formulate goals - it's just a "Front AI" that behaves like an amoeba, trying to seep through the cracks in your line. But it's fundamentally impossible to "fix" with any amount of scripting, unless they adopt a new paradigm for the AI. I suspect that what I suggested will end up being reasonably close to what Paradox intends - otherwise I don't see how the delegation-based command system can possibly work.
I won't argue with you as I don't know myself! But without a way for the overall top level of the command chain to form coherent goals then I don't see what the lower levels will do that is fundamentally any different then the amoeba and potentially worse. Attacking this province may be sensible to the lower echelon leaders and utterly stupid to the overall "big picture." Does what I just wrote make sense?
I guess in programming lingo there needs to be a top down structure not a bottom up one or worse a top disconnected to the bottom structure.