Polisurgist
King
- Joined
- Dec 13, 2011
- Messages
- 630
To those early game warmongers among us.
- Thread starter Dtekkar
- Start date
TheMeInTeam
If A implies B...
- Joined
- Jan 26, 2008
- Messages
- 27,995
Giving the AI certain scripts would make it hilariously obnoxious 
.
.
Grendeldef
Trancerelic & coffeeholic
So many good points from Cromagnus & TheMeInTeam and as I'm not sure whether the former actually disagreed with me but while I mostly agree with both of them I won't go through the pass of endless quotes but for this
the answer is simple - money. Firaxis is in the business of making money and the revenue comes from the masses, not from from those few who are seriously disappointed with the AI performance. In the case of high end chess comps those were originally an excellent show case for IBM and the market position of IBM & Firaxis is more than slightly different.
With infinite computing power the AI could be made a lot better but it'd also be a commercial suicide so it won't happen. and as others pointed out with more complex computations comes the longer waiting times and I'd assume that the vast majority don't want a longer turn times even if it would mean slightly better AI.
My main complain isn't as much targeted towards the grand scheme of AI but the multiple easily fixable annoyances which could be corrected without serious cost or manpower.
Making AI scalable with CPU power of host PC is an interesting thought and a welcomed option but sadly, I can't see the implementation of it anytime soon. It'd be seriously cool to share the task between all available machines for at least to see whether there's a visible change or not.
the answer is simple - money. Firaxis is in the business of making money and the revenue comes from the masses, not from from those few who are seriously disappointed with the AI performance. In the case of high end chess comps those were originally an excellent show case for IBM and the market position of IBM & Firaxis is more than slightly different.
With infinite computing power the AI could be made a lot better but it'd also be a commercial suicide so it won't happen. and as others pointed out with more complex computations comes the longer waiting times and I'd assume that the vast majority don't want a longer turn times even if it would mean slightly better AI.
My main complain isn't as much targeted towards the grand scheme of AI but the multiple easily fixable annoyances which could be corrected without serious cost or manpower.
Making AI scalable with CPU power of host PC is an interesting thought and a welcomed option but sadly, I can't see the implementation of it anytime soon. It'd be seriously cool to share the task between all available machines for at least to see whether there's a visible change or not.
Socratatus
Emperor
- Joined
- Jul 26, 2007
- Messages
- 1,636
Do you play Chess? Chess has set moves and pieces that works with numbers very well. A computer can be scripted to know not `learn` since it don`t actually learn, to counter every move put before it and to use new ones from its `book`. These moves are all known moves from all the best players- the AI can figure out the rest from the limited choices.
An open world strategic AI that has to follow world-type moves and have an exponentially larger number of moves to consider, but not only that it has to be able to think and think abstractly, basically recognise a HUMAN`S intentions not just from his moves, but to hypothethise and extrapolate what a Human is likely to do next, this includes deception and trickery. No computer AI can do this, not even badly against any Human with a reasonable understanding.
And if there`s even remotely such a thing, the US military probably has it deep in some maximum security vault.
CIV5 and other games AI can be a lot better, but it`ll never match a Human`s brain.
TheMeInTeam
If A implies B...
- Joined
- Jan 26, 2008
- Messages
- 27,995
I really didn't emphasize it the way I wanted, but developer cost constraints is one of the biggest reasons I'd advocate a very simple algorithm for unit movement. Making the AI actually good at it is virtually unattainable within cost/time budgets, and making AI turns longer isn't particularly appealing to anyone (some dislike it more than others). Very, very few games won by the AI are won at a tactical level, and that's usually along the lines of an awful human blunder, rather than something the AI sets up deliberately.
From a budgeting standpoint, then, heuristics make more sense, with a lot more focus on the AI applying its bonuses at the strategic level to provide it the requisite advantage to overpower the human.
While people mention tactical AI as being poor now and then, by far the more grating things (and the majority of complaints) are coming from its diplo behavior and basic choices. Its bonuses can be a real pain but the game would be a lot less volatile if they were better in the strategy sense; winning on high levels with an AI that chooses to gouge science and can't be rushed would be a real uphill climb. You don't need much movement sense to attain that at all; you just need to get it to pick the proper social policies, technologies and buildings as a priority. Compared to so many calculations required to optimize movement and get a small return, something like that is far less expensive from a processing standpoint and yet would provide an AI that does it and doesn't get attacked with an enormous lead.
From a budgeting standpoint, then, heuristics make more sense, with a lot more focus on the AI applying its bonuses at the strategic level to provide it the requisite advantage to overpower the human.
While people mention tactical AI as being poor now and then, by far the more grating things (and the majority of complaints) are coming from its diplo behavior and basic choices. Its bonuses can be a real pain but the game would be a lot less volatile if they were better in the strategy sense; winning on high levels with an AI that chooses to gouge science and can't be rushed would be a real uphill climb. You don't need much movement sense to attain that at all; you just need to get it to pick the proper social policies, technologies and buildings as a priority. Compared to so many calculations required to optimize movement and get a small return, something like that is far less expensive from a processing standpoint and yet would provide an AI that does it and doesn't get attacked with an enormous lead.
is it possible to have an AI enter "chess mode" only when at war and only considering, say, a hex circle of about 40? like only the expensive, processor-expensive code would be active in this situation and it would take all friendly and enemy units in that circle into account and make perfect moves at this time and place?
Deaths Deputy
Chieftain
- Joined
- Oct 24, 2013
- Messages
- 12
Apart from having an option to turn the warmonger penalties on or off, I would make two other changes.
1. The war-mongering penalty would fade over time. How quickly it did so could perhaps be adjusted depending upon each civs relationship to the civ that was attacked/destroyed. For example if you destroyed Alex (who doesn't want to do that?) and he was friends with half the other civs in the world, neutral to a couple others and guarded/hostile with the rest, then the war-mongering penalty would take longer to fade for some civs than others depending upon his relationship with them.
2. The war-mongering penalty should increase with each era. Conciousness of the benefits of collective security have gradually become stronger over the course of human history. I admit my analysis below applies to European history rather than global history but I think the point still stands.
In ancient and classical times most alliances were makeshift and were alliances of convienience which could be broken at a moments notice. War, conquest and betrayal were accepted as normal and inevitable; certainly there were almost never international alliances or co-operation designed to stop an aggressor.
By the medieval period a conciousness of collective security and international co-operation was beginning to emerge but it was still very weak. It got stronger during the renaissance period, epecially after the treaty of Westphalia which defind the modern state system. By the Napoleonic era states were making international alliances, often long-standing in nature, and organising conferences and various bodies designed to maintain a "balance of power". Although there were still many conflicts, those who tried to use aggression to radically change the balance of power (such as Napoleon) were faced with international co-operation to try and thwart them. In modern history aggressors can expect to face international embargoes, condemnations and war (in the guise of "intervention") if they don't stop.
Just my 2p anyway
1. The war-mongering penalty would fade over time. How quickly it did so could perhaps be adjusted depending upon each civs relationship to the civ that was attacked/destroyed. For example if you destroyed Alex (who doesn't want to do that?) and he was friends with half the other civs in the world, neutral to a couple others and guarded/hostile with the rest, then the war-mongering penalty would take longer to fade for some civs than others depending upon his relationship with them.
2. The war-mongering penalty should increase with each era. Conciousness of the benefits of collective security have gradually become stronger over the course of human history. I admit my analysis below applies to European history rather than global history but I think the point still stands.
In ancient and classical times most alliances were makeshift and were alliances of convienience which could be broken at a moments notice. War, conquest and betrayal were accepted as normal and inevitable; certainly there were almost never international alliances or co-operation designed to stop an aggressor.
By the medieval period a conciousness of collective security and international co-operation was beginning to emerge but it was still very weak. It got stronger during the renaissance period, epecially after the treaty of Westphalia which defind the modern state system. By the Napoleonic era states were making international alliances, often long-standing in nature, and organising conferences and various bodies designed to maintain a "balance of power". Although there were still many conflicts, those who tried to use aggression to radically change the balance of power (such as Napoleon) were faced with international co-operation to try and thwart them. In modern history aggressors can expect to face international embargoes, condemnations and war (in the guise of "intervention") if they don't stop.
Just my 2p anyway
PaleJackal
Chieftain
- Joined
- Feb 19, 2009
- Messages
- 78
I think that city size should be taken into account, it'd make early game warmongering less costly diplomatically, and it wouldn't punish taking small trash cities.
Cromagnus
Deity
- Joined
- Sep 11, 2012
- Messages
- 2,272
The computers that beat chess grandmasters cost like 100 million dollars to make...
If you can afford one of those, you can just make your own version of Civ. ;-)
What I mean is, even if it was theoretically possible for Firaxis to write perfect AI, the machine it would take to run the game would literally cost millions of dollars.
Simulating the decision-making of an elite player is almost an insurmountable challenge. However, they could still emulate some of the behaviors and long term strategies for a reasonably lower cost. It wouldn't be perfect, but the AI wouldn't need a 4-tech head start if they devoted half their resources towards Early NC and universities. This alone would greatly improve the deity experience IMHO, because it would fix some of the early-game issues. Right now the AI has a head start that you can overcome by t130 and by 200 you can be an era ahead. I'd much rather it always be a challenge to keep up.
Yeah, unfortunately the AI boils every decision to a mathematical equation. They assign values for certain things then do the item with the best numerical remainder... which is why they screw up their economy or place cities in poor spots, because not everything can be reduced to a number. Sometimes choices are more abstract... like if I sacrifice this unit it will delay the AI for two turn so I can mobilize additional forces and likely save my city - how can the AI compare this mathematically and make the same choice a human opponent would?
Same for using civilians as bait. I'll often use a captured worker/missionary to lure an AI unit into an ambush (usually concentrated artillery fire) and have yet to see the AI do the same because that decision is too abstract for a "is the value of x > y" type decision.
It's easier with chess because there are limited choices of moves and so many great games have been played the AI can easily assign a value to a move by adding up how many times it has led the user to victory compared to a different move.
Same for using civilians as bait. I'll often use a captured worker/missionary to lure an AI unit into an ambush (usually concentrated artillery fire) and have yet to see the AI do the same because that decision is too abstract for a "is the value of x > y" type decision.
It's easier with chess because there are limited choices of moves and so many great games have been played the AI can easily assign a value to a move by adding up how many times it has led the user to victory compared to a different move.
Ninakoru
A deity on Emperor
That is not correct. From a pure mathematical point of view everything can be measured and counted as a number.
About the Development cost and the player turns... Well, some extra evaluations shouldn't inflate computer turn times at all in any significant manner.
About Developer cost... Maybe you are right, but there's so much middle ground to fall in IMO...
Big J Money
Emperor
- Joined
- Feb 23, 2005
- Messages
- 1,141
+1 the warmongering penalty option
