A Better AI.
- Thread starter Blake
- Start date
weimingshi
Prince
- Joined
- Feb 22, 2004
- Messages
- 420
Bear in mind that the old tactical AI are basically like lemmings - they'll ALWAYS attack even massively against the odds... a big part of the tactical changes is overcoming these suicidal tendencies and ensuring the AI DOESN'T attack when doing so is suicide. So if you swap the mod out at a point where the AI should attack - yeah they're gonna attack, if you swap the mod out a point where the AI shouldn't attack - they're gonna attack.
Why don't you post a save? It's a lot more useful than a screenshot in this case.
Hum.... Save file too large, can't attach it, they are 1 meg large. Anywhere I can upload it to? The problem I have with better AI attack logic is that Hatshe stops attacking after suiciding bunch of trebuchet. With my grenadier down to 4.6 and cannon 8.9 hp, if only hashe moved his 2 cr2 maceman, I can kiss angle and say goodbye, yet she never bothered to move them an INCH. Hashe's knight and cr2 maceman definately has a higher odds than my left over defender. My game setting is huge map, 18 civs, Deity, no city razing, raging barb, aggressive AI.
You can e-mail it to BetterAI@gmail.com
Julian Delphiki
Anton's key
Correct me if i am wrong, but the first posts i remember seeing from Blake regarding these AI improvements was the fact that he found out how retarted whole AI city founding/improving code was and wanted to fix that. So i don't think thats a good idea. Adjusting bonuses according to level would be much better.
My memory might be cozy so refresh it if needed.
WuphonsReach
Prince
- Joined
- Nov 6, 2005
- Messages
- 426
I didn't think about the issue where a player might inherit less then optimal city locations. So it might make sense for even the "dumbest" of the AI designs to still make "pretty good" decisions. Not always perfect, but at least not really stupid city placement.
...
I come from the "huge" side of the fence... a typical mid-game empire, even with 18 civs can have 15-20 cities (and once you get into the top 3 dogs it could be 30+ cities). Which is an awful lot to micromanage every turn. But it's something that the AI excels at (micromanagement), which gives them an advantage. Now, I do like to micromanage a bit... but the main reason I play on huge maps is that I like the size, breadth and chaos of dumping 18 civs into a world and then spending a while living in that world. (Average game time for me is around 20 hours, give or take a few, spread over a week or three.)
Which is why I'd like to handicap the AI by making it more random and prone to mistakes at the lower levels. I'm not against giving the AI bonuses at higher levels, but I
...
I come from the "huge" side of the fence... a typical mid-game empire, even with 18 civs can have 15-20 cities (and once you get into the top 3 dogs it could be 30+ cities). Which is an awful lot to micromanage every turn. But it's something that the AI excels at (micromanagement), which gives them an advantage. Now, I do like to micromanage a bit... but the main reason I play on huge maps is that I like the size, breadth and chaos of dumping 18 civs into a world and then spending a while living in that world. (Average game time for me is around 20 hours, give or take a few, spread over a week or three.)
Which is why I'd like to handicap the AI by making it more random and prone to mistakes at the lower levels. I'm not against giving the AI bonuses at higher levels, but I
I think the AI doesn't trade resources for gtp because it believes that they are worth alot more. As such, it wont ask you for 2 gpt in exchange for giving pig or something. That is why they are ok with accepting the deal when you present it to them, and later upping it way up to 20 gpt for it later on in some cases. The AI places a high gpt value to it and as such (I think) never offers it.
Maybe the AI could become more lenient towards this, particularly the expansionists say, considering they will get more resources than normal and also have gold economic trouble due to the expansion.
Sounds like the units see the more advanced units as being too powerful to take. Maybe the war engine should take this into account and build a non-sacrificial force (much like it is) and then intentionally build a sacrificial force which has a lot of siege units in it with the intended aim of dislodging difficult (including more advanced) cities.
Maybe keeping it the way it is (settles cities well), but be a little lax in producing workers. That way, the novice conqueror will still get good cities but it will take a while for the less-than-optimal number of workers to develop the land, thus slowing them down on lower difficulties.
weimingshi
Prince
- Joined
- Feb 22, 2004
- Messages
- 420
About level of difficulty. you guys are just complicating things up. Why bother writing different behaviour codes for different difficulty level? The AI is already dumb enough compare to human player, no need to make it more so. Currently higher difficulty is achieved by giving AI significant discounts on things like tech research, unit production. if you want an easier difficulty, just give AI some penalty thats it.
WuphonsReach
Prince
- Joined
- Nov 6, 2005
- Messages
- 426
Fighters and bombers...
I saw a mention of this earlier in the thread and I wanted to echo it. My standard plan of attack for fighters/bombers is roughly as follows:
1. Start marching my SoD towards the target city
2. When I'm 2-3 turns away, send in as many fighters as I can to do strafing runs on the city. Then send a wave of bombers to bombard the defenses down to zero (when the fighters go first, most of the defensive fighters won't touch the bombers... I think).
3. On the next turn, any fighters > 75% health do another strafing run (to draw out any defensive fighters), then the bombers do a strafing run against the city defenders until they're knocked below 60% health.
4. Stack arrives at city... entire stack skips a turn.
5. Repeat fighter strafing then bomber strafing if any defenders are above 60-65%. Otherwise attack other stacks or strategic resources.
6. Attack with seige weapons first to inflict even more collateral damage on the defenders.
7. Attack with any units that have withdrawal chances and can inflict collateral damage.
8. Take the city with my 2nd best units (whatever unit has > 90% chance of winning) or attack with best units.
Steps 5-8 *have* to happen on the same turn in order to be effective. That's one thing that the AI doesn't do well yet (in addition to not using bombers to weaken defenders).
For defense, fighters and bombers should strafe any SoD within reach. Keep the attackers weakened as much as possible, going after the largest stacks first (for max collateral damage with bombers). I've held off some very large SoDs by weakening them with strafing attacks or suicidal seige weapons.
The only tile improvements I'll bomb are strategic military resources (uranium, oil, etc) or food. Otherwise the bombers are better utilized by strafing any large stacks of attackers/defenders within reach.
My air defense usually consists of 1-2 SAM infantry (depending on how strong the enemy air force is) and 1-3 fighters on patrol. Inner cities with low-risk can get away with a single aircraft (or none). Edge cities should have at least 1, and maybe 2 or 3 if there are strategic resources or the neighbor is annoyed (or worse).
I also have a pool of N fighters (divide the number of cities in my empire by 2 or 3) and M bombers (divide # of cities by 3 or 4) which are mobile. They get moved to the least threatened city and strongest city closest to the front lines. Once in place, any aircraft > 75% health will proceed to do bombing / strafing attacks against defenses / stacks / improvements. On a huge map, this might mean 8-10 fighters along with 6-8 bombers.
It's tricky to code a mobile stack like that, but air support for an assault force really makes things go faster. I can usually take a city with only losing 1-2 units which means my assault force only gets bogged down by the need to defend what I capture. Maybe the AI can simply get by with moving planes towards the current "front" and randomly dispersing them across the "front".
I saw a mention of this earlier in the thread and I wanted to echo it. My standard plan of attack for fighters/bombers is roughly as follows:
1. Start marching my SoD towards the target city
2. When I'm 2-3 turns away, send in as many fighters as I can to do strafing runs on the city. Then send a wave of bombers to bombard the defenses down to zero (when the fighters go first, most of the defensive fighters won't touch the bombers... I think).
3. On the next turn, any fighters > 75% health do another strafing run (to draw out any defensive fighters), then the bombers do a strafing run against the city defenders until they're knocked below 60% health.
4. Stack arrives at city... entire stack skips a turn.
5. Repeat fighter strafing then bomber strafing if any defenders are above 60-65%. Otherwise attack other stacks or strategic resources.
6. Attack with seige weapons first to inflict even more collateral damage on the defenders.
7. Attack with any units that have withdrawal chances and can inflict collateral damage.
8. Take the city with my 2nd best units (whatever unit has > 90% chance of winning) or attack with best units.
Steps 5-8 *have* to happen on the same turn in order to be effective. That's one thing that the AI doesn't do well yet (in addition to not using bombers to weaken defenders).
For defense, fighters and bombers should strafe any SoD within reach. Keep the attackers weakened as much as possible, going after the largest stacks first (for max collateral damage with bombers). I've held off some very large SoDs by weakening them with strafing attacks or suicidal seige weapons.
The only tile improvements I'll bomb are strategic military resources (uranium, oil, etc) or food. Otherwise the bombers are better utilized by strafing any large stacks of attackers/defenders within reach.
My air defense usually consists of 1-2 SAM infantry (depending on how strong the enemy air force is) and 1-3 fighters on patrol. Inner cities with low-risk can get away with a single aircraft (or none). Edge cities should have at least 1, and maybe 2 or 3 if there are strategic resources or the neighbor is annoyed (or worse).
I also have a pool of N fighters (divide the number of cities in my empire by 2 or 3) and M bombers (divide # of cities by 3 or 4) which are mobile. They get moved to the least threatened city and strongest city closest to the front lines. Once in place, any aircraft > 75% health will proceed to do bombing / strafing attacks against defenses / stacks / improvements. On a huge map, this might mean 8-10 fighters along with 6-8 bombers.
It's tricky to code a mobile stack like that, but air support for an assault force really makes things go faster. I can usually take a city with only losing 1-2 units which means my assault force only gets bogged down by the need to defend what I capture. Maybe the AI can simply get by with moving planes towards the current "front" and randomly dispersing them across the "front".
HardCoder
Arrrr
- Joined
- Oct 21, 2006
- Messages
- 171
I have always thought that the AI's tendency to build huge cities was more of a product of its singlemindness than a product of its brilliance. Human players don't build huge cities like that (even when they can get over the lower happiness caps), because they aren't optimal. They look impressive, but they're not the ideal way to proceed in most cases. A city needs to be well sited, not huge. When it gets bigger after biology that's just a bonus, because the important development has been done for a couple of millenia. That's my $.02 anyway.
HardCoder
Arrrr
- Joined
- Oct 21, 2006
- Messages
- 171
Heh. I don't think you're the intended audience for this patch, or even the game.
HardCoder
Arrrr
- Joined
- Oct 21, 2006
- Messages
- 171
The AI seems to do a pretty good job beating down invaders if it has the resources, including bombing and strafing stacks. In fact it's ruthless. But as you say when it comes to invading, it doesn't seem to know anything about air support. I have certainly never seen it make an effort to bring any kind of air support with it in a sea invasion except of course gunships. A carrier load of bombers would be a heck of a surprise to a human player, who rarely even messes with that because it's, well, inconvenient. Given the AI's strategy of keeping a ready mass of troops in a port city, it wouldn't be that hard to throw in a carrier or two and some aircraft and make that part of the package.
Of course, there's getting the AI to bomb stacks and cities offensively, which like I said is something that is unfamiliar to me. It probably does it but it's not something that I worry about or even remember.
Late in the game, stealth bombers are absolutely devastating and you can just roll over an entire civ's full blown mech infantry and armor defense with their support.
Julian Delphiki
Anton's key
Heh. I don't think you're the intended audience for this patch, or even the game.
That was one funny line
. Hey, masochists have their place in civ too!Thyrwyn
Guardian at the Gate
I was referring to the AI offering gpt for a player's resource. As you pointed out, the AI places a much higher value on resources than typical players do. The AI never offers gold for resources, and under the Improved AI rarely has spare gpt lying around.
Wodan
Deity
While there's some logic behind adjusting city site selection based on difficulty level, I personally think it's overcomplicating the issue.
Besides, let's get the AI doing good city site placement in the first place. Without a doubt, the first step is to get the best algorithm we can, regardless of level. Then and only then should a decision be made as to whether to muck around city difficulty level. Before that point, it's premature. Sure, toss it out there for discussion, but let's not pester the better AI guys about it just yet. IMO, YMMV, and all that.
Wodan
Besides, let's get the AI doing good city site placement in the first place. Without a doubt, the first step is to get the best algorithm we can, regardless of level. Then and only then should a decision be made as to whether to muck around city difficulty level. Before that point, it's premature. Sure, toss it out there for discussion, but let's not pester the better AI guys about it just yet. IMO, YMMV, and all that.
Wodan
Wodan
Deity
The more and more I've been popping in worldbuilder and looking at the latest build, the more I think the current algorithm is a little wacked. The AI are consistently across the board putting city sites that are IMO too close. Perhaps it's my opinion that's out of whack.
Let me pose a question: Given unlimited space, it is good to always overlap by 2-3 tiles?
Wodan
Let me pose a question: Given unlimited space, it is good to always overlap by 2-3 tiles?
Wodan
Roland Johansen
Deity
Post a screenshot with gridlines on, so that we may see how bad the overlap is in your games. Don't only show the worst cases because that's not a fair judgement.
Iustus
King
To a certain extent, I think this is a matter of taste. You basically have to choose, do you want to waste some good tiles, or do you want to have some overlap.
I know I prefer to have further apart, better cities, while Blake prefers cities with some overlap, wasting almost no space.
Unless your cities are all 20+ size, some overlap does not really hurt you, and you do pay less distance costs.
-Iustus
player1 fanatic
Fanatic
Better to overlap then to waste space...
Several extra used tiles are woth more then extra maintaince you get for having an extra 1-2 cities squeezed on continent.
Several extra used tiles are woth more then extra maintaince you get for having an extra 1-2 cities squeezed on continent.
tofof
Chieftain
Yes.. yes it would.
And it'd be shortly followed up by "OMG H4x!!oneoneeleven" and "KICK TEH MODDER".Regarding city size, agreed, there are good arguments for both. The original AI, though, certainly wasn't evaluating anything and deliberately choosing to overlap because it best suited them. I simply got in the habit of looking at tiles that overlapped, and I'd raze any city that had more than 5 overlapping tiles with other cities (at least, with others I intended to keep). This quite literally meant razing about 25% of my conquests, which was ridiculous.
It should be possible to get the AI to make this choice, right? Is the city-placement planned, or is it just a greedy-choice algorithm right now? I see situations all the time where the 'best' location for a city, with the highest overall FPC, is in the center of a stretch that I could put two cities in instead. When I can fit two in there, hit all the same 'valuable' tiles, and not have overlap, I almost always go that way. When I could fit two in but they'd overlap, I try and judge what the overall effect would be on the maintenance vs FPC (small cities cost more in maintenance, etc). In general, if I don't get at least 35% more FPC out of two cities than one on the same site, I just go with the one. That rarely happens in good terrain - it's more common in desert/mountain/snow type situations. I don't care about 'wasting' near-worthless tiles by only building one city instead of two, and if a couple good tiles end up as collateral damage, so to speak, so be it. Critically, I try never to take the 'best spot,' having decided to only put one city in, and then go back and squeeze a totally crippled second city around it after the fact. That's why I asked whether the AI currently plans multiple placements out beforehand or just takes the best choice at the moment the settler pops.
Another thought - It'd be nice to see some consideration of personality with overlap. Organized, especially, can afford to build with little overlap to maximize the output of each city, as they can afford the increased distance. Similarly, I'd say expansive's +2 health can be easily wasted by artificially capping the size of the cities by overlapping them.
Wodan
Deity
Comments to the last couple of posts.
Sure, I'll post a screenshot. Can't do it right now but will do my best later today. But... you don't have to take my word for it and you don't have to worry about me giving you a "worst case" example simply to prove my point.
Simply open up your latest game. As long as it's past 1AD or so, there should be plenty of AI cities to give you data points. I hope it's not archipelago because that would skew things. Most any other map should be fine. If you haven't discovered the whole map yet just go into worldbuilder so you can see. Piece of cake.
Yes, I agree Iustus. It is a matter of taste. My question is should ALL the AIs be doing it always? I'm looking at my latest game and out of 40+ AI cities, only one doesn't overlap.
player1... I'm not advocating wasting and leaving gaps. I'm talking about reducing what I am seeing as significant overlap happening by default.
And I don't agree with your point about extra maintenance. Most any city will more than pay for itself. In addition, often it's better to work a specialist than to work a low-utility tile such as ocean or tundra, so simply working "extra tiles" depends entirely on what tiles are there. Anyway, this is all beside the point I was making.
Wodan
Sure, I'll post a screenshot. Can't do it right now but will do my best later today. But... you don't have to take my word for it and you don't have to worry about me giving you a "worst case" example simply to prove my point.
Simply open up your latest game. As long as it's past 1AD or so, there should be plenty of AI cities to give you data points. I hope it's not archipelago because that would skew things. Most any other map should be fine. If you haven't discovered the whole map yet just go into worldbuilder so you can see. Piece of cake.
Yes, I agree Iustus. It is a matter of taste. My question is should ALL the AIs be doing it always? I'm looking at my latest game and out of 40+ AI cities, only one doesn't overlap.
player1... I'm not advocating wasting and leaving gaps. I'm talking about reducing what I am seeing as significant overlap happening by default.
And I don't agree with your point about extra maintenance. Most any city will more than pay for itself. In addition, often it's better to work a specialist than to work a low-utility tile such as ocean or tundra, so simply working "extra tiles" depends entirely on what tiles are there. Anyway, this is all beside the point I was making.
Wodan
