Lets list AI stupid behaviors

I have what is probably a way too simplified solution to my "AI immersion" issue.
https://lh3.googleusercontent.com/-...7VoufBU/s3000/BNW%20Leaders%20Spreadsheet.png
If the Victory competitiveness trait were unilaterally dropped to zero , could that possibly stop the AI from "worrying" about winning , or not winning the game ,and , make it focus more on development and conquest for the sake of itself? , or , would it make the AI more sluggish ,by way of removing "motivation"?
Since the AI can't actually worry , or , be motivated , i guess i am wondering about how this game is actually built ,and , if such a drastic change would affect stability , and , if it could make the AI be perceived as behaving better by me , and all the other gripers who complain about Civ5 AI.

Victory competitiveness is no t about winning, its about competition for your specific VC. Diplo balance is about hating the winner.

The flavors that let the AI win are what you'd expect, the ones for each VC. Since the AI picks a grand strategy, none of it matters besides relatively.

Sent from my Nexus 5 using Tapatalk
 
Worked yesterday for lots of hours to finally get ranged units to move AND attack when targets a unit. Wasn't easy because I had to check first if the units is out of range, BUT, in range adding the moves, but with movement left to shoot, counting movement, terrain, and if on such tile the unit have line of sight, then select the best tile to do the repositioning, etc. Also I had to rearrange all the combat process so first melee units come in position, then queue all the ranged units that can shoot, then position all ranged that can move and shoot, queue those and finally come in with melee attacks.

Still I have to check out all the damage evaluation before executing the attacks, as all ranged out of range but in movement + shoot range may not be finally able to attack, but still counted as sure attacks for the evaluation.
 
- AI siege behavior
- AI diplomacy behavior
- AI espionage behavior
- Stupid AI missionary spam
- Overpowered Zulu
 
Worked yesterday for lots of hours to finally get ranged units to move AND attack when targets a unit. Wasn't easy because I had to check first if the units is out of range, BUT, in range adding the moves, but with movement left to shoot, counting movement, terrain, and if on such tile the unit have line of sight, then select the best tile to do the repositioning, etc. Also I had to rearrange all the combat process so first melee units come in position, then queue all the ranged units that can shoot, then position all ranged that can move and shoot, queue those and finally come in with melee attacks.

Still I have to check out all the damage evaluation before executing the attacks, as all ranged out of range but in movement + shoot range may not be finally able to attack, but still counted as sure attacks for the evaluation.

Very impressive work! Getting the AI to attack as you described would be an extreme breakthrough for the AI on the battlefield, and add a needed unpredictability to it.

How much longer does it take the AI to calculate all this?
 
Very impressive work! Getting the AI to attack as you described would be an extreme breakthrough for the AI on the battlefield, and add a needed unpredictability to it.

How much longer does it take the AI to calculate all this?

Even if it lagged a second or so for "thinking" , any positive , and , true AI improvement is appreciated.
 
Very impressive work! Getting the AI to attack as you described would be an extreme breakthrough for the AI on the battlefield, and add a needed unpredictability to it.

How much longer does it take the AI to calculate all this?

Yeah, I'm a bit worried about that, there's part of the calculation that is a recursive function to check all possible movable and shoot tiles, that may take a while, I'll try to make some optimizations not avoid too much calculations, stop at certain number of tiles checked or something like that.

I tested the thing on a duel map and I didn't notice any aditional lag due to processing, as there's no animations involved and only applies on units out of range but in move + range distance, it MAY not lag the game turns at all, still to be checked on a huge maps where processing times go up exponentially.

Wait one moment, Ninakoru, you can mod AI? You are creating AI mod? :OO

Yep, gone straigth to the BNW core dll and I'm modding it :)
 
When you will end this, talk with whoward about merging dll mods ;)

Yeah, at least there's a minimal code that's pure bug fixing, so that code should be in any dll mod.

Some findings I did on the AI tactical code:

- For each map tile there's a tactical consideration of tile danger, but that danger is calculated solely based on the unit strength divided by the turns will take the unit to reach the tile. As moving to water takes an entire turn using embark, that makes water tiles many times "safer" in AI thinking for land units. Also may explain why AI is so poor against artillery, as it has more range than movement, danger assigned is much less than real danger.

- Air strategies are quite messy, there's an air sweep and intercept missions but conditions are poor, and little to do with real necessities. Air sweep strategy looks only cities with fighters, and doesn't take into non-air units able to intercept. Intercept look if there's any air unit in 10 tiles (yes, fixed value) from the air unit witch may intercept.

Well, here's my progress:

- I optimized the move and attack for ranged, now I search for tiles in some area formed by the target and the unit itself, so no recursivity is used, should be a fast calculation.

- AI is much more aware of enemy ranged units, if the tile distance is equal or less than a unit range, it counts as if that unit could reach to the tile in the same turn.


I am reworking all the air sweep/interception/rebase handling. Air sweep action is now part of a normal attack action, so ater looking for units in range for the target, we check if air units will participate in such attack, and then look for air sweep capable units in range, we calculate max interceptions on the target and send that many air units to sweep on the target, then execute such attacks.

I'm still thinking how to fit interception with air sweep so AI could use both actions actively without one of them eclipsing the other when both actions could take all air sweep capable planes
 
In my latest game, I tried Emperor and had quite a hard time winning, especially since I didn't go warmongering too much. The AI started the space race and was actually racing towards the win. I had to push really hard to reach William's last space ship part in Rotterdam and cut off the rails towards Amsterdam just before it could be transported. But I was unable to keep up a decent army in his faraway hostile lands, against his decent army. So I eventually had to take Rotterdam, destroying the space ship part and enabling him to rebuild it in Amsterdam and make me lose the game... I didn't care, since it had been an exciting and unique finale :)

However: William spent his last 20 (!) turns waiting for me to finish my spaceship, while adding more gold to his already huge stockpile of 10000+ gold through 'producing' Wealth (as confirmed by my spy in Amsterdam).

Too bad... I had been impressed by the AI actually trying to win, thinking for sure he would resume building his spaceship. Maybe he "thought" the job was done after issuing build orders for all parts?
 
I am really enjoying your mod, thanks for your work!

Do you have any update? Something new for us to tryout?
 
AI, "hm... you got that luxury that I want. I'll give you 180 gold, at a rate of 6 gold per turn for the next 30 turns. But since we're friends, I could give you a lump sum up front, but for this convenience I'll have to reduce the total from 180 gold to 240 gold..." For a computer, you're really bad at math.
On a similar note, why does the AI, who have ridiculous happiness bonuses, so much so that they can spam cities willy-nilly with utter disregard to proximal luxuries, have such a vested interest in paying the human player for their redundant luxuries?
 
The AI, IMO, should always want to purchase the luxuries of a neutral or friendly player. I do agree that the value they are willing to spend could be adjusted, especially on the higher levels...and that lump sums should be > gpt trades.
 
Some weird AIs like to do this: denounces u, goes to war, but then got their butt kicked and lost 2 cities, immediately offer peace, and then, BEFRIEND and support everything u do.

Eh...hello? 10 turns ago u were freaking stealing my land with the Great Generals....and now u want friendship? NO TY!

2nd AIs weird traits: I want whale x1. I have citrus x1, salt x1 and marble x1. Can i trade 1 salt for 1 whale? NO!!!!!!!!!!!!! YOU GIVE ME ALL UR LUXURIES AND UR FIRSTBORN FOR HALF A WHALE
 
Some weird AIs like to do this: denounces u, goes to war, but then got their butt kicked and lost 2 cities, immediately offer peace, and then, BEFRIEND and support everything u do.

Eh...hello? 10 turns ago u were freaking stealing my land with the Great Generals....and now u want friendship? NO TY!

2nd AIs weird traits: I want whale x1. I have citrus x1, salt x1 and marble x1. Can i trade 1 salt for 1 whale? NO!!!!!!!!!!!!! YOU GIVE ME ALL UR LUXURIES AND UR FIRSTBORN FOR HALF A WHALE

Hahaha so true!!

Well, they show friendly because they are afraid of you. You recently wiped out their army, and AI show some respect to everyone with bigger army.

But that doesn't change the bad deals you will get for them because of denunciation, covering your land, etc.
 
I am really enjoying your mod, thanks for your work!

Do you have any update? Something new for us to tryout?

Well, I'm working on a custom version of the core dll, but has little to do with balance, just making the AI more efficient. However is a pain to test things out, as I have to create a new game with the dll and log everything everytime I want to test things.
 
This is not really a flaw, but it would be an easy thing to do to make the game harder: alert the AI when you build a spaceship part, instead of just when you hook it up to the spaceship. Right now everyone knows to just stock up all the pieces and hook them all up on your last turn. If the AI was alerted as each piece was built, maybe some of them would launch a last ditch war against you to prevent you from winning. Plus it's annoying to store all your pieces around the capital to prevent the AI from learning your plans, so this is a win-win.


One other thing I didn't see mentioned was ruins. I don't know how this is programmed or if this is even possible, but a lot of times later in games you will uncover a new ruin only to immediately have an AI unit take it. The problem is where the ruin was, there is just no possible way the AI would have not noticed it until that very turn you saw it. That kind of tells me the ruin isnt even being generated until you reveal the area from the fog, but I'm not sure how the code works. Either way, the AI should not be leaving ruins around for so long.
 
I think you are giving the AI way too much credit. Even when you attach spaceship parts as they are built, the AI rarely reacts. If an AI is also pursuing a SS victory, they may denounce you, but that's about it. Parking completed SS parts around your capital just uses scarce aluminum (aluminum is freed up when the SS parts are attached) and makes the parts vulnerable to destruction if an AI with bombers already dislikes you enough to DOW you.
 
it's not gonna be a vast improvement, sure, but it seems a lot more straightforward than a lot of other fixes. even if they rarely try to stop you, why not at least give them the chance?
 
I think you are giving the AI way too much credit. Even when you attach spaceship parts as they are built, the AI rarely reacts. If an AI is also pursuing a SS victory, they may denounce you, but that's about it. Parking completed SS parts around your capital just uses scarce aluminum (aluminum is freed up when the SS parts are attached) and makes the parts vulnerable to destruction if an AI with bombers already dislikes you enough to DOW you.

Does connecting a spaceship part increase score? If so, then doing it will make diplo balance civs like England hate you more. If not, then it will only be limited to victory competitiveness for the few civs (likely none) going for science.

Sent from my Nexus 5 using Tapatalk
 
Top Bottom