As a French person I feel like it's my duty to explain strikes to you. - AdrienIer

Create an account  

 
Civ AI Survivor: Season Five

It would be very interesting to know more about how the AI decides such things. It often seems like the AI is not really trrying to win, but rather just plays its preferred game plan (cultural/builder for Gandhi in this case) until it kind of accidentally stumbles into a victory. It is not really goal-seeking the way a human player would, which explains why it does not optimize things like running engineers for Manhattan Project rather than artists in its potential third legendary city.
Reply

(June 18th, 2020, 19:11)haphazard1 Wrote: It would be very interesting to know more about how the AI decides such things. It often seems like the AI is not really trrying to win, but rather just plays its preferred game plan (cultural/builder for Gandhi in this case) until it kind of accidentally stumbles into a victory. It is not really goal-seeking the way a human player would, which explains why it does not optimize things like running engineers for Manhattan Project rather than artists in its potential third legendary city.

I'm pretty sure this is well understood because in the Realism Invictus mod there is a checkmark "AI plays to win" and they will choose an appropriate win condition and ... er, win the game. They will burn captured cities, turn on the culture slider, build alliances, etc.

I don't know anything about modding, but if someone understands that mod I'd wager they can explain to us how the base AI plays to win.
Reply

So I'm not sure yet about culture, etc.

However, I do know that on rare occasions (this should really reveal how much i suck at the game) the AI will inform the player that: "We'd rather win the game, thank you very much."

This response is available by attempting to bribe a top-of-the-pack late-game AI into ceasing a war.

How dominant does this AI have to be? See this piece of code from CvTeamAI::AI_makePeaceTrade

Code:
int landRatio = ((getTotalLand(true) * 100) / std::max(20, GET_TEAM(eTeam).getTotalLand(true)));
if (iLandRatio > 250)
{
  return DENIAL_VICTORY;
}

It appears that although the devs put in certain safeguards to keep AI's on the path of domination, similar safeguard were not put in for space/culture, at least as far as I know.
"I know that Kilpatrick is a hell of a damned fool, but I want just that sort of man to command my cavalry on this expedition."
- William Tecumseh Sherman

Reply

(June 18th, 2020, 23:12)GeneralKilCavalry Wrote: How dominant does this AI have to be? See this piece of code from CvTeamAI::AI_makePeaceTrade

Code:
int landRatio = ((getTotalLand(true) * 100) / std::max(20, GET_TEAM(eTeam).getTotalLand(true)));
if (iLandRatio > 250)
{
  return DENIAL_VICTORY;
}

It appears that although the devs put in certain safeguards to keep AI's on the path of domination, similar safeguard were not put in for space/culture, at least as far as I know.

The AI is reluctant to trade away late-game technologies if it's committed to space.  (I don't know what the exact conditions for this are.)  This is partly so we don't get Civ 3-style shenanigans of paying tons of gold-per-turn to obtain space race techs.

Tech trading is banned in AI Survivor, though, so it never comes up here.
Reply

Since I wasn't able to put in my entry earlier, I thought I would post it and my thoughts for fun.

Winner: Saladin
2nd Place: Hannibal
First to Die: Ramesses II
Total Number of Wars: 14
Victory Condition: Spaceship
Victory Date: T345

Reasoning: I think this game can go almost any way and almost any civ could win it in theory, even the poor ones like Ramesses or Victoria since if the high peace weight civs hold out (such as Ragnar going crazy vs. the aggressive civs and dragging them down) Victoria could out-tech Pericles or the like. Personally, I think the highest probability option is Hannibal first and Kublai 2nd, but I think this map is very highly set up for Saladin to potentially win.

First off, Saladin has been pretty unlucky, for example his S1 Survivor match was insanely high leader quality and he held his own. Peace weight allows Saladin to go with high or low peace weight leaders but his lower slanting should help him avoid the ultra aggro civs. I also think he has a good chance of religious spread, in particular with Hannibal (which is my personal opinion) or Kublai. Ramesses is going to found his own competing religion and have all the low peace weight leaders who adopt Saladin's religion hate him while he hates them back. And while Ramesses has capital iron he has no nearby copper or horses. It should be noted this is also true for Ragnar, who only has one far source of copper, there's a good chance he launches a particularly inept early attack. I'm thinking he launches an attack against Kublai, Stalin 2.0 style, that drags them both down.

In addition, Saladin has a bomb opening start. His double corn capital, one irrigated, needs only on tech to start getting a huge bonus, and Saladin will almost certainly found a 2-3 gold 1-2 food resource city (both high yield food too!) while having first ring copper. He's going to have an explosive early tech rate, good production, and has both multiple early and Calender happiness resources nearby + almost certain religion that will allow Saladin to grow a lot better with the happiness crunch the AI usually has. I also think it is worth noting, for example, Kublai has a poor capital: double fish resources w/o starting with fishing, only one tile is improvable with mining so poor AI pathing makes him as likely as Saladin to waste early turns, mediocre nearby city locations. Hannibal has good expansion prospects with his choice of high peace weight punching back, I predict him to become good buddies with Saladin and basically conquer around the high peace weight civs with Saladin while Ragnar causes issues.

In the end Saladin, Hannibal and Kublai (Ragnar is too suicidal to survive) are alive without Saladin having enough to win by Domination. Spaceship with mediocre finish, Ragnar drives up the war count some but not too high.
Reply

It seems that these Cold maps are more cramped, giving some advantage to AI’s who end up with a roomier starting position.
Reply

Can I just say again how much I love Sullla's involved reporting? Most sportscasters should have public bets.
If only you and me and dead people know hex, then only deaf people know hex.

I write RPG adventures, and blog about it, check it out.
Reply

Could the EP difference between observer and AI be prompting AI espionage spending? Or is the AI unaffected?
"I know that Kilpatrick is a hell of a damned fool, but I want just that sort of man to command my cavalry on this expedition."
- William Tecumseh Sherman

Reply

Thats why when i test the AI myself, i always put no espionage.
"Superdeath seems to have acquired a rep for aggression somehow. [Image: noidea.gif] In this game that's going to help us because he's going to go to the negotiating table with twitchy eyes and slightly too wide a grin and terrify the neighbors into favorable border agreements, one-sided tech deals and staggered NAPs."
-Old Harry. PB48.
Reply

(June 19th, 2020, 15:39)GeneralKilCavalry Wrote: Could the EP difference between observer and AI be prompting AI espionage spending? Or is the AI unaffected?

No it definitely has a huge effect and has been brought up basically every season. I think it really skews things since once a civ hits a breakpoint against sullla in courthouses or whatever they can turn off their espionage slider and launch their research way above their competitors still wasting a fifth of their entire GDP on the observer civ.

Can't remember which poster has posted the text edit solution to this for several years and sullla keeps saying he doesn't want to mod the game.
Reply



Forum Jump: