1.2 AI improvements

Has anyone formed a view on AI improvements in 1.2? I've just finished my first 1.2 game, and it's hard to be sure because I like to choose different nations each time, but the AI certainly felt a lot more competitive in the early game. I played as Pariden on a medium randomly generated map with default parameters and four Ridiculous AIs (evenly balanced between "good" and "evil"). I started off in a good corner of the map with lots of good town sites. The early/ middle game (after I met the AIs) was hard though. The AIs seemed further ahead of me in development than I was expecting, given my good starting location. Also, my two closest neighbours were Resoln and Yithril, who liked each other and disliked me, and the map was not particularly kind in terms of good quests or bottlenecks, so I was always having to worry about fighting the AI, I didn't have much chance to gain levels questing and lair clearing.

Even after I got a halfway decent stack I had to be quite careful attacking the AI; I had to reload twice when I underestimated the strength of an AI stack. The first time I got crushed attacking the Resoln capital when the Oracle was present; I couldn't hit her high dodge troops, and she very effectively killed my troops with long range troops. I used Guardian Wind but it made very little difference, I assume because her troops were very high accuracy (85+). The second time I lost to Yithril with Verga present. Basically I couldn't deal with his high damage and high initiative troops (including two Juggernauts), not to mention Verga himself. Both times I had to reload and bide my time slightly, picking a weaker city to target for Resoln, getting an extra unit and using Freeze before attacking Verga. Even then the battles were close (I lost several units), and in both cases the following battle was quite tough as well (i.e. Varga came back with an army nearly as strong as his first army).

It's hard to be sure exactly what the difference was, especially bearing in mind the difference in faction (so this time around I had no Beastmaster, no free mana from Consume Shard, and limited offensive spells), but it certainly felt harder. One of the biggest factors was that troops seemed to be more buffed than I was expecting. I'm not sure if this is from extra unit buffs on higher difficulty levels, or from using fortresses to build troops, quite possibly both, both are mentioned in the changelogs. In general Resoln had quite scary dodge values, Yithril had much higher initiative than I was expecting, (in addition to the normal high armour, damage and hitpoints).

I became allies with Tarth and Altar, but this was partly because their armies were quite scary too, and I felt I had more than enough to be going on with fighting Yilith and Resoln.

Has anyone else had a similar experience? Do people generally think the AI has taken a step up with 1.2?

 

18,641 views 10 replies
Reply #1 Top

Yup, there was a definite increase in AI capability with 1.2. Can't wait for 1.3 to get finished for even moar!

Reply #2 Top

v1.2 is better than v1.1 in terms of AI.  But that was mostly fixing "low hanging fruit".

v1.3 has been a lot lot more work (which is why it's taking longer).  Path finding in these kinds of games is a notorious challenge.  Computing power hasn't done much to improve pathfinding other than make its stupidity work faster.  

Every time I work on a game I think THIS time, the path finding will be perfect because we can use X computing power and we'll design it elegantly and blah blah blah.  And then 2 years later, you'll see some stupid bug in it introduced by a human developers because there's some edge case where you need the unit to do X to avoid Y and in doing so, they didn't think of case Z.

At least humans can see the problem and adjust for it when they're playing. It's a lot more problematic for the AI.

Brian Wade (Civ V AI developer) and I have talked at some length on path finding and its affect on AI.  There is no "correct" way in doing it. I like my end turns to be fast, even on large maps and so I make trade offs that affect AI quality in order to keep the game moving.  But it bites me in the butt (and it did in GalCiv).  I try to make up for that weakness by having a multithreaded AI so the AI can develop a better strategy (path finding can't be done in the background while the player is moving).

Reply #3 Top

It seems better in 1.3 atm. So you did some good work ;)

Reply #4 Top

Tonight's check-in:

+ Improved algorithm that allows AI to detect different types of threats to its cities and units
+ AI prioritizes pioneers more when there is good land its LOS (and less when there isn't)
+ AI prioritizes maximizing its army size before going into battle
+ AI more willing to sacrifice non-champion led armies to soften enemies before sending in champions/sovereigns
+ AI now makes use of a "ready for battle" flag to determine which armies it should send where
+ AI is less paranoid about distant monsters and more cautious about nearby monsters
+ AI has less ADD when it comes to targets of opportunity
+ AI generally less cautious about taking on battles as long as its armies are deemed ready for battle (i.e. AI is less cowardly)
+ Fixed AI bug where an AI would target a quest in unfriendly territory and then abort destination because of that and get stuck in a loop (result: AI sovereign would just sit there)
+ Some AI personalities won't try to build improvements early on so that they can focus on expansion while others will do the opposite in order to build up their capitals faster
+ Improved the AI's decision making for deciding how many units should stay in a city to protect it
+ AI generally smarter about when to build pioneers vs. other types. SOME AI personalities will create more pioneers than others depending on what strategy they are employing
+ firxed AI bug where AI would sometimes not train units because it had improvements in the queue
+ Fixed AI pathfinding where the AI units could not "find" their rally point and as a result would no be able to concentrate their power.
+ AI will generally notify player when they've cast an offensive strategic spell on them (complete with snark).
+ Increased odds of AI interacting with player regarding world situation events
+ Improved performance of battle engine for autoresolve (makes AI vs. AI battles faster)
+ More AI weighting if players aren't geographicaly near each other (+1 to +2).

 

This should go in the next update.

Reply #5 Top

Wow, that looks really cool. Are those changes intended for 1.31 or for 1.4? In other words, are they going to be in the current beta, or are they going to be in the version after that?

Last night for the first time I can remember I saw an AI hero head straight for a goodie hut and grab it, which was quite satisfying, and I'm still only playing 1.2. It did seem bizarre the way the AI would just ignore them.

I guess we should be careful what we wish for, at this rate I might have to move the difficulty down from Ridiculous! At the moment I'm finding though that with every release the challenge goes up and it becomes worth playing the game again to see what's changed (and how I need to modify my strategy). In general I now find I have to be slightly less aggressive, softening the AI up where necessary, e.g. by letting them attack me so I can use strategic spells.

Reply #6 Top

They'll go into the final v1.3.

Reply #7 Top

Quoting Frogboy, reply 6

They'll go into the final v1.3.
End of Frogboy's quote

Shouldn't we beta test them first for you?  The current 1.3 beta AI doesn't work.  We would hate for 1.3 to come out and still have a non-functioning AI.

Reply #8 Top

Quoting Frogboy, reply 4
+ AI prioritizes maximizing its army size before going into battle
End of Frogboy's quote

Does this mean you have fixed the 60+ stacks of single units issue? 

Reply #9 Top

Yea, there will be another (I believe) 1.3 beta update.

Reply #10 Top

Quoting sjaminei, reply 8


Quoting Frogboy, reply 4+ AI prioritizes maximizing its army size before going into battle

Does this mean you have fixed the 60+ stacks of single units issue? 
End of sjaminei's quote

Hope so!