Civilization 5 AI ? Civilization 6 AI ?

[cinattra]

3 movement points?!?! Hmmm... I think that is intriguing. Giving siege weapons 3 movement points would allow them to move and attack in the same turn.

 

[Ninakoru]

AI in Civ is interesting, but need a bit more work. There are some glaring issues there:

- AI is not able to use ranged effectively, many times it moves and do not fire. You can game this to no end. Not to talk about planes or ships, AI is terrible using those.

- The game have a lot of variety, but there are some cookie-cutter paths you should follow in order to compete with AI massive bonuses at higher difficulty levels. Most no-brainers are all related with science (whitch leads to the major problem of how science is so dominant), and AI will usually follow random paths that are somewhat covered with the massive game bonuses.

I think there's much work that can be still done without chaging the current AI philosophy, and definitively not improving turn times. Most of that time is used to movement animations and other misc issues, and not AI calculations.

I have lost faith at this point, the game balance and the AI are still pretty bad, and it seems Firaxis has not intention to take action. Can't think of Civ 6 right no, just hoping for some Civ 5 polish.

 

[feelfreetodie]

first rule is to make money. how do we make money? make the community happy playing our game. how do we make people happy playing the game, make them WIN. everyone hates to lose.

how many people win strategy games at the toughest levels? (i should have done some research before posting this.)
take totalwar franchise games, for me probably the easiest strategy game to beat at the toughest level ( shogun2,legendary). never played rome2, too many bad comments about it. if i remember correctly, only 1% have ever beaten the game at legendary by steam records. thats right only 1%. and totalwar ai bonuses at higher levels dont even come close to what civ5 ai has.

i think devs go by the numbers they see as fact, not by what a few here in the forums say. the very few talented players obviously would like a better AI but i think for a majority of the people the game is hard enough. i myself have never gone beyond monarch @civ4 or king @ civ5. for some reason i think i have the game figured out with 2 straight victories at king and then the game destroys me next time around. i hate this game LOL.

there is such a fine line between making the game enjoyable and unbeatable.

 

[PhilBowles]

how many people win strategy games at the toughest levels? (i should have done some research before posting this.)
take totalwar franchise games, for me probably the easiest strategy game to beat at the toughest level ( shogun2,legendary). never played rome2, too many bad comments about it. if i remember correctly, only 1% have ever beaten the game at legendary by steam records. thats right only 1%. and totalwar ai bonuses at higher levels dont even come close to what civ5 ai has.

From what I gather, TW on Legendary is identical to TW on Very Hard in terms of AI bonuses - it just adds player handicaps (ironman mode, battle realism) that a lot of players don't especially like (as I found out when trying R2 on Legendary because the game's so ridiculously easy, and I got bored even with that before finishing a campaign - yes, the bad comments about that game are generally spot on).

So it's not exactly comparable with higher levels on Civ V since many TW players who can beat Very Hard have little incentive to play Legendary, a mode that isn't really any more difficult, just less user-friendly.

 

[Drakarska]

What Firaxis ought to do is get in touch with Stardock. I hear their AI is phenomenal.

Agreed, and it is. Anyone who knows anything about Kael and his work (FFH fm civ IV) should be begging to have him work on CIV VI... Good luck with that one Firaxis

 

[Drakarska]

Galciv 2?


Sent from my BNTV600 using Tapatalk

I'll add 2 more to your post. And both are mods to BTS;

Realism Invictus 3.2, and PAE.

 

[feelfreetodie]

So it's not exactly comparable with higher levels on Civ V since many TW players who can beat Very Hard have little incentive to play Legendary, a mode that isn't really any more difficult, just less user-friendly.

shogun2, global gameplay stats
4.8% win on hard
1.9% win on very hard
it really is hard to believe that so few were capable of finishing a campaign.

 

[feelfreetodie]

lets take a scenario, people had computers that could handle any amount of calculations a game could throw at it, best possible video, a computer from 50 years in the future.

imagine a competent ai , one that played civ5 like a chess game, no mistakes. this would result in levels with no bonuses for the ai, the only levels would be bonuses for the human player. this would be 1v1. now what happens when you throw in one human player with 8 ai players, stnd map game. no chance at winning ever. unless you cheat with heavy bonuses.

 

[dexters]

I think the games with apparently good AI, hide the AI shortcomings behind good mechanics/game designs. For strategy games, this comes down to limiting the game scope.
There are very good chess AI's because the game mechanics and board limit the moves. There are however (i'm told) no good 'Go' computer AIs because (quoting wikipedia here:

" Although the rules are simple, the practical strategy is extremely complex....The game complexity of Go is such that describing even elementary strategy fills many introductory books. In fact, numerical estimates show that the number of possible games of Go far exceeds the number of atoms in the observable universe."

A 4x game like Civilization is not going to ever have a 'competent' AI because the AI is asked to wear several hats and like 'Go' the open nature of the game leads to so many possibilities that not only is there a computational limitation, the programmer also must anticipate and program for it which is simply not practical, though, we always want more expansions such that the AI is incrementally improved. (I am a big proponent of small $5-$10 AI DLCs that releases with wonders or new civ fillers, but with the focus being upgrading the AI)

In Civ6, What we can hope for is a fun AI on a appropriate difficulty setting that will make people really think about what they want to achieve in-game and throw up obstacles to their grand plans. More importantly, I hope we can avoid a buggy launch like what we experienced with Civ5 with an SDK and dll right at launch so hopefully the community can pitch in and improve it along the way pro-bono.

 

[PhilBowles]

I think the games with apparently good AI, hide the AI shortcomings behind good mechanics/game designs. For strategy games, this comes down to limiting the game scope.

Yes, this is precisely what Crusader Kings II does, and presumably other Paradox titles with similar engines and mechanics. It's also why previous Civ games are perceived to have had 'better' AIs - they had more AI-friendly mechanics, simple trade screen diplomacy that didn't need to account for civs' relationships with third parties, and a combat engine that allowed the largest stacks to dominate by sheer force of numbers even when poorly-constructed.

Also, there's less of a perception of "cheating" in past Civ AIs despite the fact that they got very similar bonuses, and this is largely because of the addition of global mechanics like the Civ V representation of happiness - in Civ IV and previous games AIs had happiness advantages over the player, but there was no global happiness number, and nothing that could be summarised as such in the demographics (which were instead represented as a proportion). When you see figures like 76 happiness in the "Happiest Civs" list, you know there's something 'fishy' going on - Civ IV was better-designed to hide its 'cheats'.

 

[LoneRebel]

More or less every strategy game gives bonuses to the AIs at higher difficulties. It's simply not yet feasible to program an AI that plays similarly to a skilled human player, except at something like checkers. I'm pretty sure that the more complex the game, the more difficult it is to make a good AI for it too.

In short, don't just knock Firaxis for "bad AI".

 

[Barghaest]

When you see figures like 76 happiness in the "Happiest Civs" list, you know there's something 'fishy' going on - Civ IV was better-designed to hide its 'cheats'.

I regularly see numbers in excess of that... a few times 200+ happiness. And it was MY empire, not an AI. AI regularly has happiness issues in my games (despite lower unhappiness per citizen - they get 10-20% reduction on unhappiness for population/cities in most difficulties I play) because it expands too quickly without knowing when to stop and relies too heavily on external sources of happiness (which can be easily removed).

The problem is they concentrate on making the AI to compete with the player over making the AI competing against the mechanics of the game to build an empire. The reason I can dominate the AI is I concentrate on developing my empire... not racing to beat the Joneses.

 

[bison21]

I am just wondering while reading this intriguing discussion- is this a discussion of how the AI should be better on deity, or in general? Don't forget that the vast majority of players do not move beyond emperor(if at all), and the worse you play, the better the AI is. The better you are, worse is the AI.

I am just thinking that the AI is endogenous: it is dependent on the ability of the player and also vice versa.

If we talk about better, I think "better maximization" would be the term. I think of MadJinns Poland Science LP where runaway culture egypt (I think) could have beaten MDJ but didn't due to poor management of tiles and specialists.

Other "annoying" aspects of diplomacy such as warmonger penalty, settle-cities-near-us and so forth can be fixed with adding temporal variables, which of course has already happened to certain degrees.

 

[mitsho]

It may be a good idea to what (the Civ6) AI shouldn't be:

• obvious (= too transparent for the human player)
• similar (= different AI's should differ in 'personality')
• going for the win OR role playing (which do we want? Or, see above, a mix?)
• play a different game (by getting huge gold or happiness bonus visible to the human player!)

That last point in my opinion requires that the development team thinks about the game mechanics in connection to the AI. The huge amount of gold for the AI in vanilla civ came from the fact that it didn't spend its gold on buying buildings (who then produced no upkeep and so grew the gold mountain even more).

So in short, to have a good AI for civ6, in my mind the game needs to be (rather) simple to start with.

 

[bison21]

Let me use this list to illustrate my point about the average player. I feel qualified to comment because I have played on prince and king for a year before I started to become ambitious about moving upwards.

I understand this is a list of what the AI shouldn't be, but this is also what is commonly critiqued about the Civ5 AI so my points remain the same.

It may be a good idea to what (the Civ6) AI shouldn't be:

• obvious (= too transparent for the human player)
  If the player does not use strategy guides, the game itself will not be obvious and thus, neither will the AI. If you don't know the different outcomes of choosing a particular tech to research, how could the AI's choice be obvious? It can't.
  
• similar (= different AI's should differ in 'personality')
  I think that they do. As a complete noob (settler difficulty) it didn't take me more than 50 turns to spot the difference between Alexander and Ghandi. Unless you specify what you mean by "personality", it is difficult to argue that it isn't already in the game, and the fact that certain civs have reputations here on the forums solidifies that point.
  
• going for the win OR role playing (which do we want? Or, see above, a mix?)
  I don't even understand how this is tied to the AI, nor how a strategy(!) game should implement RP elements.
  
• play a different game (by getting huge gold or happiness bonus visible to the human player!)
  As many have stated before, a computational simulation of human intelligence is not feasible and game difficulties have always been adjusted by adjusting bonuses.

 

[PhilBowles]

I regularly see numbers in excess of that... a few times 200+ happiness. And it was MY empire, not an AI.

But is that with the number of cities an AI typically spawns? I see those values in my games only when playing tall.

The problem is they concentrate on making the AI to compete with the player over making the AI competing against the mechanics of the game to build an empire. The reason I can dominate the AI is I concentrate on developing my empire... not racing to beat the Joneses.

Yes, this is very much an issue which is more specific to Civ V than to the Civ series more generally. There is far too much of a sense that the AI is out to get the player, not to look out for their own interests, and the way the victory conditions are structured doesn't help (it's not, at least in practical terms, mechanically possible for an AI to win a domination victory in Civ V because the mechanics require that the player is gone by that point).

I did see a few pleasing exceptions in past iterations of Civ V - I remember a game where Ramesses seemed as keen as I was to buy CS votes from Harald immediately before votes that would have given him diplo victory. But I went back to Civ IV recently and saw Genghis develop an empire to a consistent strategy - dominating first Korea, then Germany and taking them as vassals, before confronting his final neighbour - me. It's obviously an illusion, but it does a lot to give the sense that the Civ IV AI wants to win the game, not just beat the player, something too often missing in Civ V.

similar (= different AI's should differ in 'personality')

BNW was a huge step back in this regard, sadly. In the previous expansions the differences were so clear that some entertaining, detailed characterisations of the leaders formed the basis of two popular threads. Unfortunately BNW gives so much weight to ideology, warmonger penalties, and AI responses based on whether they like or dislike Congress resolutions, and programmed the AIs to be much less individual in their strategies (consider how many prefer cultural resolutions over Sciences Funding, for instance) that these differences become almost completely obscured past the early game and all you're left with is the rather basic couple of axes on which Civ IV AIs varied - how far and how much they expand, and how willing they are to go to war.

going for the win OR role playing (which do we want? Or, see above, a mix?)

The best efforts to make an AI "go for the win" are exactly what enhance the roleplaying - it's a false dichotomy. I want to feel I have a competitor. But, in a game with up to 12 players (and more on advanced settings), "going for the win" does not mean "beat the human", it means "win the game according to your victory strategy, and beat everyone else". This is what Genghis did in my Civ IV example - from the perspective of that stage in the game, with 2 of 5 civs under his belt and only one more neighbour, the Khmer were the next obvious target to achieve his objective. I didn't detect any sense that he considered this player civ any different from the two AI civs he'd already beaten. Civ V seems to do this to a much reduced extent, although it certainly tries in some cases (even though the AI is actually generally superior at getting non-military victories than those in past games). Instead we get a dichotomy that makes it jarringly evident that the AI knows it's an AI, it knows which other civs are AIs, and it knows which civ is a player.

 

[bison21]

Unfortunately BNW gives so much weight to ideology, warmonger penalties, and AI responses based on whether they like or dislike Congress resolutions, and programmed the AIs to be much less individual in their strategies

This is so far the only convincing argument regarding the personality issues. The ideologies and the world congress, while making a worthwhile addition to the game, "suppress" leader's flavours. However that also means that the game develops on its own rather than having the AI always say or do the same thing. I mean civs such as the Netherlands for example are less notorious for having a specific flavour, but through ideology and bickering in the congress they can become your greatest enemy or closest ally.

I don't want to glorify the AI, I am just very skeptical that there are straight-forward solutions to a better AI that does not bring a payoff with it of some kind.

 

[PhilBowles]

This is so far the only convincing argument regarding the personality issues. The ideologies and the world congress, while making a worthwhile addition to the game, "suppress" leader's flavours. However that also means that the game develops on its own rather than having the AI always say or do the same thing.

The unfortunate corollary of that is that instead the game always does the same thing - I find it much more tedious to play through endless repetitions of "three blocs square off and denounce/invade one another", especially when it ends up wrecking game-long relationships in favour of making lifelong foes allies because you share an ideology, or gamelong friends implacable enemies because you don't. I loathe the implementation of the ideology system - it's the one truly weak feature of BNW.

 

[Barghaest]

But is that with the number of cities an AI typically spawns? I see those values in my games only when playing tall.

I never go tall. Always build a huge empire (lots of cities as large as I can get them not merely 3-4 pop but around 12+)

For example, in a game I played Ottomans a while back I had around 30 cities and the smallest was 10 - but due to buildings/policies/religion all my cities were Happiness Neutral to a population of 14, 16 with circus, combined with my access to luxuries and CS alliances I had 60ish surplus happiness and was #1 in Approval Rating in Demographics (Russia was last past with a truly wide empire, several cities under 3 population and -22 Happiness according to the penalties to construction I could see with my spies).

In a later game where I was Mongolia, I had 19 Surplus happiness with 16 cities (most conquests) and the smallest was 12. Of the three remaining empires at this point in the game, I was the only Autocratic and was suffering -27 from Public Dissent due to influence upon my empire from the Order civilizations and STILL was at 19 Surplus Happiness with those penalties while the two remaining Order civs were at 6 and 2 Happiness with NO Public Dissent.

The unfortunate corollary of that is that instead the game always does the same thing - I find it much more tedious to play through endless repetitions of "three blocs square off and denounce/invade one another", especially when it ends up wrecking game-long relationships in favour of making lifelong foes allies because you share an ideology, or gamelong friends implacable enemies because you don't. I loathe the implementation of the ideology system - it's the one truly weak feature of BNW.

I actually like the Ideology system as it was implemented because prior to it, my games tended to be too passive later. With the WC and Ideologies, new "blocs" appear and stimulate dissent adding a new diplomatic hurdle to late games (or spicing them up with more wars).

 

[PhilBowles]

I actually like the Ideology system as it was implemented because prior to it, my games tended to be too passive later. With the WC and Ideologies, new "blocs" appear and stimulate dissent adding a new diplomatic hurdle to late games (or spicing them up with more wars).

I never used to have that issue: late-game war was commonplace in my older games.

My issue with the ideology system isn't merely in the way it distorts diplomacy (and indeed Congress resolutions - at least in principle tactical voting is an interesting development while it lasts, but even that is subsumed beneath the overriding influence of ideology).

It's the way it forces tourism spam. Despite the offence/defence idea of tourism vs. culture for cultural victory, your tourism output is more important as a defence against foreign ideology than your culture since overall ideological pressure is determined by the relative influence of civs with each ideology. A well-implemented game design either makes a resource available which you will prioritise for some strategies but not others (such as faith in Civ V), or which you can use in a variety of ways (you have to prioritise science, but can choose tech paths).

Tourism is both very deterministic in its accumulation, with little player input other than archaeologists, and forced on you by its overwhelming importance to ideology. This also makes ideology an "I win" for the player with the lion's share of tourism-related Wonders. If you aren't a tourism leader, all you can do is select the most influential ideology among everyone else's (and actively try to avoid being first to ideology), and become a bit player in the late game blocs constantly being bothered by allies' invitations to declare war on the other lot, and with no ability to direct the late-game diplomatic landscape, even if you're otherwise a world leader and positioned to win, say, science victory.