Sept 22 AI Discussion

[Firebug]

You know it, I know it, all us regulars at Civfanatics know it.

*cynics know it

FTFY
Seriously, tone down the cynicism a touch. feel free to be a cynic yourself, but don't drag literally everyone else down with you.

Part of Civ6's marketing has been the new engine, and the new AI and the Agenda system. Devs don't make new features if the AI can't handle it.

 

[Nick31]

*cynics know it

FTFY
Seriously, tone down the cynicism a touch. feel free to be a cynic yourself, but don't drag literally everyone else down with you.

Part of Civ6's marketing has been the new engine, and the new AI and the Agenda system. Devs don't make new features if the AI can't handle it.

You want HIM to tune it down when the devs made an entire game with 2 Expansions over several years and not one freaking time did they code an AI to move and shoot, i.e. a core, essential Feature of 1upt?

I'm a big Civ VI optimist, and proud of it, but unless your entire post is sarcasm, I am aggrieved on my "cynical" friend's behalf.

 

[phryxolydian]

*cynics know it

FTFY
Seriously, tone down the cynicism a touch. feel free to be a cynic yourself, but don't drag literally everyone else down with you.

Part of Civ6's marketing has been the new engine, and the new AI and the Agenda system. Devs don't make new features if the AI can't handle it.

I'm as excited to play the game as the next guy but, why are people so deluded? It is OBVIOUS that what x2Madda said is totally true, the AI will, for all the wonderful new mechanics Civ 6 will add, again be as incompetent as in past games.

 

[Brutus2]

The 2nd game was also played and not set up. For one thing, they said that one of Congo's cities had more war weariness because it was a conquered city, and Ed Beach said he can explain it because Kristiansan is not a congolese name... they laugh and say Pete has been a naughty person.

https://www.twitch.tv/firaxisgames/v/90778502?t=1h03m34s

They also take a look and discuss how the game has developed (Which they wouldn't if it was set up).

With regards to the Kurgan next to eachother: either the rules were changed, or the AI is not following the restriction which would be a bug.

Also interesting, I don't know if we'd seen this before but the Kurgan gives +2g +1faith, and if next to pasture +2g +2faith. This seems pretty good, interesting to see them have a non-militaristic option.

It's very possible that it was a combination of both.... Setup everything in the editor and then played a few turns from that point prior to the stream. Maybe he placed that city there and then attacked and annexed it specifically to show the difference in war weariness.

Obviously I can't prove anything but personally I feel confident most of these streams are setup in the editor to show something specific and you can't make assumptions about the AI from these.

 

[CaiusDrewart]

It's very possible that it was a combination of both.... Setup everything in the editor and then played a few turns from that point prior to the stream. Maybe he placed that city there and then attacked and annexed it specifically to show the difference in war weariness.

Obviously I can't prove anything but personally I feel confident most of these streams are setup in the editor to show something specific and you can't make assumptions about the AI from these.

I think the devs strongly implied the game was not set up. They looked at the score in order to see how well the human player was doing, and later observed with some surprise how strong Scythia's religion had turned out to be this game. Sounds to me like a real game.

 

[Seek]

Regarding the Warriors, no iron=no upgrades, maybe that has something to do with it.

If this is to be believed then the build is fairly recent, and likely to have had the pacing of the game done correctly in terms of age vs turn times.

Each era in Civ 5 takes about 50 turns on Quick.

He was researching PLASTICS in 1650 AD. which struck me as odd, so I presume Pete may have cheated a little. Or there might've been some sort of bug, since Scythia had a crazy faith bonus.

As far as I know, the Ai was on par, because Japan had Crossbowmen in 1650 AD.

Remember that unlike Civ 5 God and Kings/BNW, many units are missing, so there's HUGE gaps (once again) in the line units (Archer > Crossbowmen instead of Archer > Composite Bowmen > Crossbowmen

Even on Diety, players are able to beat the game and win a Science victory in crazy amount of time.

Good players reach industrial era around t120 in BNW and can often research Plastics by 135 or so with proper use of great scientists, Oxford, etc. Just going by typical MP games.

50 turns/era is quite slow. Don't know how the BNW AI performs on prince/quick, as I've never played with those settings, but I wouldn't imagine they are too far off what we saw today.

 

[Silverdawn]

I think the devs strongly implied the game was not set up. They looked at the score in order to see how well the human player was doing, and later observed with some surprise how strong Scythia's religion had turned out to be this game. Sounds to me like a real game.

On Reddit, Pete Murray said this about his game:
"No cheating here, but let's take a moment and marvel at how great that Colosseum location was."

The YouTube version is now available:
https://www.youtube.com/watch?v=P_F7_td7M5s

 

[Thormodr]

I'll reserve my judgement until the players that played today release their videos when their NDAs expire on September 29th.

 

[homan1983]

Pete Murrey seems to be a good player and quite intelligent. One of the reasons civ6 seems to be both more diverse civ UA wise and also much more tightly balanced.

I also love the fact that the civs aren't 1dimensional. Nothing is worse than civs that are great on one map and crap on another (looking at you polynesia). All civs, including naval leaning ones such as England, Japan or military ones such as Scythia or Rome are designed much more roundly.

The only exception so far seems to be Norway, but even then it might be their UU and UB are enough (as is the case with the faith generation of Scythia) to make them worth on Pangea and Continent maps.

 

[Xur]

You know it, I know it, all us regulars at Civfanatics know it. The AI will be as good as it was in civ5, civ4, civ3, civ2, civ1 and all the side games.

Its not that i think Firaxis is incompetent (they are very competent) but the AI has been poor in so many civ games (and even x-com which was a seperate team!) that wishing or hoping for better AI in the next game is just a pipe dream.

It has been discussed to death; the players that would apprechiate better AI are a minority. Plenty of players still lose on Prince and below difficulties so AI is good enough. As lazy and annoying as it is, if you want a challenge you will just have to enable AI cheats.

Agreed. I had no problem with the AI being a bit stupid from time to time with his units. My biggest problem with the AI was when I tried to engage in diplomacy. I think that the agenda system and casus belli will help a lot here.

I spent ~2000 hours in civ5, so I'm perfectly fine with an AI at civ5 level.

 

[Acken]

I'll launch an AI game in civ5 to see where they get at Prince in 1610AD for comparison.

Edit : Jesus didn't remember the base game prince AI to be so bad lol.

 

[Forsti]

Its not that i think Firaxis is incompetent (they are very competent) but the AI has been poor in so many civ games (and even x-com which was a seperate team!) that wishing or hoping for better AI in the next game is just a pipe dream.

Gazebo started the making of Vox Populi mod for civ 5 two years ago and with the help of only few people they have managed to improve both combat and strategical AI significantly, even though they have only used part of the time working on the said parts of code. The AI of Vox Populi can move and shoot on the same turn, it knows how to use navy, it doesn't run out of money and therefore it doesn't stop upgrading its units and so on. Mentioning only the most obvious improvements is dismissive towards the makers of Vox Populi, since the AI is all around a lot better than in BnW. The AI is so good that it's much more difficult to win as a human player than it is on BnW even though the AI gets much less bonuses.

If Firaxis with a team of full-time professional programmers can't improve the AI, while few guys working on their spare time can make a drastic change, there is every right to call Firaxis incompetent. I'd rather play modded Civ 5 with a capable AI than polished Civ 6 where the AI has warriors on modern era*. Therefore, I won't pre-order unless the videos that youtubers release on 29th show that the AI has indeed improved from what it was in Civ 5. Even if it costs me the Aztecs for 90 days if the initial reports of the game prove that the AI has actually improved and I thus end up buying the game, it is still a small price to pay compared to the risk of wasting 80€ on possibly crappy AI.

*Even if this was caused by AI lacking iron, it isn't acceptable. I consider it a design flaw if you can't upgrade the units you primarily use in the early game unless you happen to have a resource that you can't even see from the start.

 

[Stauffenberg]

I think we will know a lot more about the quality of the AI when the NDA is lifted on September 29th. Then we will get play sessions posted on youtube by quite a few people (Quill18 etc.). They would have started from the beginning and can show what things look like in the late game. If they too can breeze through prince level against the AI then we have something to discuss.

Do we even know that the games we saw today were run on prince and not chieftain?

 

[Mustakrakish]

Looks bad, VERY bad, but uh... What you see is what you get. It will get somewhat better with patches over time. It will get better with mods, which will take time.

So many people being in la-la land coming up with all sorts of excuses and magical reasons for it to be better on release is just hilarious. Keep it up! It's like you were born yesterday. Just how many games have you seen a month before release that got significantly better (if had any changes at all apart from cosmetic) by release I wonder?

 

[Seek]

I'll launch an AI game in civ5 to see where they get at Prince in 1610AD for comparison.

Edit : Jesus didn't remember the base game prince AI to be so bad lol.

Cool, thanks! And also, yep.

 

[Lord Shadow]

The truth is much simpler. Creating advanced AI actually 'good' at playing complicated games like civ is simply impossible on the current tech development level. No 4X strategy game has genuinely good AI. And we won't see any really impressive good AI unless some technological breakthrough.

Hell, modern games do not even have an artificial intelligence - just weighed lines of stupid blind code programmed to react on certain imput (sometimes with semi random elements) that is prone to the stupid simple bugs like math errors (because programmer was tired typing said code) that is supposed to make the passable impression of any intelligent process behind it.
Do not expect good strategy game AI for next few years.

I mean, what should we demand is AI that can manage game mechanics and create any (if predictable) danger for human player. But genuinely good AI, learning AI, AI using advanced tactics and traps, very adaptive AI, intelligent AI, AI inventing unorthodox ways - this is simply impossible at the moment. And certainly not in the game with extreme amount of variables, such as civ.

Firaxis didn't say anything about combat AI because they know they can't do much about it, and they can't say "we have passable mediocre ai like other games" for marketing reasons.

It's not really that technology's not up to snuff per se. The problem is the more demanding you make the AI processing, the longer it takes, and that means longer and longer AI turns. Not only is it hard to make a "smart" enough AI, it's very hard to make a decent one which processes fast enough not to become a detriment to gameplay.

So there's only so much you can do to optimize the "thinking" of an AI opponent. You need to simultaneously design your game so that it's more easily understood by the AI, too. This is something, say, the Beyond Earth team missed completely, while Ed Beach and his people, I believe, have put some serious thought into. Mainly when it comes to units and combat, given how 1UPT killed the AI's ability to conduct effective warfare.

I haven't been following the livestreams all that closely, but there's now Corps, which technicallly allow the AI to stack units or at the very least ease up traffic. And also the focus is more or less back on melee units: there's melee siege units (Battering Ram), and the average melee unit can attack cities without the massive penalties that dumbfounded AI in Civ5. At least in theory, these changes could potentially have a huge impact on how the AI civs deal with war.

 

[Acken]

Well 1UPT is what makes the AI slow during turns and is the main AI challenge. Things like city production, upgrading units etc take very little time to process and some of it is really straightforward.

The argument of difficulty makes sense for the tactical AI, it makes a lot less sense for the "economic" AI. Which is why seeing warriors in 1600AD is disapointing.

 

[Lord Shadow]

The argument of difficulty makes sense for the tactical AI, it makes a lot less sense for the "economic" AI. Which is why seeing warriors in 1600AD is disapointing.

Of course, but at the same time, playing on a higher difficulty level should mend that issue to some extent.

And at any rate, it's something that should be far easier to polish with patches over time.

 

[Acken]

Bonuses indeed help there but I believe its important to have a solid foundation of sound AI decisions before relying on bonuses. Otherwise you end up with some really weird situations. Also a decently made AI just responds better to bonuses and makes tuning difficulties an easier task.

It will indeed get better with patchs and mods.

When it comes to the livestream though the sky isn't falling. It's 1 month before release, it's vanilla and it's on par with civ5 no bonus AI (warlord level) (assuming the AI get no bonus on Civ6 prince). Although yes it's not something to be proud of... The good news is that you can mod the AI to make it better. I have almost finished the AI simulations on BNW Prince AI and my mod Prince AI to give some reference.

 

[Acken]

I'll launch an AI game in civ5 to see where they get at Prince in 1610AD for comparison.

Edit : Jesus didn't remember the base game prince AI to be so bad lol.

Here is what a 1610AD civ5 Prince look like. Prince AI has some minor bonuses iirc.

Spoiler :

The good news is that it can be improved through modding and balance. Here is a similar simulation with my mod :

Spoiler :