Unpopular opinion: people that blame AI don't play on deity

[Disgustipated]

I get the feeling the AI is working as Firaxis has intended. If they really wanted the AI to be able to beat the human player to the space race and other victories they wouldn't be introducing things like royal society, rock bands, and GDR. They really have no interest in programming the AI to beat the human player to these victories.

 

[CultureManiac]

I rarely jump into these threads, but ...

The problem is not specifically the bonuses the AI gets at higher levels, as that has been a hallmark of the system since the original Civilization - and I have played every version over that time. But something changed with CiV, whereby the AI became too passive at the end of the game and just let you win. In earlier versions, even if I was ahead in tech, I would have to prepare for an assault by a neighbor trying to burn down my capital to stop me winning. Since CiV, it's like even powerful neighbors just look over the fence and ask: "What are you building? A spaceship? Cool. Let me know when you're ready to launch it and I'll bring the family and some popcorn and we can watch you win the game together."

Without a threat at the end of the game, the last 50-100 turns become moot, just a tiring clickfest. Tension in the endgame could be maintained if we never knew whether we'd be gooned by an AI trying to prevent us from winning. Unfortunately, that threat rarely materializes, and is lame when is does.

Annual AI rant finished.

 

[PhilBowles]

I rarely jump into these threads, but ...

The problem is not specifically the bonuses the AI gets at higher levels, as that has been a hallmark of the system since the original Civilization - and I have played every version over that time. But something changed with CiV, whereby the AI became too passive at the end of the game and just let you win. In earlier versions, even if I was ahead in tech, I would have to prepare for an assault by a neighbor trying to burn down my capital to stop me winning. Since CiV, it's like even powerful neighbors just look over the fence and ask: "What are you building? A spaceship? Cool. Let me know when you're ready to launch it and I'll bring the family and some popcorn and we can watch you win the game together."

Without a threat at the end of the game, the last 50-100 turns become moot, just a tiring clickfest. Tension in the endgame could be maintained if we never knew whether we'd be gooned by an AI trying to prevent us from winning. Unfortunately, that threat rarely materializes, and is lame when is does.

Annual AI rant finished.

I actually found the AI an active opponent for peaceful victories in Civ V, and lost to science victories on Immortal multiple times (also occasionally culture and at least once to diplomacy). I remember races to build enough engines to get to Alpha Centauri first being a game in itself in Civ I and/or II, but don't recall anything similar in Civs III and IV. Civ IV - which I remember better than Civ III - was always domination or win through votes (and most often the latter - where the AI didn't exactly win 'deliberately' but just ended up voting for whoever it liked most); I don't recall losing to culture or science. But I did play those games on far lower difficulties than Civ V or VI.

And yes, I've lost to peaceful victories in Civ VI as well, if taken unaware (this is basically the only way to lose to religion outside a duel game, as I pay essentially no attention to the religious victory tab most of the time) or my lead is close enough that the AI gets a final piece ahead of me. So the AI will go for victory eventually - but it's extremely sluggish about doing so, to the extent that it can be several techs ahead and still gets its spacecraft underway after I do.

 

[MooFreaky]

Ok, now we're done with that, let me try flipping it around a bit.

A lot of people don't like the AI attacking them early , or killing CS"s, or backstabbing them. But these are all legal moves and not exploits at all.

Would you want a smarter AI going around burning down CS's that you are Suzerain of? Now, a smarter AI would require less bonuses and be less likely to rush early successfully though.

I would like to see it as one of a range of options for the AI, but done the same way a player would.
So the AI decides based on its surroundings, attributes and RNG effects that it is going to be aggressive early. So it focuses on doing that, building its units early and pushing the neighbouring CS or Civ, rather than wasting time with builders or districts. So it will be slower than currently as it doesn't have the frontloaded buffs, but it is better constructed and would be a real threat in the hands of a capable AI.

That way humans have time to play their own strategy, react to the AI's actions, etc.
It would also make recon units super useful, because you would actually need to watch the AI to see what they are doing and if you need to react. Coming from a StarCraft background recon was always something we needed to constantly, whereas Civ it has basically been relegated to "I know the neighbour is in that direction and that's all I need to know". Whereas a good AI that might strike at the CS's you are Suzerain of, or might be aggressive, or might follow one of several strategies needs to be monitored so you know whether to start reacting with units now, or you can risk powering the economy more. But not just because of what is happening next to you, but maybe Gandhi on the other side of the world has suddenly conquered Augustus and Montezuma, and will be able to power away from you if you don't act. Or Alexander has built his entire Civ around a cultural victory and you need to act now to hold him back. It would be a far more interactive game rather than the meandering and predictable slog it settles into now.

I know CS's will be conquered so I don't even care. But if it wasn't so standard and suddenly 4 or 5 were conquered by the same person, I would be engaged and wondering what was happening over there.


This would then extend into more involved diplomacy. A good AI would need monitoring, but you can't always do that. So maybe I can pay their neighbour to give me some information (maybe just about a city they can currently see. Or maybe trade a bit more and they will give me everything they know).

 

[ChocolateShake]

I get the feeling the AI is working as Firaxis has intended. If they really wanted the AI to be able to beat the human player to the space race and other victories they wouldn't be introducing things like royal society, rock bands, and GDR. They really have no interest in programming the AI to beat the human player to these victories.

It does seem like this is the case. Thinking about it, there's always a good feeling for most people when they win, new and casual players investing a lot of hours into a run and then losing isn't ideal. If that's the case though I hope they restrict this to settler/chieftain, and make it so the AI gets more and more aggressive about winning as you go up. Hopefully we can take a look at this more closely with modding soon.

 

[SupremacyKing2]

A good analogy for the AI is a 5K race. Imagine a 5K race that represents a civ game. Each runner is a different civ. The AI walks the entire 5K race whereas the human player tries to run the 5K race. At higher difficulty levels, the AI gets a massive head start in the 5K race but they still walk the entire race at the same constant pace. So even with the head start, the human who is running will eventually overtake the AI who is walking the race. Even when the human is in the lead, the AI does not try to walk faster or run but still walks the 5K race at the same pace. If you are a "casual" player who is also walking in the 5K race, the race may seem interesting since it will appear as if the 5K race is pretty competitive. Everyone is walking and may seem neck and neck. But if you are a "competitive" player or even a moderately good player, the race will seem rather pointless as the AI walks and you just run right past them every time. Even if you are not a good runner, the fact that you are running at all and the AI is walking, means that you will still overtake the AI every time.

The players who demand better AI are not saying that they want Usain Bolt to run the 5K, they just want the AI to also try to run or jog and not walk the entire 5K or at least try to run when the player is in the lead. And this analogy also explains why players enjoy the early game but get bored by the late game. In the early race, when you are still technically behind the other racers, the race might look competitive but once you are so far ahead that it is obvious you will win the 5K and you look back and see the AI still walking the 5K, you lose interest in the race. And not to sound super critical of the devs, but I think for some players, the new features are like adding a lemonade stand on the 5K race. An expansion is like setting the 5K in a beautiful national park. To the "casual" player, it is great. Now, they get to walk in a beautiful new setting and get some cool fresh lemonade during the 5K! But to other players, it is disappointing because while the beautiful setting is nice and it is nice to get some fresh lemonade during the 5K but it does not change the fact that the AI is still walking the race.

And to echo what @Disgustipated wrote, Firaxis want the AI to walk the 5K because they know that most civ players are also walkers or joggers. They add lemonade stands and parks in the 5K race so that the majority of players will have an enjoyable walk in the 5K. If they made the AI "run the 5K" most players would not enjoy it because the AI would always win. Who wants to run a 5K where you are jogging and everyone else is running and they get far ahead of you and you have no chance of winning?

 

[PhilBowles]

And to echo what @Disgustipated wrote, Firaxis want the AI to walk the 5K because they know that most civ players are also walkers or joggers. They add lemonade stands and parks in the 5K race so that the majority of players will have an enjoyable walk in the 5K. If they made the AI "run the 5K" most players would not enjoy it because the AI would always win. Who wants to run a 5K where you are jogging and everyone else is running and they get far ahead of you and you have no chance of winning?

An issue I have with this suggestion is that Civ V was massively popular (and as far as we know Civ IV was very popular - we just don't have comparable sales figures). Those games had better AI than Civ VI and the AI would sometimes win. Civ V was a huge drop in AI difficulty from Civ IV and seemed to hit the sweet spot, and as players apparently agreed based on just how long the game remained in Steam's most played list what motivation does Firaxis have for reducing the difficulty further? Civ V was ahead of Civ VI on Steam's most played list for most of the latter's life, and even now it's only a couple of spots below it.

What's more, for all the continuing issues with the AI it is a lot more difficult than it was at launch, so at least initially Firaxis obviously undershot where they wanted to be difficulty-wise. Remember when the AI wouldn't build units and wouldn't attack cities at all? Or when it couldn't fire with garrisoned units? I don't see any reason to think the AI has reached the point Firaxis wants yet. Civ VI's popularity has increased as the AI difficulty has - though since this period also includes a major expansion it's obviously impossible to conclude that it's increased because of the AI.

More likely Aristos is closer to the mark as he's spotted that the form of AI being used is cruder than that in the earlier games. If that's the case it may not get much better because it would require completely redoing it to see much improvement at this point.

 

[JBConquests]

Uhm, no offense, but why are you still here then, or rather, why did you come back just to post this? ^^" Did you just randomly check here and stumble upon this thread?

Good question, I haven't been on this site in years. I was looking at the winter sale and saw civ 6 expansions for sale and briefly wondered if they had enhanced the AI. I feel that urge to play civ again. So I stopped by this site to answer that question because I know this site is the best when it comes to civ. This is the first thread I happened upon and couldn't resist commenting. I see my answer is no. I'll keep an eye on this thread but will likely disappear again. I'll probably stop back when Civ 7 releases.

Edit: Great thread by the way. Love the discussion.

 

[Xmonger]

Google spent $10 billion on AlphaGo, if you give Firaxis that money they will may up with a fantastic AI. Unfortunately not every PC have a fast GPU for neural networks, and therefore only the highest-level computer owners can enjoy that AI.

Google uses TPU's not GPU's, and only for training AFAIK. Using the trained network is called inference and doesn't require that kind of hardware. Backprop is what takes the grunt. Otherwise I don't know about 10B, I think they paid in a few 10M or 100M for DeepMind, according to Wikipedia the hardware to run AlphaGo costs 24M (all custom).

Anyhow FWIW I'm a machine learning engineer atm. I wrote a post here about the difficulty with this, in my (possibly wrong because I'm just guessing on this without having spent the time required to try it out), the problem with a Civ DNN would be the number of input parameters. Go is a much simpler game, as the number of possible moves is much less than Civ - and the board size is much smaller. Civ looks intractable to me with today's technology, for DNN's.

 

[Trav'ling Canuck]

I suspect Civ 6 hits the mark for difficulty for Prince and King level players.

I'm not sure if they've undershot the mark for Settler to Warrior. That is, I don't know if anyone is playing these or not. My gut instinct is that they've moved Chieftain to Prince. That is, whereas in past Civ games, new players tended to need to start at Chieftain to be able to win, and then work their way up to Prince, now the game is calibrated so that you can start at Prince and have a decent chance to win out of the box. Firaxis presumably has some data on this, which would be more informative than my gut.

I'm more confident that they've undershot the difficulty level for Immortal and Deity. That is, they've left some potential sales on the table because Civ 6 doesn't offer the same challenge that past Civ games offered at this level. On the other hand, it's possible that they've gained some new sales because more players can now win on Deity, getting the Steam achievement that accompanies this, and that this is a driver of sales. So possibly, the current difficulty level for Immortal and Deity is part of their marketing strategy. Or possibly there will be an effort once the final game rules are set to hone the difficulty level and pick up sales from gamers looking for a tougher challenge.

 

[isau]

Mainly I think it's not even an AI issue, although that's part of it. The real issue IMO is how easy it is to heal a unit without expending resources. In Civ 4 when you took a rival city your army was almost always wittled down. In Civs 5 and 6 you can build an early army and then conquer the whole game with it. A better system would offer fewer opportunities for free healing. In Civ 5 and 6 you even see the units die when they are hit. Somehow they resurrect when you rest tho, as long as the last one isn't killed.

 

[Archon_Wing]

A good analogy for the AI is a 5K race. Imagine a 5K race that represents a civ game. Each runner is a different civ. The AI walks the entire 5K race whereas the human player tries to run the 5K race. At higher difficulty levels, the AI gets a massive head start in the 5K race but they still walk the entire race at the same constant pace. So even with the head start, the human who is running will eventually overtake the AI who is walking the race. Even when the human is in the lead, the AI does not try to walk faster or run but still walks the 5K race at the same pace.

That's a good analogy. Though I feel like here the AI doesn't even run the race but rather spends it time trying to trip you.

 

[Leathaface]

I quit playing civ a few years ago basically because the AI was to stupid. I get irritated/frustrated with AI's that win because they are outrageously cheating. My dream is to see a version of Civ that on Deity beats me (and most everyone else without cheating) badly such that I would have to turn down the skill level in order to compete. I know the difficulties in authoring an AI as I am an experienced software engineer but I also know that the AI could be better than it is. I may never see my dream achieved but it could be better. All they have to do is commit more resources to the AI and it can get better if they chose to make the investment. The company has no interest in making such an investment is the problem.

I just started playing and learning EU4 (Europa Universalis 4) recently and the AI is truly good. The game is cheating on my behalf so I don't lose so badly... It will be fun to see how good the AI is once I learn the game better.

I wish Civ's AI was as good as EU4.

P.S. I won't play civ again until I see some kind of news that they significantly beefed up the AI.

Have you tried Civ V with Vox Populi? The AI is very good in Vox Populi.

I've actually been fairly impressed by the AI's ability to go for science victory, but it's the most railroaded victory condition and the easiest to code. The problem is their progress through the tech path - when I check the stage 3 conditions AI civs have met, usually they'll complete one project and many turns later they'll have yet to finish researching either of the other requisite techs.

The amount of times i've seen the AI complete 1 or 2 Spaceport projects and then stall like they hit a wall all of a sudden, it's sorta disheartening TBH. I hope Firaxis has improved the AI to complete it's Spaceport projects quicker once it has started on then. (I doubt they will though.)

 

[JBConquests]

Have you tried Civ V with Vox Populi? The AI is very good in Vox Populi.

First I've heard of that. I looked it up online and it looks interesting. I'll give it a try.

Thanks for the Tip!

 

[Lily_Lancer]

Google uses TPU's not GPU's, and only for training AFAIK. Using the trained network is called inference and doesn't require that kind of hardware. Backprop is what takes the grunt. Otherwise I don't know about 10B, I think they paid in a few 10M or 100M for DeepMind, according to Wikipedia the hardware to run AlphaGo costs 24M (all custom).

Anyhow FWIW I'm a machine learning engineer atm. I wrote a post here about the difficulty with this, in my (possibly wrong because I'm just guessing on this without having spent the time required to try it out), the problem with a Civ DNN would be the number of input parameters. Go is a much simpler game, as the number of possible moves is much less than Civ - and the board size is much smaller. Civ looks intractable to me with today's technology, for DNN's.

You need people to do that. Senior machine learning researchers cost more than $1M each per year, and you may need many of them (for example, deepmind alphago involves more than 20 authors, not including engineers supporting them), that costs lot of money, too.

I don't think there's a "trained network" for Civ game and you only need to inference (also, inference is still not available for most PCs since you have to inference many times for every action of every AI each turn), and you have to train from scratch. That costs really a lot.

 

[FearSunn]

You don't really need neural network, machine learning etc. to make AI better. This regular non-senses from "experts" make me sick already. And this is the last time I am commenting on this fallacy.

 

[Xmonger]

You need people to do that. Senior machine learning researchers cost more than $1M each per year, and you may need many of them (for example, deepmind alphago involves more than 20 authors, not including engineers supporting them), that costs lot of money, too.

Yeah that's early days stupid prices because people wanted results yesterday. Today's open source libraries that we all use (TensorFlow etc) take the place of a lot of engineers. Keras is easy to start with. Most ML teams are much smaller, that example is the exception.

I don't think there's a "trained network" for Civ game

I was saying that if you trained a network (or rather if Firaxis did) then you only need normal PC resources on the client side to do inference. A typical gamers GPU/CPU should do fine.

and you only need to inference (also, inference is still not available for most PCs since you have to inference many times for every action of every AI each turn), and you have to train from scratch. That costs really a lot.

This doesn't make any sense.

Regardless the point is that in my estimation the problem isn't hardware or development costs but rather the more theoretical issue of feature size. The input parameters of the network look too large for present day capability. Just my half hour take on thinking of how I would tackle this, if I was actually working on it there probably are optimizations or localizations of the problem that would reduce the dimensionality.

 

[Lily_Lancer]

Yeah that's early days stupid prices because people wanted results yesterday. Today's open source libraries that we all use (TensorFlow etc) take the place of a lot of engineers. Keras is easy to start with. Most ML teams are much smaller, that example is the exception.



I was saying that if you trained a network (or rather if Firaxis did) then you only need normal PC resources on the client side to do inference. A typical gamers GPU/CPU should do fine.



This doesn't make any sense.

Regardless the point is that in my estimation the problem isn't hardware or development costs but rather the more theoretical issue of feature size. The input parameters of the network look too large for present day capability. Just my half hour take on thinking of how I would tackle this, if I was actually working on it there probably are optimizations or localizations of the problem that would reduce the dimensionality.

Thinking of OpenAI dota2 bot. They invest a lot of people, resources and time, and result in a failure. And that's in 2018. In fact Deepmind is lucky, even if it spend a lot of money, it succeeded.

 

[MarigoldRan]

Most people don't like it when the computer is smarter than them.

"You're dumber than a robot. Ah ha ha ha ha."

 

[_ViKinG_]

Most people don't like it when the computer is smarter than them.

"You're dumber than a robot. Ah ha ha ha ha."

But many dont like it when they are not good either. thats why whe have different difficulty levels to choose from for all players. Im thinking most of the bad AI at war