The AI isn't worse than in V, but VI is a much harder game for it to play

[jekke]

Well:
1. Considering AI dumb is quite weird itself, because it doesn't have and is not planned to have any form of intelligence.
2. That's generally the second and non-functional part of AI - the immersion. People complain about AI looking dumb - it has nothing to do with actual gameplay challenge.

Then what does I in AI stand for? Italiano?

more seriously, do you mean that AI moves which may look dumb (note how I have never called AI dumb! only its moves or actions) can in the long run be good for the gameplay challenge, is that correct?

 

[olduser]

I am against this if it means putting in mechanics that are uninteresting to the player. Remember that the primary purpose of a single player game is to give the human player an enjoyable experience. This means giving the player some challenges, some fun, some highs and lows etc. The game should be designed around the player experience. The AI is part of that experience certainly but the human player should always be front and center.

That is what the Civ4 modders did in the last ten years. Now you have mods with an endless amount of additional content like C2C and it's 930+ techs, 3000+ buildings.......

But they forgot about the AI and it's not really possible to finish a game because everybody gets bored way before finishing. Mostly because of bugs and the fact that the AI is really bad. They even consider 3-5 minute turn times to be ok.

If that is what you want try that instead of Civ6.

 

[Rathelm]

I used to hate how reliant Civ was on giving the AI flat-out bonuses, but to an extent, that's almost necessary for a decent game. I don't expect the AI to capably land eurekas, so I'm fine giving them a flat boost instead. I don't expect the AI to pull off strats like worker steals, so I'm fine giving them the production to build a couple extra workers.

I do hope when the game comes out (or when the balance dust settles) the AI bonuses don't take away too many gameplay options at higher levels. One of the sad things about Immortal/Deity is how many things in the early game are off limits to the player. Many wonders are impossible or near-impossible to build, and it's very difficult to found a religion unless you pick a civ like Ethiopia or Celts that comes with strong religion bonuses.

But we'll see. Something like "the AI doesn't produce enough units" should be easy to tweak, and will help with side problems, like getting dominated by barbarians. Other things, like sending settlers off without an escort or laying down the optimal city configuration, might be more difficult. I'm holding out for build updates between now and then.

I would think the holy grail is to make the game difficulty levels outright mimic human potentials. So at the higher difficulties you're just playing against an AI that is better at the game instead of one that plays the same as Prince, but with massive production/research/money bonuses.

 

[SupremacyKing2]

That is what the Civ4 modders did in the last ten years. Now you have mods with an endless amount of additional content like C2C and it's 930+ techs, 3000+ buildings.......

But they forgot about the AI and it's not really possible to finish a game because everybody gets bored way before finishing. Mostly because of bugs and the fact that the AI is really bad. They even consider 3-5 minute turn times to be ok.

If that is what you want try that instead of Civ6.

No thanks. And I never said that the game should just have a huge amount of content just for the sake of content. I was talking about game mechanics. My point is that it seems like some players don't want new ideas like districts, separate trees for tech and civics, governments etc on the grounds that the "AI can't handle it". With that attitude, we'd still be playing civ1. I want the new features that civ6 brings to the franchise.

 

[Septentrion]

I would think the holy grail is to make the game difficulty levels outright mimic human potentials. So at the higher difficulties you're just playing against an AI that is better at the game instead of one that plays the same as Prince, but with massive production/research/money bonuses.

Smart IA without tons of units doesn't seem to be option for medium difficult. That would be ideal for Prince. I'm never convinced that swarms are a prerequisite of intelligence.

 

[Biz_]

I never seen a single person complaining about "bad AI" to actually define what "bad AI" or "good AI" is.

cheating = bad
not cheating = good

there's more to it than that, but firaxis AI = cheating AI = bad AI
you can debate whether it's a F- or a F or a F+ or a D- based on its other merits, but it's just a cheater

 

[SupremacyKing2]

I don't think the civ6 AI looks super terrible but there is definitely room for improvement. From the previews, it does need to upgrade its units better and it needs to mount proper attacks. Upgrading units should be relatively easy to code for. Mounting proper attacks is the tough part because of 1upt. I do think the new movement rules in civ6 make things a bit harder for the AI since there will be terrain that a unit cannot move to when they used to be able to in civ5. With 1upt, attacking cities becomes a very complex puzzle because not just is there the terrain to consider but each time you move a unit, it changes what tiles other units can move to. So the puzzle keeps changing every time you move a unit. I am hoping the corps and army mechanics will help the AI if it can use them. To make the AI better at attacks, there needs to be fewer units on the map.

 

[Jabulani]

Considering how new is civ vi and its complexities I am pretty sure the vanilla will be plenty of exploits, unbalances and bugs.

The first and second patch will mitigate them all.

 

[Gort]

Considering how new is civ vi and its complexities I am pretty sure the vanilla will be plenty of exploits, unbalances and bugs.

The first and second patch will mitigate them all.

What, like in Civ 5 and Beyond Earth?

 

[Ulthwithian]

Well, I can say this from an AI / Expert Systems / Machine Learning / 'trying to get a computer to mimic human intelligence' perspective.

No phenotypically clear computable algorithm can avoid being exploited.

By 'phenotypically clear', I mean that the outcomes of the algorithm are understandable by the observer of the algorithm.

E.g., 'the AI attacked me. He seems to attack when he gets N units.'

By computable, I mean 'able to be executed by a computer'.

And algorithm I think is clear in this discussion.

This isn't even restricted to computers. This is an issue because pattern matching is a basic trait of human sentience. The only way around this is to conceal the pattern. Unfortunately, in games it is much more difficult than normal to hide the patterns. So you get exploits.

I'll also note in passing that it is by no means required to give a computer player of any game more information than a human player is given. That may be the standard these days, but that doesn't mean it is a requirement.

 

[Ovarwa]

Example of good game AI? 1830.

Like just about all game AI, its mistakes and tendencies are hardcoded, so after a while a good player can predict what it will do, take advantage and almost always win solidly at hardest difficulty.

But even then, there is a solid sense that the AI is playing the game. It knows how to lay tiles and stations to benefit its own position and mess with yours. It usually has a good idea about what stock to buy, what railroads are good, when to pass on buying. It can even do all of this on randomly generated maps.

A novice and even intermediate human player is likely to lose at the hardest difficulty, in the face of successful and continual AI collusion to hurt the human without unduly hurting themselves.

Programming an AI for Civ is not the same problem. Yet reasonable game AI does exist even outside of games that have received exceptional attention and research: Chess, Go, Reversi, etc.

Subjectively, I like a game's AI to the extent that I have the feeling that it is playing, and that I also need to play in order to win, even if I win far more often than I would against a real intelligence.

To the extent I feel like I am playing in a sandbox, I am better off playing Pokemon Go.

Example of a decent 4x game AI? Imperialism 2. I can crank the difficulty and still usually win, but probably not with a bad starting location, and not if I get too aggressive, and not if an AI just decides to declare war on me before I'm ready, and not if I don't *really play* using most of the tools available to me...

 

[phryxolydian]

As you said, programming an AI for Civ is entirely different than programming a chess engine. Chess is only 64 squares yet modern engines will still use a lot of your CPU to calculate further than depth 20 or whatever. I think, as others have said, the AI's current issues are easily fixable, it's just up to Firaxis to deliver. I don't really get a good feeling when reading modders' comments of "just give us the game and we'll fix it" but, it is what is...

 

[stealth_nsk]

more seriously, do you mean that AI moves which may look dumb (note how I have never called AI dumb! only its moves or actions) can in the long run be good for the gameplay challenge, is that correct?

The AI has 2 unconnected functions:
1. Gameplay challenge where AI is "active" part of game mechanics to let player use it. For example, AI need to sometimes attack player cities, so player use city defense mechanics. If AI doesn't provide challenge, it's bad.
2. Immersion, where AI tries to look like an opponent the player is wining against. Looking bad for AI breaks this part, but it's important to know what immersion is:
- Subjective.
- Not consistent as sometimes players expect AI to act as competent player targeting winning and sometimes as roleplaying character.

And yes, sometimes AI actions which look dumb are actually not, the reasons are just invisible to player. In Civ5 vanilla a lot of players complained about AI backstabbing at random, but that was just lack of information (however such situations mean bad AI design actually).

cheating = bad
not cheating = good

there's more to it than that, but firaxis AI = cheating AI = bad AI
you can debate whether it's a F- or a F or a F+ or a D- based on its other merits, but it's just a cheater

Since modern hardware doesn't allow creating AI which could somehow compete with human, in your definition Good AI = Weak AI = Not interesting game.

 

[Biz_]

Well, I can say this from an AI / Expert Systems / Machine Learning / 'trying to get a computer to mimic human intelligence' perspective.

No phenotypically clear computable algorithm can avoid being exploited.

By 'phenotypically clear', I mean that the outcomes of the algorithm are understandable by the observer of the algorithm.

E.g., 'the AI attacked me. He seems to attack when he gets N units.'

By computable, I mean 'able to be executed by a computer'.

And algorithm I think is clear in this discussion.

This isn't even restricted to computers. This is an issue because pattern matching is a basic trait of human sentience. The only way around this is to conceal the pattern. Unfortunately, in games it is much more difficult than normal to hide the patterns. So you get exploits.

I'll also note in passing that it is by no means required to give a computer player of any game more information than a human player is given. That may be the standard these days, but that doesn't mean it is a requirement.

there's plenty of fog of war in civ which would counteract much of this theory

Since modern hardware doesn't allow creating AI which could somehow compete with human

this isn't true
99.99% of humans are awful at playing civ. beating them is completely within reach of standard personal computers.

with cloud computing, you can reach the other 0.01%
or just small amounts of cheating, not the absurd levels of cheating that civ uses

 

[stealth_nsk]

this isn't true
99.99% of humans are awful at playing civ. beating them is completely within reach of standard personal computers.

with cloud computing, you can reach the other 0.01%
or just small amounts of cheating, not the absurd levels of cheating that civ uses

Only this year a supercomputer was able to beat one of the best Go player. And Go has way less variants than Civ.

EDIT: Not to mention it's just a on-goal for a game like Civ.

 

[Acken]

It must be notted that Firaxis also makes an AI capable of playing mods with a minimum of adjustments. This will always make it less efficient at playing vanilla only since they give it a lot of soft rules rather than hard rules.

 

[Biz_]

Only this year a supercomputer was able to beat one of the best Go player. And Go has way less variants than Civ.

EDIT: Not to mention it's just a on-goal for a game like Civ.

there's a big range of skills between best player in the world and how we'll be approaching a game for our first hundred hours with it

it's okay if the AI needs to compensate against someone who has dedicated his/her life to mastering a game.

 

[stealth_nsk]

there's a big range of skills between best player in the world and how we'll be approaching a game for our first hundred hours with it

it's okay if the AI needs to compensate against someone who has dedicated his/her life to mastering a game.

Ok, let's put it that way. Creating such AI will require neural network, because brutforce approach of Chess doesn't work even for Go. The game will cost about $10/hour to compensate cloud calculation expenses. The game will have no gameplay patches, because because training AI for particular rules will take months. The AI will have no roleplay, because it will be optimized for winning. Turn times will be about 10-20 minutes. In those conditions the AI will probably be able to compete with average civ player on equal terms, but I don't see any single reason for developer to try something like this.

 

[Biz_]

Ok, let's put it that way. Creating such AI will require neural network, because brutforce approach of Chess doesn't work even for Go. The game will cost about $10/hour to compensate cloud calculation expenses. The game will have no gameplay patches, because because training AI for particular rules will take months. The AI will have no roleplay, because it will be optimized for winning. Turn times will be about 10-20 minutes. In those conditions the AI will probably be able to compete with average civ player on equal terms, but I don't see any single reason for developer to try something like this.

you don't need any of that except for the top 0.01% of players
but keep pulling fake numbers out of nowhere

 

[stealth_nsk]

you don't need any of that except for the top 0.01% of players
but keep pulling fake numbers out of nowhere

Yeah I can. With working 10 years in software technical analysis these numbers are surely have at least some substance. Let's calculate. The AlphaGo in it's match against Lee Sedol used 1202 CPUs and 176 GPUs. If you use Amazon EC2 C4 instances (which are focused on computation) with largest discounts (reserved instances), that's about $50/hour for CPU only (without GPU) for 1 game against 1 player. Of course this could be somehow compensated by longer turn times (that's why I've put 10-20 minutes), but in general my numbers are more underestimation than overestimation.