AI?

[stealth_nsk]

Have you played Vox Populi? Gazebo and Ilteroi are the living proof that you don't need anything crazy in terms of resources or money to achieve an excellent AI.

Oh, and they did it for free, and in their spare time.

Shame on FXS, I guess...

I believe I've posted similar comment before - define "excellent AI". It's not that easy as it sounds.

Also, just to note:

1. Civ5 had very limited resources. Most of vanilla AI programming was done by Jon, the lead designer. Investing more resources in AI development wasn't difficult task.

2. Jon made several mistakes which he described publically. It wasn't that difficult to follow his steps.

3. One of the problems you had in AI game development is constantly changing rules. You can't make AI too complex as it will be impossible to adapt. Mods on already complete game have much less balancing iterations, so they could afford more complex AI.

 

[Staler87]

It is possible for AI to beat humans, I mean there is a program that is better than any man alive at chess. It's just not feasible for a game due to a couple reasons:

Programs like these take the top computer scientists years to design

Chess is a bit simpler than civ

The program had to play the game repeatably to get good at it

It would take way too long to sim turns

Plus it would, theoretically (after many games), be a perfect AI and thus impossible to beat.

 

[stealth_nsk]

Chess is a bit simpler than civ

This class of tasks is solved by bruteforce (with some optimizations, but still), so with complexity of calculations growing nearly exponentially with the amount of tiles, "a bit" means "several billion times" here. Games like civ have to use completely different approach to AI, which is not designed to beat human, it's designed to simulate sane behavior.

Another point you forgot - chess rules didn't change for more than hundred years. Rules for a game like civ change during development roughly once a week during development and once a month after release.

 

[Staler87]

This class of tasks is solved by bruteforce (with some optimizations, but still), so with complexity of calculations growing nearly exponentially with the amount of tiles, "a bit" means "several billion times" here. Games like civ have to use completely different approach to AI, which is not designed to beat human, it's designed to simulate sane behavior.

Another point you forgot - chess rules didn't change for more than hundred years. Rules for a game like civ change during development roughly once a week during development and once a month after release.

That's why I used the keywords theoretically and not-feasible. I mean we have the technology and expertise to do it, just not the time and maybe not the computer specs.

The a bit part was a purposeful understatement. I know it is a bit more than a bit.

 

[Aristos]

I believe I've posted similar comment before - define "excellent AI". It's not that easy as it sounds.

Also, just to note:

1. Civ5 had very limited resources. Most of vanilla AI programming was done by Jon, the lead designer. Investing more resources in AI development wasn't difficult task.

2. Jon made several mistakes which he described publically. It wasn't that difficult to follow his steps.

3. One of the problems you had in AI game development is constantly changing rules. You can't make AI too complex as it will be impossible to adapt. Mods on already complete game have much less balancing iterations, so they could afford more complex AI.

Excellent AI in this context: an AI that is tenfold superior than what Shafer and his sucessors were able to create.

As for Jon, I don't want to even start talking about him and his "work". My opinion about all that has been posted more than once, if you want to know it, just look up my posts.

 

[stealth_nsk]

Excellent AI in this context: an AI that is tenfold superior than what Shafer and his sucessors were able to create.

As for Jon, I don't want to even start talking about him and his "work". My opinion about all that has been posted more than once, if you want to know it, just look up my posts.

That's pure emotions. No real arguments here.

 

[Aristos]

That's pure emotions. No real arguments here.

You are, plain and simply, wrong. I will not repeat my arguments, I told you. Go and look for them if you want.

 

[stealth_nsk]

You are, plain and simply, wrong. I will not repeat my arguments, I told you. Go and look for them if you want.

I mean how you could argue about "better" AI if you don't have any definition of "better". And you reply with something like "better means 10 times better".

 

[Aristos]

I mean how you could argue about "better" AI if you don't have any definition of "better". And you reply with something like "better means 10 times better".

Oh, OK... do I need to list the dozens of details that make the Vox Populi AI ten times better than vanilla BNW? I don't think so. If I remember correctly, you are usually around in the VP forums, so why do you ask for details you know very well if you play the mod? Is it a rhetoric question? Do you want me to give details for those that don't know the mod? If that is the case, then I invite all those that don't know it to go to the VP forums and try themselves.

You know very well why the VP AI is ten times better than vanilla. Why are you asking for details then? I still don't understand your "hidden agenda". Are you the sentient AI of Civ 6 that your praise so much? That would explain things...

 

[stealth_nsk]

Oh, OK... do I need to list the dozens of details that make the Vox Populi AI ten times better than vanilla BNW?

No, i don't. I'm asking about how you measure whether the AI is good or not.

There's quite common misconcept of "good AI" equaling "smart AI", where "smart" means being able to reach its gameplay goals. This approach was already proven wrong at strategic level, there effective AI was marked "gamey" and not interesting.

The only real measure of the AI quality is how fun is to play against it. It has nothing to do with smartness. Packman ghosts are absolutely mechanical, still within Packman rules it's fun to play against them.

So, things which matter:

1. Ability for AI to provide challenge. If defending a city against AI don't require any efforts from human player, that part of AI is bad. That part of Civ5 AI us bad. Funny thing is - even without knowing AI details, we could assume Civ6 AI is better here, because when defending city in Civ6 you need to protect districts as well, which requires more efforts from player.

It's important thing as it demonstrates what AI is just one of the gameplay systems. It's not necessary to improve AI coding itself to make playing against AI more fun.

2. Variety of tactics. Again, that's the area where Civ5 AI was weak, but it's not weakness of AI itself. With ranged units dominance the game had 1 primary tactic to use and AI was just unable to counter it effectively.

For Civ6 we could hope for general tactical improvements first, so there will be more than 1 way to fight against AI. I could say the retreats we saw in gameplay videos are really good, because they look like a thing which could help a lot in the area. Of course it's too early to tell.

3. Immersion. AI doing obviously stupid things like going back and forth don't look like a serios opponent, so it becomes less fun to fight against.

So, that's my definition of good AI with breakdown, explanation and samples. What's yours?

 

[qwerty25]

*snip*

So, that's my definition of good AI with breakdown, explanation and samples. What's yours?

Not the person you're replying to... but:

Honestly, if the AI can at least move then shoot with ranged units and place melee units in the front of them I'll call it a success.

 

[ftl]

I'm just jumping in to this discussion.... The way I'd think about it: the AI needs to be good enough that when a human player beats it, they feel like they outsmarted it rather than exploited it. (It would be your #3, immersion).

In Civ 5, there were a bunch of things that felt like really dumb ways of beating the AI. After the first few times, I didn't feel particularly smart when I took an army of 6 ranged units and destroyed any huge invading armada without losing any cities or units... it just felt like the AI had no idea what it was doing.

Diplomatically, it always felt dumb that I could extract vast quantities of gold from the AI by selling it resources I didn't need and it didn't either. Felt like an exploit and not a reasonable strategy.

It's not really about how good the AI is in an absolute sense; it's pretty easy to make an AI appear 'better' by giving it bonuses at higher levels. I'm sure that the AI was swimming in gold despite giving me like half of my gpt income in trade. But the goal should be to make the strategies that beat the AI feel smart and not lame. That's a pretty low bar, and Civ 5 didn't really meet it, hopefully Civ 6 does...

 

[stealth_nsk]

Not the person you're replying to... but:

Honestly, if the AI can at least move then shoot with ranged units and place melee units in the front of them I'll call it a success.

Well, that's surely not enough for me

 

[Gort]

Not the person you're replying to... but:

Honestly, if the AI can at least move then shoot with ranged units and place melee units in the front of them I'll call it a success.

With the unit balance of Civ 5 being what it is, a good AI would be one that has a massive army of ranged units with a single fast unit for capturing cities.