Artificial Intelligence

My understanding of Thyrwyn's post is that The Operational AI may sub-divide the units into "Western Front" and "Eastern Front" and each front would have its own "Tactical AI" that directs the units within its Operational front; almost like you see currently within HoI:3 (the WWII game). Such a sub-division strategy is probably but whether it is being used we have no idea (but it is not precluded by the list you quote).

If it does then the operational AI would need to be able to reallocate existing units between sub-Fronts in order to operate effectively - but would be a complex trade-off mechanic (sacrifice the east for the west or put the east into a "holding action" while the west attacks).
That was indeed my point. Although, as Kozzer said, we are all just "thinking into our keyboards" at this point.
 
I've said this time and time again but here it is. Their are two different ways to create an AI for civ. 1) creates an AI that role plays a historical view point, values friendships and its citizens lives, ect... ect...
2) create an AI that plays exactly like MP and wants to win at all cost. Ie, back stabs you if you're about to win a culture victory.

You can have one or the other, but not both. Personally, I prefer 1. More immersive.
 
Well Sid at GDC was clear that he and Firaxis aim the AI to not beat the player but provide a challange, but too high of a bar, the player has to feel that he can win with a reasonable effort, so the AI can't be dumb or the player loses interest, and the replayability is gone. But the AI can't be to hard or the player never gets threw his first game and never plays again. A fine balance for a game especially in one with so many different facets as Civ.

CS
 
Actually, my feeling is that the AI will behave in the way that JonoLith is describing while giving the gameplay effect that Psyringe is describing. It really comes from the fact that the AI will know its own strengths, based on Civ special ability, UUs & UBs, map location, etc. If the AI is done well, it will cater its game "management" toward maximizing those things, which means that and Egyptian AI should build lots of wonders, Monty will act like the Monty we all know and love, England will dominate the seas, etc.

Also, while I know they've explicitly said that there are no more "leader personalites", they at other times have also talked about some of the leader personalities in the game - I remember the talk about Napoleon specifically.

So, as long as the special abilities are appropriate for the civ (and for the most part they really seem to be), and all the other factors are appropriate, then you should get a more goal-oriented AI but which is less determined by essentially random factors such as religious affiliation (since religions are out) and so on. Since the AIs will act to take advantage of the advantages their specific civ endow, they'll play in the style that their civ is designed to engender.

Edit: tl;dr - if the game designers create approprate civ-specific attributes for each civ, the AI should be able to accomplish both an immersive "alternate history" experience, as well as play the game in a goal-oriented manner (i.e. try to win the game).
 
I've said this time and time again but here it is. Their are two different ways to create an AI for civ. 1) creates an AI that role plays a historical view point, values friendships and its citizens lives, ect... ect...
2) create an AI that plays exactly like MP and wants to win at all cost. Ie, back stabs you if you're about to win a culture victory.

You can have one or the other, but not both. Personally, I prefer 1. More immersive.

Incorrect!!

You can have both. Just make two buttons letting you choose between a historical Civ or a Civ that only tries to win. Of course making the AI's is a problem seeing that making 2 different AI's would take a long time to make. (Which can be modded in of course)
 
They should probably add an additional ending type, "Draw", which is neither a victory nor a loss. Anyone allied (or at least very friendly) with the ultimate victor (especially or at least in a diplomatic victory) would be considered to have won a draw while everyone else would be "defeated".

Both AI and humans could be programmed to accept and try for a draw if outright victory is unlikely or impossible. In the "trying for a draw" state the AI would support their leader civ and vote for them in order to get a diplomatic victory before someone else gets spaceship or culture.
 
I'm one of those who want an AI that "plays to lose" or "plays to be fun" if you want to phrase it like that. I still want a challenge. The challenge I seek is written on the box: "Build an empire to stand the test of time". I do not seek the challenge of "Beat the AI".

I think this is the key point for me too. I prefer "realistic" empires to "gamey" ones.
 
They should probably add an additional ending type, "Draw", which is neither a victory nor a loss. Anyone allied (or at least very friendly) with the ultimate victor (especially or at least in a diplomatic victory) would be considered to have won a draw while everyone else would be "defeated".

Both AI and humans could be programmed to accept and try for a draw if outright victory is unlikely or impossible. In the "trying for a draw" state the AI would support their leader civ and vote for them in order to get a diplomatic victory before someone else gets spaceship or culture.
They already do have a "draw" - it's called a "time victory"! :p
 
Also, if the computer players automatically attack you whenever you're about to win a Science or Culture victory, then all victories are really military victories.
You can certainly have an AI that "tries to win the game" without having it resort to Mafia tactics like this! In fact, you'd probably have to explicitly code that behavior into the AI. I know they talked about the AI attacking a Civ as it's building spaceship parts and transporting to the capital, so maybe they did put that in there. Or maybe that's the only victory condition they coded that for. I find it hard to believe that they'd code the AI to start bombing your wonder-filled cities as you approach a culture victory. But maybe they did (if anyone's seen anything like this, I'd be interested in a link).
 
Also, while I know they've explicitly said that there are no more "leader personalites", they at other times have also talked about some of the leader personalities in the game - I remember the talk about Napoleon specifically.

I had though they implemented a system were each leader was given weights in various categories for ai behavior, and each game the weight were randomedized by +/- 2.

So you could have a very warlike Aztec (10) up to a relativity passive one (6).

Is this now gone?
 
I had though they implemented a system were each leader was given weights in various categories for ai behavior, and each game the weight were randomedized by +/- 2.

So you could have a very warlike Aztec (10) up to a relativity passive one (6).

Is this now gone?
No, that's exactly what I was referring to with regard to Napoleon. But then I read that leader personalities are gone, which I take to mean that there was some other implementation in Civ4 that isn't going to be in Civ5. Though what you described is definitely in Civ5 based on what I've read.
 
On the contrary, I think huge stacks are MUCH easier for the AI to manage than 1upt units. But I hope for the best.
They aren't. Mainly because a huge stack is best countered by a bigger stack which itself is counterd by a bigger stack and so on and so forth. This forces the AI into a never ending arms race on the global level, we all know how well that works out in the end :)

At the tactical level a stack is definately not easier for the AI to manage than a collection of single units spread over multiple tiles. Consider the most basic decision- do I attack? Some of the units in the stack may have a bonus or penalty for a particular tile so when is it a good time to attack? Is one units penalty small enough that it is overcome by another unit's bonus? These things are easy for you and I to spot and decide on but teaching a computer to understand them, especially odds and acceptable losses is quite complicated.

Several of these decisions are greatly simplified when you can't stack units because the AI now only needs to consider the effect of moving a single unit. It's simple to decide if a tile is good for defense when you only have one unit to place there.

I'd say it's a wash really, from the AI's standpoint anyway. However stacks are definately not 'MUCH easier' for the AI.

The main hope I have for better AI is a computer that doesn't have to cheat as much to provide a challenge.
It'll still have to cheat unfortunately. As nice as it would be to have good AI that didn't need any help I don't think it'll happend any time soon. One of the biggest problems is teaching the AI to understand what a player can figure out in a few seconds of looking at a map.
 
If the goal is to make the AI play as a human would, trying to win at any cost isn't exactly the correct approach. Going for the win isn't always the best thing to do.

Take for example a situation where you are playing a game with several opponents and are currently sitting in second place. Going after the leader might damage them, but in some situations it could hurt you just as much, if not more, thus allowing your other opponents to gain on both of you.

In boardgame circles this is refered to as king making, and is generally seen as poor play.

Instead, in the above scenario, it might be better to go after the person in 3rd place to try to secure your 2nd place finish.

In other words rather than simply try to win even when it's not feasable, each player's goal should really be to place as high as possible.

Hopefully, the AI is aware of that.
 
There's a few points I think is truly important for the sake of clarity.

1) The Game should never be built so that it is the AI versus the Player. The Game should be built so that it is the AI versus the Player AND all other AI.

In general, all the conversation that I've been seeing here has assumed that the AI will work together to see the demise of the player, and that is the only way it can be. If such is the case, then I make the argument that the AI has been built poorly, and is a failure. The AI needs to constantly be aware of it's own self interests, and aware that pouring resources into destroying one Civilization will hamper it's own ability to pursue it's goals.

2) The AI should pursue it's own goals, and pursue them well.

The wide concern that people have been throwing around is that all AI will simply "nuke" you when you attempt to win any kind of non-military victory. (I assume "Nuke" means "Attack aggressively.")This carries with it a few assumptions.

A) It assumes that all AI will have access to the Units/Technology available to be able to successfully wage a war.

To put it bluntly, if they don't have it, they don't have it.

B) It assumes the AI is willing to drop whatever it was doing in order to wage a war, rather then pursuing it's own victory.

For example; If an AI is pursuing a Cultural Victory, and you are pursuing the Space Victory. You begin building components, it has been "researching" social policies. Does it make sense that the AI move away from pursing that victory to ensure that you won't get yours? Or does it make more sense that they would continue pursuing their goal, confident in their success. Why make a major shift, if it doesn't have to?

C) It assumes that the AI is stupid.

If I see all my neighbors around me start waging war against a guy who's started to build space ship parts, my inclination isn't to join right in. My inclination is to wait for my neighbors to commit to full scale war, and then attack them when they are at their weakest. If the AI is a warmongering AI that's pursing a Military Victory, this is the course of action that makes more sense.

D) It assumes the AI is cheating.

I will admit that there is definitely precedent for this claim, even within this franchise. However, it is poor to simply assume that the second you start building any space components the AI will always 100% of the time be able to, with pin point accuracy, and flawless execution, nuke the exact specific tile where your component is at currently. If such were the case, then the AI would be a failure.

3) A distinction needs to be made between difficulty levels.

This is where the major breakdowns are starting to happen, I feel. There is a group of people who would like an extreme challenge, with an AI that feels and plays like a person, and there is a segment of people who want to simply experience the game world as it is, breathing in the reality of their game. I strongly believe that this is specifically what the difficulty levels are for.

If you wish to play a relaxing game where the AI plays more lax, then put the difficulty on Settler. If you wish to encounter a difficult challenge where a group of cutthroat AI work to secure their own success, then play Immortal. Or any range in between. Find which setting is within your comfort zone, and play at that setting.

Neither side of this conversation is in the wrong, and both are looking for the same thing, ultimately. Both wish to enjoy themselves while playing Civilization 5. I strongly feel that this is where the difficulty level will facilitate both groups.

It is very important to note, that I do not hold up the idea that the AI should be trying to win at the cost of people's enjoyment. If you do not wish to play against an AI that is trying to defeat you, then don't. Enjoy the game how you wish to, but please allow me to enjoy the game how I wish to. And my desire is that the AI tries to win. If I need to slide that bar over to Immortal to make that happen, then rad.

4) If the computer is not trying to win, then why have victory conditions at all?

The entire point of calling something a "game" is that there is the possibility of defeat. When we sit down to play a game of Catan, or Risk, or Monopoly, or any other board game, there is no certainty of victory. If Civilization wasn't a game, then it would simply continue going, as history does. The goal of the developers wouldn't be to create a specific end point, but rather to facilitate an ongoing experience that could last forever. That is the distinction between Civilization being a Simulator and a Game.

And Finally,

Well ... weren't you actually the one who drew that strong distinction, by labeling in-character (but strategically suboptimal) AI decisions as "stupid" and to be done away with?

Calling something that is stupid, stupid, does not mean that I believe that a game cannot be both strategically strong, and also provide a well rounded role playing experience. It only means that I am capable of looking at something for what it is, rather then what I wish it to be.
 
There's a few points I think is truly important for the sake of clarity.

I completely agree with you here. Also, about personalities vs. rationalism - remember how AI was described in earlier interviews? For example:
- Catherine will try to take as much land as possible, without a lot of care for individual city development.
- Elizabeth will try to reach the ocean and gain sea domination. Of that can't be reached, she just plays for domination.

That sort of AI targeting combines both rational specialization using and real world references.
 
You can certainly have an AI that "tries to win the game" without having it resort to Mafia tactics like this!

No, you can not.

An AI designed to "play to win" is by definition an AI that always does what if feels will give it the biggest chance of winning the game. If that means nuking a friend, so be it.

As soon as you deviate from this rule you are compromising and no longer have a pure "play to win"-AI.
 
An AI designed to "play to win" is by definition an AI that always does what if feels will give it the biggest chance of winning the game. If that means nuking a friend, so be it.

Playing to win and win at all costs are different things.
 
No, you can not.

An AI designed to "play to win" is by definition an AI that always does what if feels will give it the biggest chance of winning the game. If that means nuking a friend, so be it.

Not if nuking the opponent will cause them to end the game in an even worse position or allow one of your other rivals to win instead. Attacking the leader isn't always the rational course even when playing to do your best (as opposed to strictly playing to win). Sometimes it makes more sense to play to finish 2nd or 3rd.
 
This conversation started with the pros and cons of the AI having a pre-programmed personality, versus a random one. One recent week-long review addressed this, saying that a civ programmed to be aggressive in most cases, has a small possibility of going for a cultural victory, for example. This ought to satisfy everyone, because it makes the game both immersive and unpredictable (until one develops a justified, historically-based sense of a civ).

This sort of thing – that one civ will try to win via alliances, another culturally, another by war – but not the same way all the time – has me very psyched. Civ5 promises to be significantly more layered that Civ4, while being simpler to play.

With regard to the big discussion – simulation vs game, overlaid with degree of challenge – Sid Meier did indeed address it by creating difficulty levels (as Jonolith points out). But nothing he said makes me think people prefer a “sim” tilt over a “game” tilt. It’s quite likely that when many people say they want to be “challenged,” it’s not because they want an immersive simulation experience, but because they want to win the game in an entertaining manner, and can’t do it at certain levels.
 
I've said this time and time again but here it is. Their are two different ways to create an AI for civ. 1) creates an AI that role plays a historical view point, values friendships and its citizens lives, ect... ect...
2) create an AI that plays exactly like MP and wants to win at all cost. Ie, back stabs you if you're about to win a culture victory.

You can have one or the other, but not both. Personally, I prefer 1. More immersive.
Totally agree. I want to play against nations, not deathmatch bots.
I want to be challenged, but in a way that feels real. If I manage to get someone to be a close ally (eg by giving them stuff they want, defending them, attacking their enemies, etc.) then unless they're a specifically backstabbing AI, they should act like they're my ally and try to help me out.

Playing to win and win at all costs are different things.
How?
 
Top Bottom