Civ IV

[Darth_Schmarth]

I demand smarter AIs!

<{POST_SNAPBACK}>

Apparently a greedy demand, since I've yet to come across one.

 

In fact every strategy game falls into the same trap, which is that of giving the AI cheat bonuses in order to make it competitive. For a history buff playing EU2, fighting a 80000-headed HRE city state army and beating it without much effort isn't all that exciting.

[Deraldin]

I demand smarter AIs!

<{POST_SNAPBACK}>

Apparently a greedy demand, since I've yet to come across one.

 

In fact every strategy game falls into the same trap, which is that of giving the AI cheat bonuses in order to make it competitive. For a history buff playing EU2, fighting a 80000-headed HRE city state army and beating it without much effort isn't all that exciting.

<{POST_SNAPBACK}>

 

Have you tried playing Galactic Civilizations? I can't say that I noticed any obvious cheating in the game although the game just lacks that vital something to keep my interest. Should be able to pick up the Deluxe Edition for $10-$20.

[SteveThaiBinh]

Have you tried playing Galactic Civilizations? I can't say that I noticed any obvious cheating in the game although the game just lacks that vital something to keep my interest. Should be able to pick up the Deluxe Edition for $10-$20.

<{POST_SNAPBACK}>

GalCiv set the standard for AI, I think. GalCiv 2 is in the works, too.

 

The only cheat the AI gets is that it knows where all the habitable planets are from the start and doesn't need to explore - but they explain that by saying they're older civilisations than the humans and have already mapped the skies successfully.

[alanschu]

I really enjoyed Galactic Civilizations.

 

I'm looking forward to the sequel.

[Guest Fishboot]

Have you tried playing Galactic Civilizations? I can't say that I noticed any obvious cheating in the game although the game just lacks that vital something to keep my interest. Should be able to pick up the Deluxe Edition for $10-$20.

<{POST_SNAPBACK}>

 

Part of the design of GCiv is that many of the 4X elements of the game were simplified to facilitate the AI. For example, the planet quality dealie is precisely intended to help the AI which would otherwise get caught up in a cattle call of planetary traits and improvements and things that it can't properly execute, since AIs have terrible troubles trying to manage the military/infrastructure/utility production balance in games like Civilization and MoO. I dunno if I want to take the tradeoff of playing a simplified AI-facilitating game rather than a more complex game that just gives the AIs raw bonuses.

[Walsingham]

Interesting to hear about GalCiv. I've not played it, but I think I will make an effort to now.

 

Fishboot: I may be being unfair since I'm not a programmer, but I do know a little about the design of expert systems. I really don't think you have to be that sophisticated to create smarter AIs. I take your point about thinking ahead, but ther are some clear stratagems that the AI uses, such as thos pointless one unit seaborne raids behind the lines, that could be bulked up to create serious problems. Feints, such as an apparent emphasis on one flank, followed by a massive push elsehwere woud also work. That kind of thing, would add spice. Atlower difficulties the Ai behaves a lot 'dumber', maybe.

 

I was also thinking about the strategy I've been following towards the AI - i.e. kill them all, all the time. They simply don't make very good friends, whereas they help each other just fine. They never ever trade favourably with a human, and hardly ever trade tech at higher levels, yet swap such goodies between themselves with gleeful abandon.

 

Bah.

 

Humbug.

 

And so forth.

[alanschu]

In case anyone is interested, an interview with Sid Meier:

 

http://pc.ign.com/articles/632/632379p1.html

[EnderAndrew]

Grr. Not much for info in the interview.

[alanschu]

No...I posted it before I read it. Oh well

[EnderAndrew]

It is nice to know the game is in the home stretch however.

[alanschu]

There's a link to a video interview at the bottom of the interview though.

 

 

And it is good to hear that they're on pace for their late 2005 release.

 

 

EDIT: D'oh, you said the same thing as me

[EnderAndrew]

I didn't watch the video interview as the corporate proxy server blocks streaming video.

[alanschu]

Meh, I downloaded it

[metadigital]

Fishboot: I may be being unfair since I'm not a programmer, but I do know a little about the design of expert systems. I really don't think you have to be that sophisticated to create smarter AIs. I take your point about thinking ahead, but ther are some clear stratagems that the AI uses, such as thos pointless one unit seaborne raids behind the lines, that could be bulked up to create serious problems. Feints, such as an apparent emphasis on one flank, followed by a massive push elsehwere woud also work. That kind of thing, would add spice. Atlower difficulties the Ai behaves a lot 'dumber', maybe.

...

<{POST_SNAPBACK}>

I concur; a simple Expert System would learn the tactics from the human players, or from its own performances, over a short period of time. It needn't be a supra-genius out-of-the-box, either, I would find it interesting to watch the AI learning how to play better ... it could even provide lessons for the player, as well, either in real time or on a "replay" (what ever happened to that feature ...)

[SteveThaiBinh]

Ooh, this sounds familiar, but I can't put my finger on it. I'm sure there was a strategy game released some years ago which recorded all your games and sent them over the internet to a central computer, which would then analyse the best strategies and constantly update your computer's AI, making it increasingly more challenging at the higher levels of play. What was this game called?

[EnderAndrew]

Vaporware?

 

[Deraldin]

Ooh, this sounds familiar, but I can't put my finger on it.  I'm sure there was a strategy game released some years ago which recorded all your games and sent them over the internet to a central computer, which would then analyse the best strategies and constantly update your computer's AI, making it increasingly more challenging at the higher levels of play.  What was this game called?

<{POST_SNAPBACK}>

 

I'm not sure what game you're referring to but I think that was one of the things they said they were doing with GalCiv. Although I could be wrong...

[metadigital]

Dunno Steve, I don't recall it. But there are numerous Expert Systems (and Decision Support Systems, or Executive Support Systems, etcetra) on the market, and some (like Internist) fit perfectly well on a 386, so I don't see the hurdle in making a game engine able to learn.

 

Even if the limiting factor was using LISP (or some other Neural Natural Language processor), the developers could create a seperate "learning engine" that operated either in parallel or batch (at the end of a game) to perform the analysis of the moves and attribute rewards accordingly.

 

Or am I too hip for the room?

[SteveThaiBinh]

Vaporware?

<{POST_SNAPBACK}>

More than likely.

[metadigital]

We could start a rumour that it's Homeworld 3.

[Loof]

Don't know how many actual games there are where teh AI uses a leanring system. From what i have heard it's a pretty common feature in th eearly planning stages but it generaly gets scraped when the developers find they don't have time to do all they want.

In some ways learnings sytems are actualy worse then the alternatives. The reaosn I say this is that the goal in game AI is to make a "good enough" oponent, as we want the player to have a reasonable chance of wining. Now if we realy implemented a learning system that did its job well it would only be a matter of time untill the player could no longer win.

 

For this reason I think that I would prefer some kind of finite state machine AI, but as im far from an expert on the subjekt you should take this with bucketfull of salt.

[metadigital]

You can always switch the AI off. Or reset it. It would interact with a database of some sort, and that can always be reset. Calibrating a difficulty setting might be more problematic, although you could have fin with other settings, like irrational-type emotional equivalents, like impulsiveness (e.g. short circuiting the calculation cycle when the pressure is on) ...

[Loof]

True but having to reset your database sort of defeats the point of the learning AI.

Impulsiveness in coputer algorithms is realy just inserting randomness into the codepath, and I would guess its more usefull for AI personality, or to create replay value then to set the difficulty.

 

One use that machine learning can (and I beleve has) had in computer games is for the developers to train the AI before release and the lock it so it stops learning. With this technice you could train your AI a bit then copy it lock it and call it easy, keep training the original copy lock and call it normal and so on. This is a technic I defninetly think could work, I'm just not a big beliver in the free learning machine in games.

[metadigital]

That's what I was alluding to in my previous comment.

 

You also wouldn't necessarily need to wipe the entire database, either; just like any transaction record can have set checkpoints that can act as rollback stops. Like in an electronic banking transaction, if the comm link to the bank clearing house goes down, you only lose the current transaction, not your entire banking history.

[Calax]

I do know that Starcraft's AI learns. They've started to rush me and my friends with lings.