Difficulty... too high?

[Frank327]

I think it was just a design decision to make the ancient era involve a lot more fighting. You can see it in the way warmonger penalties work and the increased power of barbarians. It's logical that in the first few days people play this game like it's civ V and get punished for it. That said, when you get an unlucky with horse barbarians it really does get difficult. Once you've got a decent military established you can switch your focus to the economy.

It's nice to see that the AI does less wonder rushing, or at least it seems so. That way you don't completely have to gimp yourself in the early game to get one.

 

[Dracul JOSHI]

Didn't realize you were an experienced AI programmer. If you're serious about writing code for a superior AI, you will be a rich man (or woman -- need to beware of unconscious bias).

Nah, It's not true AI as it doesn't really learn, it just keeps track of events and adapts to them, not quite the same thing and much simpler, particularly with such limited possibilities. True AI programming when dealing with anything in the real world is on an entirely different level, but for something in a closed system like an application, all you really need is patience and a clear understanding of the system itself. The most difficult part is quantifying the possibilities and putting the result into an efficient form.

You might laugh, but I did once write a tic-tac-toe game during an application week, the AI took me 3 days, and it's kind of a blood-force-trauma approach, but it's rock solid. It's written using a really bad propriatary software called 'neoBook' but I could do something like that easily enough in Java, but would take a little longer. C# would take a while longer, because I ain't used to it... but I worked with it once... isn't that different.

Doesn't look like much, because I ain't good at graphics, but I can send it to you if you want? The only thing I ain't sure right now is whether the mod-scan routine was in the final or if I had to cut that due to the deadline. But it does boost the ability to play as either symbol, two sets of symbols, three difficulties, a scoreboard and a 2 Player mode.

 

[Dracul JOSHI]

This thread is confusing;

First you ask people to go easy on you, but when you complain about getting outplayed by the A.I. and people point out glaring problems with your play - you essentially say that you shouldn't have to follow their suggestions in order to be able to get what you want done. Also, the guy that plays on settler takes issue with Deity lvl cheating?

you even toss in;


like your problem is everyone's problem.

Build a settler or two by turn 50, and have a standing army of at least 3-5 units; a couple of ranged and whatever else. If you don't want to do that even on settler then the problem isn't the game, man.

I did ask to refrain from insults, because I usually don't play on higher difficulties... exactly because I don't want to play a game where my enemy cheats. Civ V and BERT say they don't on normal difficulty, and please let me keep that illusion if they do.
I also don't play competitively and don't give a hoot about playing this game victory-oriented like a boardgame... it's not fun to me like that.
I am here for the roleplaying.

also, I play games for fun, I don't always want nailbiting challenge.

no, I meant that like this: I want it to be better when I come back for me, who had issues but ALSO those, who didn't... as in, I hope to like the game when the add-ons are released and I hope they will still like it.

Because patches/add-ons can ruin a game for certain people... that's what happened with me and Galactic Civilization III... that and a personal argument with the Director because of his attitude... but that has nothing to do with this here.

 

[Dracul JOSHI]

Well that was a quick and easy way to show everybody that you have no idea what you're talking about.

First off, that wasn't code.
Secondly, I can't put UML into a textfield.
Thirdly, even if I could, it wasn't an actual algorithm but just a sample to get an idea across... and I don't know about you, but If-Else switches and threated subroutines are some of the most basic building blocks of object oriented programming.

The actual code for this very basic bit would probably be around 2-5k lines. which isn't actually all that much, but definitively doesn't belong into a forum... and would also require the exact variables and documentation.

 

[Dracul JOSHI]

Dracul: Then write an algorithm. Present it to me, and I will have a Human strategy that can tear it apart in far less time than it took you to implement it.

Also, I am glad that you have a realistic view of how long this would take. You are saying it would take one FTE a year to build a credible algorithm. Not a flexible system to improve the AI over time. Not anything but the algorithm. I agree.

Do you know if 2000+ hours were budgeted for that in CiVI development?

First off... I can't write an algorithm without the code from the game because I don't know where I need to make which variables or how to get their information... this isn't XML editing, you know.
secondly... you literally do that with EVERY GAME EVER... are you also posing that enemies in other games shouldn't actually behave like anything because you're beating them anyway? Have fun hitting on motionless damage sponges.

I HAVE worked in the field, you know.

That's... actually not that long for a core mechanic in a AAA game. It could probably also be used for future releases... retweaked of course, but still.

 

[Dracul JOSHI]

I think it was just a design decision to make the ancient era involve a lot more fighting. You can see it in the way warmonger penalties work and the increased power of barbarians. It's logical that in the first few days people play this game like it's civ V and get punished for it. That said, when you get an unlucky with horse barbarians it really does get difficult. Once you've got a decent military established you can switch your focus to the economy.

It's nice to see that the AI does less wonder rushing, or at least it seems so. That way you don't completely have to gimp yourself in the early game to get one.

I can see that... although that should be mentioned, particularly because ancient and classical era span a good 4-6 hundred turns on Marathon... plenty of time to get wiped out... particularly by AI.

 

[Ryika]

First off, that wasn't code.
Secondly, I can't put UML into a textfield.
Thirdly, even if I could, it wasn't an actual algorithm but just a sample to get an idea across... and I don't know about you, but If-Else switches and threated subroutines are some of the most basic building blocks of object oriented programming.

The actual code for this very basic bit would probably be around 2-5k lines. which isn't actually all that much, but definitively doesn't belong into a forum... and would also require the exact variables and documentation.

That would be a lot of work to end up with an AI that still requires massive boosts, because you misidentified the problem in the very beginning.

The major problem is not that individual units are not behaving intelligently, the major problem is that the AI is not able to move armies in a cohesive way, is not able to understand what is "in the fog", where units have likely moved since they went into the fog, has no long-term memory about what units it encountered. The real problem the AI has is getting their units into formations and "thinking ahead" about how to move units around to make space for other units so it can attacking, instead of scattering units all over the place. And no unit-based solution will do anything to solve those problems.

Even if you were able to write code that makes single units behave intelligently within the limited scope of that one unit - which I don't think you are - the AI would still mess up the second a bunch of additional units are on the screen and suddenly the "optimal" move for the Unit does not depend only on the unit in question, but also on everything that's happening around them.

 

[Dracul JOSHI]

That would be a lot of work to end up with an AI that still requires massive boosts, because you misidentified the problem in the very beginning.

The major problem is not that individual units are not behaving intelligently, the major problem is that the AI is not able to move armies in a cohesive way, is not able to understand what is "in the fog", where units have likely moved since they went into the fog, has no long-term memory about what units it encountered. The real problem the AI has is getting their units into formations and "thinking ahead" about how to move units around to make space for other units so it can attacking, instead of scattering units all over the place. And no unit-based solution will do anything to solve those problems.

Even if you were able to write code that makes single units behave intelligently within the limited scope of that one unit - which I don't think you are - the AI would still mess up the second a bunch of additional units are on the screen and suddenly the "optimal" move for the Unit does not depend only on the unit in question, but also on everything that's happening around them.

I am well aware of that and that's why I was talking about quantification and implementation earlier.
That was a single unit example, but all of what you have listed is actually possible if one uses the leader as a super class for a couple of arrayLists containing the position (or last known position of all known unit) Heck, one could probably make one above the Leaders and just tag each unit with a Boolean Array as to whether it or its location is known by a specific leader. wouldn't use much space. After that it's mostly a question of numbers and formations and while complex formations are out of the question, some basic stuff is very possible. Heck, even stuff like Age of Empires 2 had basic formation and while the movement is implemented differently, the calculations are actually quite similar.

 

[Galbias]

On diety the AI starts with 3 settlers and 5 warriors.

Yep, that just confirms my suspicion that they're ratcheting up the AI freebies even higher than they did in Civ 5 (which was already pretty bad) to make up for the AI being clueless. Emperor is the first difficulty that gives the AI 2 Settlers so it seems like Emperor is the new Deity and Deity is now Deity +2 (meanwhile, King is a joke). Pretty disappointing.

Curious, does anyone have info on the specifics for all the difficulty levels? I'm curious as to what the other bonuses are, I did some searching in the XML files and got a couple references but no actual data.

 

[Frank327]

Yep, that just confirms my suspicion that they're ratcheting up the AI freebies even higher than they did in Civ 5 (which was already pretty bad) to make up for the AI being clueless. Emperor is the first difficulty that gives the AI 2 Settlers so it seems like Emperor is the new Deity and Deity is now Deity +2 (meanwhile, King is a joke). Pretty disappointing.

Curious, does anyone have info on the specifics for all the difficulty levels? I'm curious as to what the other bonuses are, I did some searching in the XML files and got a couple references but no actual data.

I thought at emperor the AI got one settler? Then at immortal they'd get 2. I agree that it's mostly to balance for the fact that AI is really bad at expanding. It's astonishing how long they stay at one or two cities at king level even while settlers have become easier to produce, at least that's how it feels to me.

Also it makes higher difficulty levels even more about fighting wars than previous versions. I think it's odd that king seems to be the "standard" difficulty where AI gets no boosts though. It implies that the difference between King-Deity have increased but the differences between settler-king have become a lot smaller.

 

[ExemplarVoss]

First off, that wasn't code.
Secondly, I can't put UML into a textfield.
Thirdly, even if I could, it wasn't an actual algorithm but just a sample to get an idea across... and I don't know about you, but If-Else switches and threated subroutines are some of the most basic building blocks of object oriented programming.

None of that was the issue with it. It was that the legion would kill itself, every time, unless it just happened to run out of enemies.

 

[Ulthwithian]

First off... I can't write an algorithm without the code from the game because I don't know where I need to make which variables or how to get their information... this isn't XML editing, you know.

You can write pseudocode, can you not? I'm not asking for much. If you need data structures to do what you want, simply posit that they exist.

secondly... you literally do that with EVERY GAME EVER... are you also posing that enemies in other games shouldn't actually behave like anything because you're beating them anyway? Have fun hitting on motionless damage sponges.

No. I am not posing that. I am posing that 'acting like something' in CiVI is much more difficult than 'acting like something' in Tic-Tac-Toe, Chess, or similar. I think the AI plays the game decently, particularly in the light of this thread where you original complaint, I believe, was that it was too difficult on low difficulty settings?

I HAVE worked in the field, you know.

Then you should have a realistic impression of both what is possible and what is actually useful for a game AI, no?

That's... actually not that long for a core mechanic in a AAA game. It could probably also be used for future releases... retweaked of course, but still.

It is possible that Firaxis does not consider the AI as you see it a core mechanic of CiVI. It would be useful to define what you think the purpose of the AI in CiVI is. (I.e., is it supposed to give a human player a challenge or is it supposed to play the game 'intelligently'? Doing both at once is beyond the budget of more or less every game.) And what I was saying is that 1 year FTE would be just for the construction of an AI that could play the game intelligently. If you want to ensure that that AI is also able to be modular and flexible, that would take more time. And without that modularity and flexibility, it wouldn't work for modding.

Bottom line, Firaxis clearly decided that creating an AI that could work 'well enough' (according to an internal definition of such) and could be modded, tweaked, and prodded by the community into what it wanted was superior to creating an AI that would play the game intelligently.

 

[comatosedragon]

On Emperor, 75% of the civs will have a second city up by turn 12-15.

 

[paulmclem]

Barbs seem infinite, but as long as you have enough units i.e. some Archers, spearman etc it's valuable XP.

 

[Dracul JOSHI]

You can write pseudocode, can you not? I'm not asking for much. If you need data structures to do what you want, simply posit that they exist.



No. I am not posing that. I am posing that 'acting like something' in CiVI is much more difficult than 'acting like something' in Tic-Tac-Toe, Chess, or similar. I think the AI plays the game decently, particularly in the light of this thread where you original complaint, I believe, was that it was too difficult on low difficulty settings?



Then you should have a realistic impression of both what is possible and what is actually useful for a game AI, no?



It is possible that Firaxis does not consider the AI as you see it a core mechanic of CiVI. It would be useful to define what you think the purpose of the AI in CiVI is. (I.e., is it supposed to give a human player a challenge or is it supposed to play the game 'intelligently'? Doing both at once is beyond the budget of more or less every game.) And what I was saying is that 1 year FTE would be just for the construction of an AI that could play the game intelligently. If you want to ensure that that AI is also able to be modular and flexible, that would take more time. And without that modularity and flexibility, it wouldn't work for modding.

Bottom line, Firaxis clearly decided that creating an AI that could work 'well enough' (according to an internal definition of such) and could be modded, tweaked, and prodded by the community into what it wanted was superior to creating an AI that would play the game intelligently.

you... have clearly no idea what you are asking me to do.
pseudocode still takes time to do... in case of this, probably a few days and I ain't being paid for it either.
also, you obviously know nothing about A.I., otherwise you wouldn't post such crap.
Sure, Tic Tac Toe is simple... but it took me still 3 days to make a list of all the possibilities and putting it into a code structure, particularly because I didn't have the luxury of disregarding redundant possibilities due to the fact, that such a code would have taken x-time longer than just brute-forcing it, so it would have been inefficient. Now do the math... if you can.
As for chess... that's one of the holy grails of AI coding... there's still no computer powerful enough to calculate the perfect game of chess because of the insane number of possibilities.... yet, still there are chess computers... know why? Because it's not entirely about playing the game perfectly, but playing it so that it provides the difficulty the human opponent desires.

oh, and A.I. is part of the GAMEPLAY of the game... which means it is a Core element, just like controls, render cycle and the rest from the Controller-group.
Most of the less important things are in the Model and View groups.

In case of the Tic-Tac-Toe I achieved that by having it draw pseudorandomly on easy... which can still result in a perfect game, just less likely so, so a victory isn't guaranteed... it's just that, easy to achieve.
On medium it plays perfectly for the first half of the game and then starts going random and for unbeatable it... well... played perfectly. This was pretty much a debug setting... or to train to stop making the same mistakes again and again.

 

[Acken]

So will you still be there when/if they release the C++ code ?

 

[Ozisl]

One thing that it is hard getting over is not improving every tile. Before it was OCD trying to get your farms and roads laid out. With Districts, though, there really isn't as much of a need to improve terrain, which matches nicely with the new Builders.

Gonna wait a while and see what happens with mods. I don't think it is too hard - just a different skill set.

 

[Dracul JOSHI]

So will you still be there when/if they release the C++ code ?

C++? Dear lord, that's ancient... they are really still using C++?
I never used that... I only ever coded in fully integrated object oriented languages... meaning Java and the Microsoft copy of it, C#... and websites, but I hate that (html, php, JavaScript, css)... and assembly... which was fun but complicated as all HIFL
Also my attention span is... questionable and I have my own projects, I am still going to need to renew my Sony Vegas and install it on this new pc of mine so I can continue my Let's Plays where I left off.

But I am here for now and I can keep and eye out for it, sure.

 

[Buccaneer]

C++? Dear lord, that's ancient... they are really still using C++?

You just jumped the shark. Nearly all game engines today are written C/C++.

 

[Galbias]

I thought at emperor the AI got one settler? Then at immortal they'd get 2. I agree that it's mostly to balance for the fact that AI is really bad at expanding. It's astonishing how long they stay at one or two cities at king level even while settlers have become easier to produce, at least that's how it feels to me.

Also it makes higher difficulty levels even more about fighting wars than previous versions. I think it's odd that king seems to be the "standard" difficulty where AI gets no boosts though. It implies that the difference between King-Deity have increased but the differences between settler-king have become a lot smaller.

Both games I've started on Emperor, AI has had an additional city up while I was building the second or third thing in my capital, so I'm relatively certain they do.

On top of that, they had a big enough starting army to take a city-state immediately. City-states probably need more starting units on higher difficulties so they don't just roll over and die as soon as an AI looks in their direction.