The AI is still horrible.

[trooth]

I play marathon and the ai most certainly uses religion, missionaries, and great prophets. They definitely align with city states. Militarily it is still lacking. Some games you will see more action than others and it depends on the civs you play against.

Sent from my Galaxy Nexus using Tapatalk 2

 

[tf21]

I'm curious to hear of an example of a computer AI that is considered brilliant in a game/simulation that has land, sea and air components and can utilize all elements in a coordinated manners, is tactically competent, operationally and strategically brilliant and pull off an occasional surprise. Oh yes is also open ended, not scripted and able to react in a rational and competent manner to moves I make, and doesn't cheat.

I've played hundreds of games/simulations/hard core military sims over the years. Some are better than others, but to mind only chess is something that can be programed to be very difficult and often unbeatable, but that's a 2 dimensional and simple game.

 

[ptoss1]

I'm curious to hear of an example of a computer AI that is considered brilliant in a game/simulation that has land, sea and air components and can utilize all elements in a coordinated manners, is tactically competent, operationally and strategically brilliant and pull off an occasional surprise. Oh yes is also open ended, not scripted and able to react in a rational and competent manner to moves I make, and doesn't cheat.

I've played hundreds of games/simulations/hard core military sims over the years. Some are better than others, but to mind only chess is something that can be programed to be very difficult and often unbeatable, but that's a 2 dimensional and simple game.

In my opinion, the goal of a computer AI in a strategy simulation game such as this isn't to become a "perfect" player. Instead, we seek a human-like companion that will make the same mistakes such as us and to formulate plans based on concepts and ideals we understand. In theory, Firaxis could make an awesome and unbeatable AI opponent with the calculated ability to know which path is optimal and how best to handle a situation, but what would the fun be in that? What is more fun? Playing against the best chess AI, or playing against a human player who has consistently bested you?

Not saying CiV AI is good. At launch I considered mediocre. Now I say it's average. One of the underlying problems is that the AI constantly send units to die against the walls of a player's city in ill-plotted attempts to win "the game". A human player would calculate the losses and back out, but an AI will lose even more resources before it becomes increasingly clear that it has "lost" the battle (which has become less common in G&K from my experience, but occasionally still happens). Then they will offer peace treaties which do not reflect their situation - as if I would not take their cities and raze them.

It is the lack of rationality in CiV AI that undermines the gameplay, not the other way around.

 

[Andulias]

I'm curious to hear of an example of a computer AI that is considered brilliant in a game/simulation that has land, sea and air components and can utilize all elements in a coordinated manners, is tactically competent, operationally and strategically brilliant and pull off an occasional surprise. Oh yes is also open ended, not scripted and able to react in a rational and competent manner to moves I make, and doesn't cheat.

I've played hundreds of games/simulations/hard core military sims over the years. Some are better than others, but to mind only chess is something that can be programed to be very difficult and often unbeatable, but that's a 2 dimensional and simple game.

People weren't complaining that the AI isn't brilliant, nobody expected it to be. People were complaining because, especially at launch, the AI had absolutely no idea how to play the game whatsoever. Thankfully it has been much improved, but it still sometimes does things that are borderline stupid.

 

[Kabloosh]

I'm staring at Korea's massive fleet which is like 15 turtle ships and a few frigates and he just declared war on China which is literally right across this small channel for him to attack and all I saw him do was send in 1 frigate which got obliterated by a city cause it wanted to hug it for some reason.

Granted China has battleships and ironclads... correction "did" cause I was also at war with China(along with like 5 other civs) so I obliterated his 5 or so battleships and ironclads with essentially a battle carrier group consisting of carriers, subs, battleships and destroyers.

I'm just disappointed by how the AI sits on their units.

 

[smallfish]

I'm staring at Korea's massive fleet which is like 15 turtle ships and a few frigates and he just declared war on China which is literally right across this small channel for him to attack and all I saw him do was send in 1 frigate which got obliterated by a city cause it wanted to hug it for some reason.

Granted China has battleships and ironclads... correction "did" cause I was also at war with China(along with like 5 other civs) so I obliterated his 5 or so battleships and ironclads with essentially a battle carrier group consisting of carriers, subs, battleships and destroyers.

I'm just disappointed by how the AI sits on their units.

Me thinks its the difficulty level that's inhibiting their willingness to expend units. I was playing an Immortal game where upon taking Sejong's coastal capital, he literally kept kamikazing Turtles in to take and retake and take and retake ad 50 turns...

 

[ptoss1]

I'm staring at Korea's massive fleet which is like 15 turtle ships and a few frigates and he just declared war on China which is literally right across this small channel for him to attack and all I saw him do was send in 1 frigate which got obliterated by a city cause it wanted to hug it for some reason.

Granted China has battleships and ironclads... correction "did" cause I was also at war with China(along with like 5 other civs) so I obliterated his 5 or so battleships and ironclads with essentially a battle carrier group consisting of carriers, subs, battleships and destroyers.

I'm just disappointed by how the AI sits on their units.

Korea in my game:

Is Korea supposed to city spam?

Spoiler :

 

[smallfish]

Korea in my game:

Is Korea supposed to city spam?

Spoiler :

Yes, yes he will.

He will wait until he has a higher tech level than you (read: Hwachas) before he dows. Don't ever let him get there.

 

[machival]

I've said before(not to an individual mind you), if you're so good that deity isn't a challenge for you, then it should be a "no brainer" to mix up game play to make it a challenge for yourself.

Just worldbuild yourself a scenario and give whatever extra advantage to the AI that you think will be "fair' to you.

For example, give them extra starting settlers, units and gold.

I dare you to start a huge pangaea, with you in the middle and give all 22 AI's a starting advantage of 4-8 settlers, 4-8 units and say 5K in gold. If you win it at all it would be a miracle. At the very least you would be sweating bullets every turn.

You miss the point completly and absolutely. If you must give the AI 1000% extra everything in order for it to be competitive, then you know you're dealing with an extremely poor AI.

This isn't about playing Deity. It isn't about giving 100x gold, settlers and what have you. Even if I give the AI 1000% extra everything, it will still be the same ******** AI.

Yes, your scenario would probably result in a loss, so? What's that, proof of how smart the AI can be? No, in fact it's the opposite.

Again, setting handicaps for humans DOES NOT mean that the AI is better. It just shows how poor it is.

 

[AlextheGr8]

I'm curious to hear of an example of a computer AI that is considered brilliant in a game/simulation that has land, sea and air components and can utilize all elements in a coordinated manners, is tactically competent, operationally and strategically brilliant and pull off an occasional surprise. Oh yes is also open ended, not scripted and able to react in a rational and competent manner to moves I make, and doesn't cheat.

I've played hundreds of games/simulations/hard core military sims over the years. Some are better than others, but to mind only chess is something that can be programed to be very difficult and often unbeatable, but that's a 2 dimensional and simple game.

Panzer General
Hearts of Iron

 

[Andulias]

You miss the point completly and absolutely. If you must give the AI 1000% extra everything in order for it to be competitive, then you know you're dealing with an extremely poor AI.

This isn't about playing Deity. It isn't about giving 100x gold, settlers and what have you. Even if I give the AI 1000% extra everything, it will still be the same ******** AI.

Yes, your scenario would probably result in a loss, so? What's that, proof of how smart the AI can be? No, in fact it's the opposite.

Again, setting handicaps for humans DOES NOT mean that the AI is better. It just shows how poor it is.

But it will offer you a challenge. So, fire away.

 

[Tokira]

The AI isn't very great especially in wars. They hardly ever protect their units and they declare wars in the weirdest times, such as just after paying 1000 gold for a bunch of luxury resources. As for their expansion, some of them spread cities like mad. Some othe civs tend to have very small empires, though.

 

[hbar]

...In theory, Firaxis could make an awesome and unbeatable AI opponent with the calculated ability to know which path is optimal and how best to handle a situation, but what would the fun be in that? What is more fun? Playing against the best chess AI, or playing against a human player who has consistently bested you?

This is patently untrue, for several reasons. First, the best chess AI can't run on a gaming rig, period. The bigger issue is that chess is a relatively closed system. Each turn, the AI can move at max 16 pieces (and usually less). Each piece has a possible 8ish moves on average, making a total move space of 16^8, or 4 billion possibilities. That sounds like a lot, but for a computer it is manageable. Chess AIs aren't smart at all - they barely qualify as AI. They brute-force compute the possible outcomes of all possible moves for ~15 turns, then pick the best. There is no strategy involved, just as many processor cycles as can fit in a standard chess turn.

Now let's take the example of a typical invasion in CiV. A decent army will have about 16 units - so far, so good. Each unit has 19 possible moves (we'll limit ourselves to 2-move units), making a move space for the army of 16^19, or 75 billion trillion moves (7.5E22 for all my homies in the sciences). Also, each move has a less certain outcome - in chess, the result is binary. The space is empty, or the move results in a kill. In CiV, units have hit points, and the combat is random. This means that subsequent moves in the same turn rely on what happens previously. So take that 75 billion trillion and multiply it by 16, because you have to repeat it for all the units. And, the units don't come for free at the start of the game. There are cities which produce and spend resources which must be balanced between military, growth, and gold. All of these things are open ended and complex, with multiple interacting factors. In short, CiV AI is bad exactly because CiV is the kind of simulation all AI, real or theoretical, are bad at.

Of course the AI could be better, and honestly it should be better - it's been the weakest point of the series since the beginning, and it arguably hasn't seen drastic improvement since Civ 2 (although it has kept up with the myriad new features that have been added since then, for the most part). But for you to say it is easy to implement trivializes decades of AI research.

 

[machival]

This is patently untrue, for several reasons. First, the best chess AI can't run on a gaming rig, period. The bigger issue is that chess is a relatively closed system. Each turn, the AI can move at max 16 pieces (and usually less). Each piece has a possible 8ish moves on average, making a total move space of 16^8, or 4 billion possibilities. That sounds like a lot, but for a computer it is manageable. Chess AIs aren't smart at all - they barely qualify as AI. They brute-force compute the possible outcomes of all possible moves for ~15 turns, then pick the best. There is no strategy involved, just as many processor cycles as can fit in a standard chess turn.

No, not period. There is already one that does and does very fine. Fritz is the engine, Homm5 is the game. eternal essence is the project. And it's seen as a big success by the heroes community.

Brute force was used for Deep Blue that beat Kasparov in 1997.
The modern chess engines rely on advanced algorithms in order to shorten the calculation process. In any case, brute force isn't what makes the best engines like Houdini, Rybka, the best. There is strategy involved.

But for you to say it is easy to implement trivializes decades of AI research.

Nice straw man, but the things I suggested really are easy to implement. The combat part is harder to perfect, but it sure is piece of cake to block AI from wasting money on useless open borders. I bet it can be done in minutes, surely not decades.
Tactics must be hard to improve, but there are a lot of things to be done strategically to improve the AI.

 

[Unthinking Pain]

How about a "Don't go in the water if the enemy has boats around" flag? That'd be a good start.

 

[Atwork]

wasting money on useless open borders

I'm not convinced it is a waste. I believe that open border agreements factor in to the AI's feelings toward a player. Therefore, the AI is essentially offering money for an agreement, that if accepted, will in turn result in the AI having a more favorable view of the relationship. But, if rejected, the positive modifier isn't triggered. I'm not sure that's worthless.

 

[Andulias]

No, not period. There is already one that does and does very fine. Fritz is the engine, Homm5 is the game. eternal essence is the project. And it's seen as a big success by the heroes community.

And is also completely inapplicable to CiV. Movement around the map is hardly as complex, while combat is constricted to a relatively small battlefield with limited movement and attack options and no terrain modifiers. The variables simply aren't as many.

What's more, and that's something you for some reason chose to ignore, it took the fans SIX YEARS to come up with that AI. Six years. Six years of having a game with an established rule-set, six years of people writing a code for a game they already know inside and out. Meanwhile, CiV's AI was for the most part written during development after incorporating hexes and 1UPT, both of which completely change the flow of gameplay. Essentially they had to scrap almost everything they had accomplished with previous Civ games.

So a comparison in this case is hardly fair.

 

[Securion]

Why dont they build settlers anymore?!? They have very good land but dont build any new cities.
Sometimes they build a settler and move it out in the water, and then it just move one tile and then back again, until it gets killed. Or it just stays there for the rest of the game.
Hilarious.

 

[Falk]

My impression is that the AI got even worse. I was a King player in vanilla - I usually won, but it was always close. With G&K King is more like Noble was in vanilla. Not even close to being a challenge.

I think it's mainly because the AI tends not to build new cities for way too long. In vanilla, the AI often spammed cities early on - with its bonuses this worked nicely and snowballed quite well, so it could keep up with the human in later stages of the game. No more.

As for combat AI - the army composition is improved, but since the AI has no clue what to do with its units, it makes no difference at all. Or in fact it does: the AI now builds more ranged/siege units and these are easier to kill than melee units, so the combat AI actually performs worse than before.

 

[Adjuvant]

I'm curious to hear of an example of a computer AI that is considered brilliant in a game/simulation that has land, sea and air components and can utilize all elements in a coordinated manners, is tactically competent, operationally and strategically brilliant and pull off an occasional surprise. Oh yes is also open ended, not scripted and able to react in a rational and competent manner to moves I make, and doesn't cheat.

I've played hundreds of games/simulations/hard core military sims over the years. Some are better than others, but to mind only chess is something that can be programed to be very difficult and often unbeatable, but that's a 2 dimensional and simple game.

Starcraft 2 on "very hard" or "insane" is fun to me. Alot is scripted, early on, like they won't do "stealth unit surprise attacks" like humans will, but for the head-to-head, bare-knuckle brawl of an AI having competent tactics with a slight edge over the player in precision and resource gain, it's worlds apart from Civ 5... and it's RTS, not turn based. Just seems to me programming turn-based AI should be easier.