Internal build 0.78 is the first build that the full strategic AI has been checked in. From then on, it’ll be balancing it based on play testing and listening to players discuss how they beat it. But feature wise, the AI can now fully play the game with one exception – we don’t let the AI win via the master quest. But it will try to win through other means if you have it selected.
The next public build will be where we begin the tactical AI implementation. We’ve been writing the APIs for tactical battles for months but we haven’t made use of them yet since we first have to decide how tactical battles are typically fought. That is, knowing the rules is only part of the challenge.
AI Wars: Let it begin. LET IT BEGIN!
To see how effective the AI is at decimating each other, we put a ton of them on a small map. Specifically, small map, 18 players.
I, on the other hand, will be going out of my way to kill them all quickly.
As you read this, please understand, I’m not trying to play the game. I am trying to win the game as efficiently and, truth be told, as cheaply as possible. Any scummy tactic I can find I’m going to use.
It’s a three way battle: FE’s AI vs. My cheese vs. Kael’s design.
So let’s the cheesing begin…
Since I am scum..
So the other players are out there happily exploring the world. Getting gear, probably training pioneers with dreams of expanding their empires.
What they’re really doing is starting my cities for me because while they’re out, I’m going to take their cities.
So…
What should we do about that? What should they do to stop me?
Here’s the data we have to work with:
What sort of evil should we inject into the function above?
Getting ganged up on
To answer some of the things above, we can create some interesting consequences that will help “mix things up”.
For example, I can take advantage of the ideological differences between Kingdoms and Empires. So if I’m a Kingdom and I attack another Kingdom, that’s a big deal. Or I can make it a big deal if I attack early depending on the faction traits. For example, a player with a Diplomatic trait might freak out a lot more if I go and attack someone early on.
Go cheese!
If that is balanced anything like .77 you can just get a hero up to blizzard level and rush a large AI city. Even though the population will be 0 when you capture it, the faction wide bonuses make you very powerful at little cost. Good luck.
I have to go make a post about removing the mountains from the game. The we can rename it Plains Duelers.
Ha sounds like fun, good luck with that!
Any specifics on why the AI doesn't pursue the Master Quest? Will this be implemented later on or are there definite reasons?
Maybe the other players in a game should know when a player is advanced in the Master Quest?
Oh, and did you set anything up for the AI to know that you are raping them and start ganging up on you? I loved the random event in GalCiv2 where the Drengin or Yor would get so powerful everyone started to attack them.
Of course, it's not balanced like 0.77 so that won't work.
You tease.
Don't blame me, blame Derek. He's diabolical.
Plus, the AI is much better now about getting experience. It already knows to split up its forces some, especially early, to maximize XP.
Yay cheese!
edit: I mean Go cheese!
Does it know how many movement turns you are away from attacking a city?
Does it know the strength of your potential attack stacks?
Does it know the bonus it gets for defending?
Does it compensate for magic and other bonuses when calculating the stack attack potential (and its own defence potential)?
Does it know the importance of its individual cities?
Does it know you are aggressive before you declare war?
If it doesn't do anything until you declare war then it has already lost. It needs to know your threat level before you declare war because it needs to know how far away from its cities it can explore with its own stack. If it is too protective then it won't advance and level up because it will be hovering around home base. If it is too aggressive then it will lose its cities.
I eagerly await the next beta with all these mysterious changes...
Maybe it should know the structure of your army\civilization? If it sees you're relying on powerful heroes, it can split it's forces so you can't meet all their attacks. The AI usually can't tell how strong I am based on my power rating, so having more details about the player might give it an edge. Sometimes I just cut through their lines like butter yet they have double my score. A potential idea:Upon visual confirmation of an improvement, it can get an idea of how far you are along the 3 tech trees. Each tech tree progression would determine how the AI attacks you. Lots of civics: It knows you will dominate later, so it must overwhelm you now.Lots of Magic: It knows that you will be a focused threat to him, so it should spread out it's army and attack multiple targets to since you can't be everywhere at once.Lots of War: This would mean you've slowed down the civics tech and have the potential for a more spread out offensive\defensive. So it should focus it's attack.Also, maybe it could adjust it's power rating analysis of you based on information on how well you do in combat. If you're "medium" strength and you're taking out "strong" targets constantly, maybe it should treat your unit\hero as such? The better you do, the more your power rating goes up.Also, for allies, maybe it can adjust it's own standing to your power rating by having allies willing to help them crush you. This way it wouldn't screw around with you if you're too powerful, but if it has an ally helping, or another AI fighting you, it would jump in.Obviously all this would require programmer "optimization" and "approximation" to make it computationally possible. Those are my two cents!
EDIT: More ideasAnother idea that I had was giving the AI a goal or a strategy. Maybe if it has a magic based hero and the starting conditions are good for magic, it should focus on building an ultimate mage to decimate the enemies with magic, etc. An idea would be to have a set of pre-determined strategy biases that the AI can switch between depending on the environment\conditions.One other issue about magic is the chaos spell. I think the AI\auto battle spams it too much and it's somewhat useless to do so. Maybe it should randomly do good things to one of your units or randomly do bad things to one of the enemy. That way spamming it is justified. I haven't seen the AI rape me with terrible spells yet either. Will we expect evil and cruel magic casting eventually?
The AI should focus on getting a basic production building (workshop) and training units.
Then they should know how to counterattack once you forces have been weakened.
They also might need to know how to attack in concert, with the strongest of their armies directly attacking your city while the others lay siege to your outward resources.
Something like a check to see if there are multiple, substantial threats nearby before vacating your city.
Not sure if that makes sense, but what I was able to do in .76 is:
Send a 3-man squad of horsemen near an enemy city and raze their elemental nodes, mines, etc. while keeping my real army right outside their border. The AI would send their garrisoned army out to deal with my pillaging group and I'd basically just walk into their cities (the militia would be there, but meh).
I wouldn't like it if the AI did the things I do to it to me. AOE spells are brutal. It is very difficult for troops to survive even weak ones. I don't defend any of my cities really, I rely on Firestorm and Cloud Walk for protection when necessary. If I got hit with a strong Firestorm (or multiple Firestorms!) the moment I stepped in to my opponents territory that killed a stack, or Cloud Walked a hero stack of doom next to me, I may feel it is unfair. Yet that is what I try to do to the AI. Maybe that means somethings is wrong, I am not sure.
I think troops need to be buffed (or made available more quickly), or the AI needs to be more savvy using its heroes, or the AI needs to use the spells the player knows works. Maybe all of the above. It's current troop heavy focus gets decimated by fireballs, blizzards, dirges, cloud walks, and firestorms.
If I am the AI, which I wrote a whole report on how I wanted to be AI when I grew up, I go into panic mode.
1.Rush Buy defender tank. Cast spell to buff city defense. Summon a magical unit.
2. Gauge speed of attack. If it can defend city in time, send unit. If it can't, build army near city and counterattack.
3. If that battle is lost, decide whether a full force attack with Sov can take back city. Maybe even decide if city is crucial to winning the game.
Responding to Burress and his idea about AOE spells being too brutal for the players. I think I'd agree with him, maybe we need spells to help protect against the map based spells the AI could potentially use?
What i always wished to see more in strategy games AI was some Randomness. Starcraft 2 seemed to have some of it when i played against the AI.
What I mean is like, one game the AI decides to turtle up and defend from an expected attack.
Next time It makes an all out and go straight for your city while neglecting his own defense. Potentially dooming itself, but also potentially ruining your plan or you game.
AI should automatically declare war on players of opposing faction. It's what I would do.
AI should know which monster spawners to leave along and farm when they get strong and which to kill right away. For example, farm butchermen in the hopes of a maul weapon dropping.
AI should purposefully trigger strong monsters near enemy territory and run away or suicide (preferably with expendable units).
Really evil AIs will powerlevel strong monters near enemy territory by feeding them lots of cheap units.
AI should be able to rush build defensive buildings or units if it thinks a city is about to be attacked.
AI should be able to design or build better spear wielding units such as poison coated spear wielding units with bonuses to movement, strength and dexterity.
AI should be able to designate a city for producing elite units. Stack all unit boosting buildings and city enchantments on that city.
AI should build some cheap units with high movement for executing scorched earth tactics in enemy territory. IE scouts to destroy all your undefended shards and outpost.
AI should pull heroes out of danger if it knows that its stack or city can not deal with a near by hostile army. Units are expendable so it is ok to let those do a rear guard action.
AI should designate heroes as high priority targets and kill them first when possible. Especially if it knows it will lose the fight anyways.
I would like an AI sov in a mana rich region to spam buff spells all over their units and cities and summon both strategic and tactical allies
A more material strong sov to create large armies of well armed and armored troops.
I would love both types of sovs to harry my flanks with raiders to keep me honest.
I would love an AI sov to hit and run with the Escape spell and annoy my stack of doom...
I would love for unit gifting ala GC2 from one AI sov to another: When the Elven Archers arrived at Helm's Deep in The Two Towers (I know it didn't happen in the book) but still... kewl!
And what Zygwen says just above...
What is the point of testing like this? Serious question. I don't think anyone sits down to play a 4x game with the intention of winning as quickly as possibly, whether or not the AI can survive a turn 10 rush is kind of irrelevant I would think.
I don't see how it could fare well anyway, regardless of intelligence. The way the game is set up your main force is your sov, and your sov is also the unit you explore with, so if you stumble across another civ in the first few turns of the game you are going to have your best invasion force right there ready to sack the city. No level 1 militia is going to stand in their way. So either the AI gimps itself by camping it's sov in it's starter city and it's immediate viciinty for the first few hundred turns, or it doesn't and leaves it's city ripe for any other wandering sov to grab. I see no third option with current mechanics.
I use my Soverign until I get my first Champ then I send him to the capital for the rest of the game. Let my champs do the work
YES. An interesting AI needs to be able to balance contingencies and predict threats beyond progressing linearly toward what they're programmed to focus on.
The player also needs to be able to see these plans in a way that they can understand; the AI's choices translated into plots and concerns.
There are many great features available to you once you register, including:
Sign in or Create Account
