My point is that this is the best AI in terms of a game so far. It isn't a comment on how well it actually plays against a particular strategy .

Again how many games did it take you before the AI needed twice as many ships as you have to win? I doubt anyone was able to wipe the galaxy with the AI in their first few games. I know I sure as heck couldn't. It took me several games to understand the underlying mechanics well enough to have a shot in heck of making the AI look silly.

There are a lot of things that a human is just simply better at doing. Long term strategy is one of them. Deciding what planets to attack, which ships to use, calculating the relative strength of your enemy, routing your forces, and dealing with a changing battlefield are all things that humans just do better. And if you really expect AI, in a game, to be able to match a human on those terms you are going to be disappointed for a long, long time. All you have to do is look at the current state of military weapons to see how far we've got to go. And those weapons systems are specifically designed to fight wars.

Another example: driving a car. If computer AI was to the point that some seem to expect we'd no longer need to drive ourselves anywhere. Ever hear of the DARPA road challenge? Basically teams have to build a robotic car that can traverse a desert course autonomously. Needless to say during the first several years of this challenge no one even came close. It has been only recently that a team has completed the entire course. And those vehicles cost tens of millions of dollars with completely custom software.

Brad has said in dozens of posts that one weakness of the AI is the fact that it cannot defend itself against an aggressive steamrolling human opponent. But in terms of computer AI this is probably the most difficult thing to accomplish and balance. The goal of any AI is to challenge *most* players. I'd say for the target audience this AI is spot on.

[quote]
Again how many games did it take you before the AI needed twice as many ships
as you have to win? I doubt anyone was able to wipe the galaxy with the
AI in their first few games. I know I sure as heck couldn't. It took me
several games to understand the underlying mechanics well enough to
have a shot in heck of making the AI look silly.
[\quote]

There's no doubt that AI programming is pretty tricky and we shouldn't expect a real intelligence any time soon. Still, we can expect that the person who designs the AI (Brad) has a much stronger understanding of the game that a totally novice player. Thus the Ai shouldn't be compared to someone a first-time who doesn't even know all the rules. As players describe their strategies, they can be incorporated into the AI. My post above was an attempt to describe a fairly simple offensive fleet management strategy that is none the less dangerous.



I agree with you on most of these items. People are going to be better at most of these spatial task than we can expect an AI to be. However, calculating the relative strength is a sore point for me. The AI seems to go by the military graph, which is extremely misleading. There's no particular reason why an AI can't predict fairly precisely (through simulation) the outcome of any particular battle. Simulating battles between it's own designs and the enemies designs should also give the AI an excellent idea of how the two militaries actually stack up. Going by the military graph is a really bad idea because it can't take into account the pairwise interactions, such as when the player has designed ships specifically to counter the AI.

Simulating battles to gauge relative strengths of fleets and designs is an excellent idea. (and of course you know the *but* is coming ). However this is an impractical solution in terms of how long the AI actually has to make that decision. Think about chess AI for a moment. The longer you let it think the stronger it plays. Sometimes it can take it upwards of an hour or more to make a single move. And that is a *known state* game. No way that would work in this game. Turns have to be over in a few seconds not hours. Multiply that by 9 opponents and all of a sudden one turn takes almost 4 weeks to complete. (9 * 8) * 9. And that is not counting minor races.

The military graph works as intended. The problem arises in how most players actually play the game. I don't design ships and build them all until the need arises. This puts the AI at a severe disadvantage because it cannot know what I'm planning to build until I actually do it. Sometimes I purposefully build inferior ships for defense and then build the best ship I can for my attack fleets normally with a completely different weapon type. So what happens is the AI is now geared towards one set of weaker ships and I smack it around with my new ships. Flaw in the AI or superior strategy by me?

The real question becomes how much time and effort should be put into the AI to combat player strategies knowing that the player will adapt to utilize even newer strategies? No matter how *clever* the AI becomes the human will always find new ways of taking advantage of it. That is until the first AI is developed that can learn. And that is going to be a very bad day for us inferior humans

okay.



it's spelled 'intelligent'.



this phrase is a bit awkward. i'd suggest "sometimes just one ship as an escort" or "somtimes the escort is just one ship."



you should use "too" here. "to" is a preposition, whereas "too" is an adverb and would correctly modify the meaning of "stupid" to "excessively stupid".



the expression is "crush them like a bug." and it's spelled "irrational."



you should say "more smart" or "smarter," but "more smarter" is redundant and improper.



i think you meant "babbling," which means to speak excessively. it's an odd choice of words, since i think your intent was say Brad's claims were exaggerated. a more direct way to have said it would have been "I see that was an exaggeration," or you could use the idiom "he was just blowing smoke."

Am dissapointed with DA[/quote]



i'm not sure if it was a type-o or a genuine spelling error, but it should be "why," and you should end the sentence with a question mark. also, even when asking a rhetorical question, you should switch the verb and subject. it should read, "Why can't Stardock make a special button to change all the AIs' aggression levels?" also, when you say "It is true," it's unclear what "it" is.

overall, your English isn't bad. and frankly, there are plenty of native speakers who abuse the language more than you.

there are a few more pervasive habits, though, that I'd like to point out. You often neglect to use a pronoun when you could, and while that's not a bad as some other mistakes, it can be confusing or awkward for a native speaker to hear and understand. it's much better sounding to say "I am so disappointed," and with a lot of other verbs it's also very important, since most english verbs don't have unique conjugations for each subject.

also, you should use articles (the, a, an) in front nouns when you're talking about a specific one. for example, saying "AI is stupid" makes it sound like the statement is about all AI in all games and other programs. saying "the AI is stupid" makes is clear and obvious that you mean the AI in the game.

also, please don't mind the fact that i don't capitalize. it's an aesthetic choice i make as a creative writer. it's very brave of you to welcome help, and i'm glad to offer what i can.

still, i don't agree with you assessment of the AI entirely. it still has flaws, but it's still a lot stronger than the AI in most other games i've seen.

Precisely!

Your on planet Earth, the planet of all work and no play... the minute you learn to talk, you get shipped off to school, and get told what to do all day by teachers. And just when you thought you can have fun after school, they invent parents and homework to stop that loophole!. Ok you finish school, escape getting told what to do all day by teachers, and soon after, your parents. Get a job and then get told what to do all day by your boss instead, arrive home and yippee - don't forget to feel too tired to have any fun after work.

Just incase your in any danger of having fun during your working life, they invent marriage and children to block that loophole! Eventually you get rid of the kids, and the job, just in time to be to old to enjoy whats left of your life and then you die!

Ah isn't life on Earth wonderful??



Only 2 to 1??

I don't think the time to simulate battles is that impractical. I find it hard to believe that simulating a typical fleet engagement, striped of all graphics, would take more than a few milliseconds. One this ancient computer I'm using right now, for instance, simulating 10,000 random dice rolls takes half a millisecond, and that's not even in an efficient language like C. If we assume an extremely luxurious 5 milliseconds per simulation, and gave each computer player half a second of CPU time, that's 100 battles it could simulate. Since battles in DA are more predictable than they were in DL, you could probably have stripped-down version that runs in well under half a millisecond but is still a very good predictor. Heck, I'm very tempted to code it up in C myself just to see how fast it could possibly be.

Now, I don't know how much CPU time each AI thread actually gets per turn in DA, but a couple of thousand reasonably accurate battle simulations per second would be a pretty good basis for evaluating your chances in a few engagements.

Chess AI's can take a long time because they are trying to evaluate so many board positions many moves ahead. Such an approach is totally infeasible for GC2, so Brad must be using some heuristics. It's hard to know for certain, but my sense looking from the outside is that using some simulation results in these heuristics would not be CPU-crushingly expensive.

I didnt play long enough to give a final opinion but after not playing GC for few monts i started a game in DA on painfull ( i used to beat the strongest AI) and it seems the AI got better, maybe its becouse i dont remember how to abouse this game to the max but, i've never seen AI build comparable ships, by the time i build a 50 dmg ships they already have few on their planets, previously i would roll over any AI on any besides the two highest difficulty levels And usually it was over by the time i had some 20dmg ships, but now everytime i declare a war on the strongest AI i win but still another one rises to power ...

But there are still some exploits, like anomalies ... i usually reasearch sensors/ engines first and make 4-5 fast survey ships to get all thye goodies, with all the cash i get from anomalies i usually don ahve to lower my 100% spending whole game and keep pumping ships/improvements to the max. The Ai is also too easy to bribe, you give them one colonization tech and theyll attack someone thats 2xtimes their military strength :/

Whelp I'll tell ya what. You go right ahead and code the C yourself to simulate the complex battle system including all the possible ship combinations along with fleet compositions and get back to us. Feel free to work from known items such as current tech level along with economic potential. I seem to remember (and I'm sure it is still up) that someone coded one and posted it on the forums as a combat simulator.

Right now the AI does a pretty good job of doing this which is all your thousands of simulations will show. It can design ships. I'm not disputing that. All I'm merely attempting to point out is there is a lot more involved in intelligently creating fleets and moving them around a board in an effective manner. That is what would take 4 weeks to accomplish for the same reason it can take a chess AI an hour to move a single piece.

I think you may be interpreting this as a bit broader than I intended. I'm merely asserting that there is no particular reason why the AI cann't know with a fair degree of certainty whether it can win any particular battle or short chain of battles. I also suspect that something along these lines could help the AI make better decisions in terms of how it's existing fleet stacks up against another empire's existing fleet.

Of course, this is not a panacea. As you say, there's a lot more the AI has to do, and a lot more that has to go into it's decision-making. But doing a few simulations is not prohibitively expensive.

I did implement a battle simulator in matlab script. Execution time for medium-sized battles (5 to 8 mediums a side) was 3 to 5 ms. These sorts of scripts (using lots of for loops and if statements) aren't particularly fast in matlab, and I've typically achieved speed ups of 30-100x by carefully reimplementing them in C. So I definitely think that C code that simulates an average battle in half a millisecond is plausible.

Its just a bunch of random numbers applied to a fairly simple ruleset, so I don't see why it wouldn't.

Pointing out spelling and grammer errors made by someone who admitted English is not his first language is kinda lame, no?

I'm all for slamming British, Aussie, Americans, and Canadians who can't figure out how to use "there, their, they're", and "too, two, to", however.

Me i just know that the pirates hate me and kick my economy's butt . I was playing on challenging and aside from the pirates which left me bankrupt the kyrnn sent spys to tap my industry needless to say they wiped the floor with me. Besides in time every game can become better with fixes more ideas and more improvements .

The AI is not really good at research/development. It ignores 700% and 300% research and production tiles. I would not call this really good, I would call this really poor.

Add to this that the AI does not know that a influence starbase will filp his planets, and is unable to understand when a reseach victory is about to occur and you have AI holes that should have been fixed long before the developers released an expansion.

Why are there influence starbases in the game if the AI cannot understand them? A patch should remove them from the game because they eliminate the AI's ability to compete. Just build some influence starbases and take all of the peaceful races' planets. The only thing that might stop you is an aggressive AI declaring war on you and destroying them for the fun of it, and not because he seems them as a threat.

What it boils down to is a game of only combat. Peaceful races cannot deal with culture and research. They are not smart enough, so they should not exist. They serve no purpose in the game.

The AI should have been tweaked in a meaningful way before releasing an expansion. At least have them use a 700% reseach tile. Because of poor AI, all we have is a game of warefare, not diplomacy, research arms races, influence fights, and the other options that would make this a deep, enjoyable game. Maybe if game reviewers actually examined what the AI is doing before praising the "brilliant" AI, Stardock would acutally work on it.

Hmm I don't know wonder what the developers have to say about this issue? Or would they comment? Best to ask them when they get into the chat room or email them. If something is wrong then hopefully they will fix it tweak it or something.

Entropy Avatar:

Ok now I think we are on the same page. I know the AI does some form of what you talk about simply because it redesigns its ships to counter whatever I'm doing with mine. So it obviously knows how its ships stack up against mine. It used to only redesign its ship every game year. This put it at a severe disadvantage. Now it seems it will redo them every few game turns if needed.

Snyder:

Ummm... have you read any of the posts about the planetary build AI? It is not meant to be *perfect*. Also just because it ignores those huge bonus tiles at times doesn't mean something is wrong. A lot of players ignore 300% farm tiles. Is that wrong? The AI simply could have determined that it did not need the bonus. Remember you do have to pay BC for that extra production. Its possible to conclude that the cost does not justify the gain in certain circumstances.

For the record I've skipped 700% manufacturing tiles when they are on an out of the way planet with no hope of being a major ship building facility and I've already got most of the TG and GA I want.

The AI does understand influence starbases. If you don't believe me have a look at the report page that shows why they are friends/enemies with you. At times you'll see -Your Alarming Influence. And just because they don't do anything about it isn't a negative either. If they are a passive AI or you greatly out match them militarily then why in the world would they go to war with you and lose a lot more then the few planets you threaten with an influence starbase.

The AI was adjusted a while back to understand when someone is going for a tech victory. Again doesn't necessarily mean they are going to do something about it. But they will if needed.

I guess my real question is have you played the game recently or are you going by what you remember from the initial release of DL?

Well, I am playing the latest version of DL.

I have DA at home for me begging to be installed.

I started complaining about this when I was deciding whether or not to purchase DA. One of the factors that was influencing my decision was whether or not they fixed civilizations responding to Culture wars. They just let themselves die rather than declaring war. the - entry in the relations tab that you are refering to is ridiculous. When you destroy a pesky scout ship, the civ goes to war with you immediately. When you spam influence starbases around one of their planets and steal an entire planet, they don't even seem to mind much. That is absurd. When testing this, I did not have even 1/2 of the military rating that they had and they still didn't attack. Placing influence star bases near a planet should be a hostile action in the Ai's eyes. If the AI does take steps against culture influence, it is not one that is even noticeable, even when culture is completely erradicating its civilization. Without an appropriate response to culture attacks, Galciv2 should not have any peaceful races, as they are just effortlessly flipped by culture starbases, placing other civilizations at an insurmountable disadvantage.

I posted on the forums, asking if DA fixed the ai's reaction to influence starbases. people told me it did not. Without reacting to culture, the game really is not a deep stradegy game of Diplomacy, influence, and culture, it is just a military operation. Just kill the guy that is not alligned with you. (Which is still quite strategic, obviously, but nowhere near as deep as it could be if the AI defended itself against culture attacks.

I bought DA anyway, hoping that it will be someday fixed.

Brad's said he's trying to fix the AI's passivity to culture in several other threads on the topic, for what it's worth. He's also said it's a careful balancing act because you don't want it to be impossible to win culturally. I can see where he's coming from, certainly, although I have to admit that right now it seems much too easy to win culturally while the AI stands idly by and smiles at you. It's gotten bad enough that i just don't build influence bases anymore, period, since it feels so cheap to me.

I wish I could find the posts you are refering to. Other people have told me the same thing, but my searches for "Frogboy" aren't showing me any of his comments about culture.

Hopefully they will improve it sometime. In my opinion, an influence starbase should never be very near a planet without its owner becoming alarmed/angry. At present, he doesn't even notice.

Not when he asks to be corrected. Read the posts you quote, man!

Right here:

https://forums.galciv2.com/?forumid=357&aid=143914#1125596

Also don't forget he sometimes posts under Draginol as well.

There are countries who would go to war with other countries just for destroying some bothersome ship, plane, building, or even single citizen, but if a country just spams them with KFC's, Micky D's, and Hollywood tripe, they just sit passively by and do nothing. Go figure!!!

Works in real life just as in the game    muhahaha, everything is going as Ronald McDonald planned.

You seemed to have missed the "and steal a whole planet" part. It's one thing to spam your culture at someone, it's another thing for the citizens of a foreign country to decide they like it so much they secede from their own nation and join yours. The nation they secede from should NOT respond well to this, and apparently in future AI updates it will be regarded appropriately by the AI.

Can someone MOD a McDonalds race?    Think about it; +50 Influence, and the Super Isolationist Super ability (Colonize barren worlds (tons of McDonalds in the slums), and the "no foreign ship can travel more than 3 spaces per turn in their area of influence" - this is due to +400 cholestrol level in the arteries of all who eat the local food