AI in need of a tune up, or a boot to the head?

Hi!

Sophisticated diplomacy compared to other games in the genre.

Hovewer I understand your feelings. I'm a very strong Stars! player. When I bought the game, I had similar expectations: finaly no need for a live opponet. But when I wiped the floor with the tough AI in my second game, I realized I can't expect a real diplomacy from a game for 40 USD. If you'd realy like a challenge, join us on Stars! Autohost.

BR, Iztok

All of these (limited wars, block alliances, binding treaties ala the UP, more sophisticated analysis of when to go to war and taking actual outcomes into effect when assessing the progress of wars) are really very nice ideas that I would naively assume to be reasonable to implement. However, nobody's done them, and that leads me to expect that it's actually a good deal more difficult than it would appear to the casual observer.

The fact is that this game has the latest iteration of (basically) the same diplomacy options we've had since Civilization-- Ally, go to war, trade goods. It's clearly more sophisticated than any of its predecessors, and a commendable addition to the field-- even if it's not a revolutionary transformation.

It's really NOT any better than any previous CIV games out there, namely Civ4.

The AI is just as stupid as any other game of this genre that I've played. They're simple - kill the largest threat, or be afraid of the most powerful player.

Bully the weak, obey the strong. That's all the do. When a war IS waged, they don't seem to consider the effects of it at all.

The GalCiv devs basically needed an excuse as to why they didn't implement multiplayer, and they chose to say it's because their AI is groundbreaking - News flash, it's not.

People have been asking for more sophisticated Diplomacy since DL first came out. It's been getting better, but other things like graphics and other features always get the no.1 spots on the Dev-to-do-list.

This is one of the things that is difficult to program in. Could you please explain here (in mathematical terms perferably) how to have the AI identify that they are losing the war, and that their "superior" military isn't up to the job? If you could, the devs could work on it right away.

For us, its easy to identify problems, but the AI has no such luck. They need to be programed first.

Um come on it's not rocket science. All it would need to do is keep track of say the last 10-20 battles with each race in the last x weeks. If it sees that it is losing 80% of the time, it should :

1) Rethink ship design. For each of those battles the key element of why it lost could be recorded. (out-numbered, too weak offensive, thier defense too good, your defense too weak, ...)

2) Rethink tech choices based on above. Face it, the AI doesn't use defense correctly, or prioritize researching it.

3) Consider upgrading ships to better design.

4) Consider peace if it doesn't make any progress.

The reality is the AI does not design good ships. It still for the most part designs hords of crappy ships, and on top of that it uses a completely flawed calculation of military power.

It is completely pathetic how the turn your power is more than them, they will almost always take a peace deal.

This is rich. All it needs to do is 'think' and 'reconsider' its choices based on 10-20 battles. What is the precise formula hear about how it will decide what to do? What if there were multiple reasons it lost? How processor intensive would it be to make a good decision? How does it weight its builds based on feedback from multiple races?

But your attitude ('face it!') indicates that you think the devs are either stupid or they haven't done this ("pathetic") because they hate you. Nevermind that I can't think of another game that does this well. I doubt you can either.

I would assume that if you wanted to do this, you would define a war-status variable, and modify it depending on the outcome of battles during the war (say, by reducing it in value by the cost of everything the enemy destroys and increasing it by the value of everything you capture or destroy from your enemy--- or possibly setting it to the ratio of the former to the latter). You probably wouldn't want to make it the sole defining factor, but at least the computer would know when it's getting its butt kicked.

Programming considerations aside, perhaps playing on "Suicidal" means the AI is suicidal as well? Did Hitler surrender when he was losing? Did the Roman Empire? Suicidal, indeed.

I happen to be a programmer myself, and I understand the argument that what is obvious for us to percieve can be quite vexing to program into a machine. But gallagher's argument seems reasonable to me. I think it would be possible for the AI to look at a win/loss ratio and based on that alone, modify it's statistacal assessment of relative military strength accordingly. I will give a simple example here, not with the intent that it is to be condsired as a real solution, but just as simple proof that it can be done. If the AI rates my military at 80 and its own at 150, there is a 70 point difference. Let's say I've won the last 9 out of 10 battles, so it has a 0.1 win rate. The AI could determine that war is profitable by using the following calculation:

(AIRating - MyRating) * WinRate

In my example this would give (150 - 80) * 0.1 = 7

Now we set up a scale for the AI to use. Say anything less than a 15 is a losing position. Of course, my numbers are not balanced to good gameplay and the devs would have to find out through testing what the real threasholds should be.

While wholly agreeing with you in principle, I think you're going to want to take the scale of victories into account as well. Losing a couple freighters or scout ships needs to count less than losing a major planet.

Agreed. I made my example extreemly simple just to show the general idea of how it could be done. You can then add numerical weights, say, relative to the cost of the ships involved to further modify a formula.

I'd also like to add that I'm sure that GalCiv does something far, far, far more complicated than I am capable of conceiving. I just wanted to give an example of the mathematical approach to solving a problem like this.

The AI should calculate military power not only based on ship's power, but strategy efficiency. If, during the war, it has lost ships in a 10:1 ratio against you, then it should evaluate your military strenght 10 times than it actually is.

That way, the computer will actually compute your strategic genius when considering if to go in a war or not. Makes me think of Rome's reputation in the Mediterranian.


Also, I already suggested that, for the computer to stage a war, it should do it in that order:

1) Decide that it will go to war. It will begin a military build-up, and the computer will have to have an agenda (capture planets X, Y and Z in priority, destroy starbases A B and D). It should have a PREPARED STRATEGY, and have some sort of priority evaluation that will allow the computer to stray out of this prepared strategy only is better targets of opportunity present themselves (and still, the computer should evaluate the value of said target of opportunity, compared to sticking to it's actual strategy)

2) The computer should deploy and concentrate it's fleet in order to achieve the previously agenda

3) The computer NOW declare war, and will strike within 2 weeks it's main targets. The computer still should have a valid strategy of offense, and probably settle for what it just won if he could not swallow anything larger. At the same time, the computer is supposed to have many fleets either tracking your fleets, or protecting it's assets.


I don't know how such programming be possible, 'cause I know nuts about programming. However, I do know that any dev who actually achieve such STRATEGIC artificial intelligence could actually be heralded as the one who will revolutionnise the strategy industry.

Jasamcarl. Why do you always get personal in your mindless fanboy posts?

My problem with the devs is they spend thier time on new bells and whistles, and leave problems like this laying around. I won't buy TA. Mainly because it looks like it will essentially have the same problems as DL and DA.

DA's AI is better than DL, but not in the war department.

I would think just a small change here could make the AI much better, and I just can't understand why resources don't go this way.

Once again let me say, the game is great. However, to me some criticism is necessary if this game is ever going to meet it's potential.

Well if that's what's confusing you, let me make it simple...

For the majority of the players, the current difficulty is fine. There is always a tradeoff between the complexity of game mechanics and the ability of the ai to exploit those mechanics, and the specific combination of those two elements is not obvious from a marketing point of view, especially when your vantage point is that of a hardcore gamer.

You should consider this before using strong language such as 'pathetic', or claim ignorance of the developers motives, or, as you did in an earlier thread, claim the game was in constant 'beta', or that they are leaving a complex ai challenge 'laying around'. Until you can cite a game that does this better, much of the tone of your posts will have an air of entitlment.

The AI's ok with the diplomacy, but could be better of course. The real problem is that the diplomacy system is far too limited and it's more like trading.
I want more Civ-series styled diplomacy (yeah GC is different from SMcivs but anyway)

Oh and the AI is not active enough when it comes to dealing with the player.

Civ Diplomacy? I always read Civ as A war not war system and I like that game.

I agree in general: We always get told how great the AI is and how much better it is than in any other game out there but in reality it's not that bright at all (and not better than the Civ4 one, for example. In fact it's worse, if only due to poorly balanced and exploitable features that the AI can't handle too well). It's still pretty good but it could be so much better. And should be, if we are to believe Stardock marketing.

Oh, and yes, I am quite disappointed with how the AI enhancement was supposed to be a major development point for TA and now it's all about new features (bells and whistles) while the AI isn't even mentioned anymore. Read the latest dev journal/TA preview? One lonely and vague bulletpoint about AI but pages on useless features like new planet graphics. Not that I don't want better graphics (and the memory reduction is very nice... well, would be if it actually worked now) but given the choice I'd rather get a decent AI instead.

And if we don't get that... then maybe at least multiplayer.

I seem to remember something about Galciv II 2.0 coming out next year, after TA, that will include diplomacy and espionage improvements. Also, I thought up to this phase they still havent started tuning the AI for TA, they've just been busy implementing the new tech trees.

From TA Beta 2A Announcement:
Lastly, the AI hasn't been touched yet. So as you play and see the AI using the new techs and the new improvements, that's one of the reasons we don't use scripted AI. When players start modding the game they don't have to worry that the AI won't be able to use their mods. That said, a bunch of AI work will be done in December based on new strategies and ideas players have given us.

I kinda agree with both sides.

The AI is good, but it behaves quite stupidly. The AI shouldn't learn when to make peace but how to fight a war better. Right now, it fights with brute force, not strategic thinking. i.e. Send any ship that isn't meant to be guarding a planet fleeted into the player's core areas and engage. Masochistic is very easy and even though the AI gets more resources to work with, it uses those resources very badly. The single and only trick to beating higher difficulties is to manage your own resources better then the AI can waste theirs. The strategic element of the game is getting stale and only once in a while I get a game that I really enjoy.

Perhaps each race getting their own tech tree will be a precursor to better AI.It would be nice if you could create a shell strategy for each race although this may come close to prescripting.
One thing that bothers me is that the UP vote never correlates with your diplomacy relationships.You would think that an ally would either vote something positive for themselves or you.
Also when a civ surrenders couldn't this be influenced by a UP vote as to who gets their worlds assuming you have no ships or transports attacking the left over planets?
And why not expanding conditions for war?Like remove your starbase from mining resource no.xxx or it will be attacked.Why always go for the whole enchilada of an allout war?
I mean it could be up to you to go to war over an incident.
And besides the points which effect your relations I would like to have sliders to change my stance on AI's.

It would be nice to have
a) the announcement of the agenda
4-5 weeks of negociations
c) vote on the topic

the b sequence is supposed to either:

1- Convince the other players to vote your way
2- Milk the #1 peoples for as money as they want to give you

Off course, like in SMAC, if you are #1, you could simply say: "Vote for me, or else..." (a) I will stop trading with you I'll smash your teeth so hard you won't know what happened

The Civ 4 AI is not only benefited by easier mechanics, i.e. how cities are conquered, but cheats like hell at the very early levels. What you call 'poorly balanced mechanics' are inherent result of increased complexity. And no, there are not other games with mechanics like this where the ai performs as well with relativly little help. I'm still waiting for someone to name an alternative. And there is so much vague talk about how much better the AI could be, but very few specifics, atleast ones that will make a significant difference and are at all simple to impliment.

As to the diplomacy..its an inherent part of the game. There are diplomacy enhancing modifiers that result from researching techs. These mechanics wouldn't work if the AI acted as a human does, nor would any type of cultural victory or non war related strategy. I may wish that the game would go in that direction, but there many other people prefer to be able to manipulate the ai.

I know I'm coming off as dickey in these replies, but I guess what I'm looking for is a bit more empathy. I agree with a lot of the problems with the AI noted here, and I'd be all for making the AI more agressive (i.e. i'm a big proponent of diplomatic relations suffering as a result of holding an increased proportion of habitable planets, or having shown a high degree of tactical ability), but the devs have other priorities that aren't necessarily in line with mine. It's what I'd like, but it isn't what I expect to get.

Anyway, that's my piece...sorry for the lectures.

True. But that is the design choice of GC2. They decided to have a completely open game that the AI was supposed to be able to handle. If you're going to make ship design so powerfull, you'd better do a good job of programming the AIs to use it.

Also I believe the modding flexibility has hurt the base game. It's a great feature, but because of it the AIs need to be immensely more complex.

Again, that is a design choice GC2 took. Personally I think they bit off more than they could chew, and with TA it looks like they're taking an even bigger bite. I would think people would want better AI over what's in TA, but people are dumb...