AI brought tears to me eyes

[Bereta_Eder]

A spooky thought just crossed my mind. But I dismissed it. I thought for one second that if the AI could simulate coordinated attacks players would get exasberated and civ as a game would be destroyed. Virtually every wonder packed 30 size city would be easy target...

But this is rubbish. Civ would only get better with these attacks. Multiplayer proves it. Better attacks bring better defences.

Even though the name «General Grant» sounds only very vaguely familiar (american civil war? or maybe I just ridiculed myself ) I can't help but thinking what is so hard about programming attacks on one round. Or no-attacks on mountain fortresses.

Then again when a unit is in a higly defensive terrain square the AI usually bribes it, if it can. If not then it attacks suicidally. I think the ideas were there but they were not fully developped?

Maybe the lack of emphasis on AI combat is due to the intention of the programmers to set as a goal the «building of a civ» as a whole?

[Cyclotron]

Yes, General Grant from the American Civil War. Became one of our presidents, too.

Case 3: (this was brought to mind after you described bribing units on mountain squares) You guys must all know the WWII scenario that comes with CivII, right? You must also remember that the Alps (switzerland) is inhabited by two alpine troops in fortressess, on mountains. Talk about defense! Anyway, usually waht happens is France ends up bribing them and spiriting them away to their other cities, leaving Hitler a good mountain fortress to use. When I was the French, however, I let the Germans get to them first. They bribed the alpine troops too, but then moved a howitzer into one of the squares that overlooked a newly built road of mine. Remarkably, they kept the howitzer there, stopping my engineers trying to build a railroad and totally destroying my chances of a sucessful land invasion of Italy. I had to settle for Berlin.

Ming, your point is valid. Of course the AI is not the smartest one on the block, and certainly it does not share in even a sliver of your mental prowess. I do think, however, that the AI does have some ability that you don't give it credit for. After my intitial thrill of CivII and playing many games, I was dissapointed by this same AI and devoted my games afterwards to trying to see how the AI reacted to various tacics and how it worked in various unusual circumstances. Although most of these field studies of mine bore no fruit and reeked of stupidity, I dare say there are a few diamonds in the rough that you may be overlooking.

[Ming]

I fail to see the gems

Simple AI programming:
A unit next to a city will attack the city, no matter what else is around it. (nice to know when you want a barb king... just move a unit out of the city next to the barb unit if it has a king with it. It makes it easier to hunt it down after the attacking unit losses)

AI units are in love with fortresses. They love to camp out in them and attack anything that comes along.
While the traditional programming tells a unit to keep moving toward some destination, target, or searching... they will camp forever in a fortress.

The AI has no clue that it has NO chance to win a combat. A human will know what it's chances are of winning an attack. But the AI was not programmed to look at the attack realistically. What a simple task that would have been to add to the programming. Gee, do a simple look up... a warrior can not beat a phalanx fortified on a mountain square... DUH! Granted, an AI unit can get lucky... like a non vet horse taking a fortified phalanx off a mountain... but when the odds are closer to impossible, the AI shouldn't HAVE TO ATTACK!

There are tons of examples of just how limited the AI combat programming is.

So again... a seemingly well thought out attack by the AI... just dumb luck!

[Bereta_Eder]

Well some of the gems that are programmed (because it happens a lot) are:

1. The diplo out of the blue tactic.

You got a small city that's far away from your capital. A trireme (or something) appears. A diplo comes out and your city is bribed. Just like that. Answer is to have an ironclad (or something) patroling the waters (courthouse blah blah) but that doesn't guarantee it won't happen.

2. The spy with a smile fortress tactic

You got highly defensive units stationed in higly defensive terrain. The first thing the AI does is to try to bribe the. THEN it attacks suicidally.

Agreed these are not enough for a challenging game. Maybe in civ 3 there will be more? (I KNOW there will be )

[drake]

I had an experience similar to leftys in one of the first civ games that i played. (and this was only at prince level!)

The Japanese coast was just across the ocean and they made my civ look like a pimple on thier ass. I was at peace with them, but I soon grew impatient and sent over a small fleet to try and take a few cities. This I suceeded in doing, but it made them very angry. Within two rounds their massive armies liberated both cities i had claimed. Three or four more rounds from that, two ai transports chock full of armors arrived at my shore, pouring out 15 or 16 units on my coast. Within another 3 turns this batallion had taken my capital and surrounding cities, for I didnt have the technology to do anything. Needless to say it was a hopeless cause from that point and I called it quits.

However, I've not experienced such a rush in a long time, because my opponents never get as far ahead as they did this time.

[Blaupanzer]

Ming, just because much of the combat programming is primitive, doesn't mean it all is. I, too, have seen naval assaults that looked coordinated. Not often, but once every four or five games. It is also at sea, as I noted earlier that I saw the AI use nukes against one of my fleets, something it never seems to do against land units. Seems to me the programmer for naval warfare routines may have been a little more together, or had a lower number of branching options, or some such. The AI can be (should have been) programmed better than it has been, but not every single case is idiotic.

[Ming]

Blaupanzer... again, I'm not an expert, and I didn't write the code, or am I qualified enough to look at it and determine the capabilities of the AI...

However, I would still chalk up any seemingly coordinated attack as dumb luck. The AI floods an area with units, and they attack. That's seems to be the limit of their strategic thought process. Early in the game, when it doesn't have enough units on the board, the idiotic sucidial attacks are laughable at best.
Later, AND ONLY because they have built enough units, does it appear like they actually know what it's doing.

The programming seems simple enough to me. Build units and attack. When they are lucky enough to many units in the same area...

Maybe that's the best they could do with the programming. If I designed it, and knew that it would be limited in what strategy you could give the ai, I would probably do it the same way. Just keep building units and hope you can overwelm the opposition. I have seen nothing in my many games to think otherwise.

But, I have been wrong in the past
I would really love to hear from an expert who can take a look at the code and tell us just what "strategy" the AI uses... Until then, I'll continue to subscribe to the dumb luck answer

[Wiglaf]

quote: However, I would still chalk up any seemingly coordinated attack as dumb luck.

I don't see why. You mentioned the AI doesn't take in to account hit points in its attacks, but I fail to see how this would affect its ability to wage coordinated warfare.

Six battleships bombarding two riflemen that would otherwise block the path of four tanks that would pillage a railroad is beyond the AI, but this does not rule out a simple coordinated assault. A few ships, a few marines, and the AI can take a base - while not much, that is more than a computer that randomly tosses units around, as you seemed to imply. And if it's luck, it happens a whole lot, and at the perfect time.

Do you believe the AI does use some strategies players used from civ1 or would potentially use in civ2, as the manual suggests?
------------------
The fault lies not in our SMAC, but in our stars.
~ Vanguard, February 2, 2000.
[This message has been edited by SMACed (edited February 02, 2001).]

[Ming]

I'm not saying that the computer doesn't know how to load some marines on to transports and attack a city.
I'm just saying that there is no overall game plan except for to randomly send out units on hunt and destroy missions.

Let's get real... they didn't even program the AI to know how to use a carrier. If you want a good chuckle, just gift a carrier, fully loaded with planes, to an AI... and watch how it uses it. I will bet you the first act will be to attack some costal city, and go down in flames before a plane ever leaves it deck
Now that's strategy

[Bereta_Eder]

I really wished somebody would give unlimited money to Sid and Co. at Firaxis so they could never ever leave anything unexploited because of financial and time difficulties.

The carrier example is a disgrace. Don't tell me they couldn't program it, they propably didn't have enough time or money.

I am saying this just after I won my first Deity game and was a little bit dissaponited because I discovered that as soon as you have dealt with the unhappiness problem (that terrified me at my first two deity games that I abandoned - God Bless HG on my last one ) everything else is the same.

Still, I dont think you have to go to multiplayer to find challenge. A horribly disadvantaged start coupled with vikings and mongols next to you is challenge enough.

---Paiktis22 wonders off mambling something about damn expensive greek ISPs---
[This message has been edited by paiktis22 (edited February 02, 2001).]

[Theben]

I agree with Ming. The number of incidents involving what seems to be a coordinated assault by the AI pales compared to the number of stupid, pointless, &/or suicidal attacks.

It's like the infinite # of monkeys syndrome. Given enough time eventually they'll write Sheakespeare. Given thousands upon thousands of attacks by the AI vs. hundreds of civ players, you'll get a couple dozen example of what seems to be coordination on the part of the AI. You're just rationalizing random outcomes.

[Simpson II]

quote: It's like the infinite # of monkeys syndrome. Given enough time eventually they'll write Sheakespeare. Given thousands upon thousands of attacks by the AI vs. hundreds of civ players, you'll get a couple dozen example of what seems to be coordination on the part of the AI. You're just rationalizing random outcomes.

Doubtful. The AI would be unlikely to have more than about 20 ships, so these attacks presumably encompass it's entire navy. Say there are 20 targets it could attack with any given ship from a group of 20. The odds of them all going for the same target are ~1 in 20^19=1 in 5242880000000000000000000. That's ignoring the fact they may not arrive at the same time - even if there's only a 3-turn possible window for some strange reason, the odds go to something like 1 in 60^19 = 1 in 6093597400104960000000000000000000. I have played somewhat less than 6093597400104960000000000000000000 games of Civ, but I've seen this happen once. The AI turned up with 42 (!) ships. Shame they didn't have any AEGIS cruisers, though it still took 3 cruise missiles to get a kill...

The probable reason that many people haven't seen this is that they play a more effective game than others. I mean, Ming, how often can you remember seeing the AI build a battleship? Let alone 20 of them. You have to have a slow tech-rate so that the AI has time to build all those units (probably at random) and yet not be attacking them to avoid triggering off the infinite-riflemen code.

Btw, the Smac AI is perfectly capable of this sort of coordinated assault. The best I ever saw was a feint of one front with ~10 units which withdrew, quickly followed by another 20 at another point. It doesn't change much defence-wise, because it's still messed up in other areas.

[Ribannah]

The AI doesn't play totally randomly. Defensive units are always put in a city or a fortress. In wartime, all attacking units get a target assigned to them. This is partly based on the available information, and partly random.

As a consequence, if the AI only knows the location of one city of your empire, or the route to only one of your cities, it will send lots of units to that city. They simply start out from where they are, so normally you can expect them to arrive one at a time.

However, in some cases they will all arrive together, for instance when the AI had just broken an alliance that made all its units return to the capital. This may look impressive, but it's merely a co-incidence.

------------------
If you have no feet, don't walk on fire

[Roman]

I have to agree with Ribannah and Ming. The AI does not create taskforces "intentionally".

BTW: Simpson 2, your math is not correct on this matter. You cannot simply do 20^19 to get the number of places where 20 ships to appear.

[rah]

quote: Originally posted by Ribannah on 02-09-2001 08:21 AM As a consequence, if the AI only knows the location of one city of your empire, or the route to only one of your cities, it will send lots of units to that city. They simply start out from where they are, so normally you can expect them to arrive one at a time.

While I agree that the AI is just luckly if it looks coordinated. (will appear more often if their empire is linked by railroad allowing the units to arive at the same time....

The AI knows where your cites are even if it hasn't discovered them yet. When I was testing the AI, I would often go into cheat mode and look at their goto commands on units. They can target cities that if you check out the map from their view, they can't see.
In fact for one test, I checked goto commands, went back a turn and created a new city without defenders and checked the cheat menu and "low and behold" most of the units had been retargeted to the new city. Considering that there were no units within 10 squares of the new city, They must have found out about it by magic

So feel free to trade your map to the AI, get something for something they already have.

RAH
If anyone else has any other explanation for this type of behaviour, please feel free to post it so it can be tested.