Best AI I have ever seen hands down!

[dexters]

Originally posted by vmxa1
I will with hold my admiration for the AI. Last night Hammie sent 28 Calv to one square and 26 INf/Rifle to another square in the same city. No one was fighting at the time so I suspected I was in trouble. I only had about 60 units total. Hammie had another 75 or so Inf in reserve. I asked him to remove and what do you think, he decalres war. I was worried, but not much as I knew he would not send those reserves all at once.

He is why I am not so sure the AI has a clue. .

The AI's chief weakness currently is its difficulty in handling 3 movement units. Those Cavalryies could have parked in their borders, perhaps even out of sight.

After the Declaration of War they will then move in and attack your cities. If it is successful, cultural border exapands and it can then move in their 1 movement units and hopefully close enough to defend their offensive forces. Once they have a city under their control, their 2 to 3 movement units will then use it as a base of operations. cavalry will move out, attack, and retreat back to a defensive position either in a city, on a mountain and always behind A STACK (not one) but A STACK of the best defensive units they have.

As mentioned before, the AI needs to switch strategies. If the multiple SoD moving at a snails pace and getting slaughtered because they park 1 tile from you city inviting the human player to destroy it, it may stay 3 tiles away next time and use its multiple movement units to attack, and retreat back to a fortified position.

That's their chief weakness offensively.

The 2nd weakness is the AI's general risk aversion. It will keep a minimum # of units in their cities at all times. These units are the non offensive tagged units. Whereas humans will tend to shuffle their units around to meet emerging threats, frontline AI cities tend to become helpless islands that can be cut off and destroyed one by one after you've destroyed the AI's mobile OFFENSIVE force.

While I think it was imaginative and efficient to divide AI units into offensive and defensively tagged units, there is something to be said about giving the defensively tagged units special instructions on how to move and react in times of invasion, so that fronline cities can receive some semblance of a co-ordinated "relief" force from the back-line cities, instead of the common phenomena of the 1 offensive unit the Civ managed to produce that turn doing that attacking and the moving, while all the non offensive units simply sit in their cities twiddling their thumbs.

With RR and multiple movement units on roads, the AI offensive units like to end the turn in an exposed position. Cavalry can usually move from city to city (even on roads) and have enough movement left for an attack, and then move back to a city to hide. The AI has trouble doing this. The strategy is only used when A) pure chance B) The unit is damaged and runs to a city to heal.

If Soren fixs this for the next PTW PATch OR perhaps adds it as a bonus for Conquests XP, I'd be very happy. Although there are always other things I like to see improved, such as AI's use of navies. But I'm speaking hypothetically of course. I'm pratical here and understand that Soren can only fix so many issues given time constraints and his involvement in other projects.

Bottom line though, if we're expected to pay another $40 this Fall, improved AI is a must.

[vmxa1]

What you say is correct. I don't know when I posted that, but if it was before 1.14F for PTW, then the AI a bit better in at least managing it empire.

[dexters]

This thread is far too interesting to let die. I hope no one minds my attempt at reviving it.

That said, I paid pretty much release price for PTW mostly for the AI. The new units are nice. New civs and buildings add variety. But it's the AI, and I'm more or less please with the improvements.

What worries me is that Soren was not involved very much in PTW. Granted it was supposed to be about multiplayer.

With Conquests XP the fixed alliance thing is really making me excited about the diplomacy possibilities. That would imply some major AI work in the diplo area. I just hope Soren and whoever is helping him also get around to AI defense/offense stance, and add some more strategies into its repertoire. CAPTIAL Isolaiton was a nice surprise. AI leans

[Daz]

Where did this thread come from?

Anyway... shouldnt we wait till conquests is out? Are there any AI improvments in Conquests?

[Virdrago]

I'd like to see some, though I can't think of any right now except for how the AI defends it's homeland...

[Gen.Dragolen]

I can think of one major change to the diplomatic model: reduce the previlence of RoP agreements between AI Civs. This would do wonders for the AI´s offensives.

I have been trying to think of a way to make it a little more realistic when you tell an AI Civ to take a long walk off a short pier when they try to extort a tech and a map. The more militaristic civs like the Romans or Japaneseare bad for this. They will be on the other end of same continent, 3 civs over from your ´pathetic little civilization´ which only happens to be about 100 points more powerful, and they will march 15-20 units, almost all of their forces through the heart of the other civs.

Now if they were all allies, I could see it. However, the usual rule of thumb in state craft is that any foreign troops on your soil is unacceptable. The threat of them taking up residence is just too great. Logistically it used to lead to the starvation of the people along the march´s route.

I would like to see the AI consider any foreign troop movements through its land to be an invasion and treated as such. Only allies that have declared war together should be able to have their allies move through their lands, and only in the direction of their stated enemy.

As for its offensive strategies, yes, it still hits piecemeal and without supporting units. If the AI brought up some artillery and attacked with more than just 5-6 units when going after a city, it would be as devastating as a human player. Though I have seen some wonderfully agressive attacks using cavalry to its full potential.

But the success of these attacks have always been rather dubious: they still need to redo the combat system. I have seen cavalry wear down mechanized infantry in the strongest positions. Imagine seeing 8 cavalry attack a mech inf. unit fortified on a mountain. Only 1 cavalry unit died in the attack. And this happens all the time. I have read all the postings about how the system is trully random and is designed to give technologically inferior troops a chance, but that tears at the heart of the technology race. The whole point to research is to find ways to improve your units so they can destroy the enemy. And if the last action in Iraq isn´t enought to demonstrate this, the previous one was.

It still galls me to see units like cavalry or knights knock off modern units. The game designers should have paid more attention to the impact that one technology has had on war: the machine gun. The main change from the US Civil War to WW I was the machine gun. Any civil war infantry unit going up against any WW I unit would have been slaughtered to a man. Same with Cavalry. Just like an Ironclad would slaughter any wooden ship. ANd just like any destroyer would slaughter an Ironclad.

Technology progresses at an exponential rate in the world, where what is bleeding edge today is on the junk pile next year. Canada is finally getting around to updating the technologies on our fleet of CF-18´s. The were purchased in the 1980´s and are two generations of avionics out of date.

This may only be a game, but reality really is a break neck ride. From our wandering off the savannah´s in central Africa 40000 years ago to just making it through traffic to get to work. We need a sophistacted game to provide some diversion from that reality.

D.

[dexters]

The combat system is not a problem. If you've got succifient troops its more or less the same as having average performance since the streaks tend to cancel each out and the majority of the time, you get expected performance. We just hear a lot of *****ing because of the human nature to remember the outliers.

If you've got a few units however, and you ask yourself if you could afford to go attack A or B, then it is a kind of gamble. But you know, war is a gamble. Bad things happen and the RNG simulates uncertainty well.

In anycase, the problem with the AI is not the combat system but how it handles its units. The combat algorithms appear to have been perfected during development with basic 1 movement units from the ancient era. The AI do relatively well with them, but go beyond these one movement units and to more modern units, they start to run into trouble.

One thing they absolutely must fix is the city production system. The AI is unable to check in on their cities during mid production and re-evaluate a build. In some of my test games, I would give an AI building the Great Library Education Tech, but the AI would continue for another 30 turns, seemingly not knowing that Education makes the GL useless. This is a great disadvantage to the AI. If there's a concern over computing time required, I would suggest the basic idea of management by exception. If a city is taking 120 turns to build something, the AI would go in and have a look. Usually the problem is because the AI's Civ traits force it to build A library in a corrupt city instead of going Courthouse first, reduce corruption and then build the rest of the infrastructure faster.

That said, there's a lot of things the AI do very well. There are times when I was almost sure the AI would do A, but they surprised me and did B. I think there's a lot of things under the hood, either programmed or created implicitly by the AI subsystems interating with each other that we don't know about and that is making decisions that can still keep veterans of the game guessing.


Anyways, It would be nice if they would keep patching the AI indefinately. It's the only way I think to make the ultimate learning AI. I recently loaded up my first full game of Civ III and was surprised by the sloppy AI. was Poor expansion, keeping tons and tons of legacy units around, poor map exploration and that are just the stuff I could see because the patches doesn't correct mistakes made by the AI in the past. It underscores how far the AI has come as it has been tweaked, adjusted and I'm sure new lines of code added to help it cope with various situations, strategies and conditions that wasn't envisioned during the design phase.

[Gen.Dragolen]

dexters,

I agree the AI has progressed massively since the first version of CivIII was released.

And there are alot of things that it should be doing, like re-evaluating it´s production each turn, but we will see shortly (well in a few months) how they have modified the programming to take even more details into account.

However, I still think they haven´t captured the essence of the tech race: in my latest round of testing, I have managed to get to Military Tradition about 40 turns ahead of the rest of the continent I´m on. That has left me at least 4-6 techs behind everyone else, but it allowed me to finish off one smaller civ and do some serious damage to another. I won´t comment on the casualty rates for cavalry attacking longbowmen in open ground, but needless to say, the AI should have been either trying to steal Military Tradition or buy if off me, just to keep pace in the arms race. Instead, they were demanding it from me or offering a World Map and 1 gold.

Now I don´t know if this is due to the personalities the AI Civs assume or to circumstances in random maps, but there is still some work to be done. One thing I really miss from Civ II is the Diplomat unit. It was annoy as all h&*l to have units bought off just before you were about to launch them into a city, among other things. The capacity to investigate cities from the one interface was a welcome change, but it is still too hard to execute spy missions, both in terms of cost and the chances for success.

I have yet to see the AI employ spies in any game I have played. When I have done looking for their spies, and the witch hunt fails to turn up anything, all I have afterwards is a war with however´s spy I tried to find. Same thing with setting up spies. Using execute safely has only succeeded a couple of times, but has started most of my mid-game wars. And no one else in the games at the time had Espionage. Again the advantage of having a tech before anyone else has been erased or hobbled. And having Espionage restricted until late game takes away much of the possible intrigue from the mid-game. There have been official spies in use since the Ancient Egyptians. Court intrigue relied on prying eyes and sticky hands. Granted the AI knows where evey one of our units are, there should be some penalties and benefits to employing agents.

Well, at this rate, we may wind up making an AI Civ that will be classified as a state secret and used by the Pentagon...


D.

[DaveMcW]

The AI does use spies if they have waaaay too much money.

[dexters]

Abstracting espionage is the best decision Firaxis has done. I know some people hate not having spies around, but like the trade system, in Civ III these two aspects were moved UP to the macro level. In Civ III trade is done so that you approach it from the perspective of a leader dealing with foreign nationals and it takes the principle of "if you want it, it is done for you" instead of "if you want it, spend 20 turns moving that camel from your cities to the rival capital" The foreign trade system in CivII was so tedious that I just gave up after a while. The same goes for espionage.

On that topic of spying, is remains a mystery how the AI benefits from it since some of its decision making subroutines do have access to information like where all the units are. Part of the reason why human players use espionage is therefore somewhat defeated. That said, the AI do use the spy mission when it has a lot of gold. But usually, if it has a lot of gold, it is maxing out its science, rush building improvements and trading like mad. Catt has a nice collection of screenshots showing AI espionage attempts in one of his games. If i manage to find the thread I'll link it here.

On the technology front, the AI does value different techs differently and are know to gun for certain high value techs. I'm not sure what can be done, other than tweaking their preferences to make it more comeptitive techwise.

I would be nice if each AI civ were given a tech goal, that is, the AI decides to get to a certain tech and builds a strategy around it, stealing the techs along the way if neccessary.

The one thing that the AI clearly lacks is foresight. Granted computers can't really think, but I'm saying it in terms of a programmed routine that looks beyound the current state of its existence. It does this during a surprise attack. It will move its units into position several turns ahead of time. But if would be interesting if the AI was given a goal on the macro level as well. An AI will for example strategize against another Civ, and use proxy wars, direct aid and other tricks to get their goal.