AI Naval Power Play

[dexters]

Reading through some of the older threads, from the greats (Theseus, Catt, Alexman et. al) the issue of the AI use of navies have coem up numerous times.

I was wondering if there are those of you out there who have come accross a strong AI naval power.

I define strong as

either
1) A strong navy that rules the waves
or
2) Capable invader

Both need not require (I'm lowering my standards a bit) But I am trying to start games that will feature such AI opponents as I feel the naval aspect of the game is sorely underplayed.

Thanks.

[Catt]

Yes and no on both counts.

1. I've seen massive AI navies such that they rule the waves. But -- they don't necessarily rule the waves effectively. For example, I've seen many a continents map where the KAI on the other continent has a huge navy -- during war, I can often do no better than take the relentless bombardment and strike with artillery from the coasts, with an occasional foray out from port. But while I can't take on their navy ship for ship, I can assemble a convoy that can make the crossing with transports intact and land a sizeable force.

2. The AI is usually not very adept at intercontinental invasions -- this topic gets a lot of play in the forums. I have, however, seen the AI land 16 units (modern units!) at a time at one concentrated point. Sadly, this seems more random than anything else since I can never predict it and I also see the much-lamented landings of one infantry, two longbowmen, and a warrior from an 8-capacity transport. If the AI could only land with concentrated force and then keep the reinforcements coming fast and furious . . .

Catt

[Jaybe]

... If the AI could only land with concentrated force and then keep the reinforcements coming fast and furious

But they DO! Unfortunately, they do NOT take into account that the 1st wave died the turn after they landed.

Their definition of "fast and furious" just isn't in our league.

[dexters]

Thanks Catt for your well thought out input, as always

Jaybe, Naval invasions are always tricky. Against AI, its one of the toughest forms of invasions a human can play.

That said, I think the AI can do well if it learns or errr... is prodded by the Soren AIprod Co. to value large naval landing parties highly.

It can already surprise human players by landing in the strangest of places. Imagine 2 or 3 such landings over the span of 3 turns, and you have the recipe for a dazed and confused human player who must divert forces to deal with these. Cities will be lost. People will die, and the best thing of all, the AI may just own us.

It very nearly did in my recent regent game when the Greeks launched a surprise attack. Their land forces moved in from two fronts. From the north, came a stack of Knights, swordsman, and Med. Inf. And from the SouthWest They had a stack of about 15! I kid you not, 15! medieval infantry through the Jungle, supported by a Hoplite. A border I shared not with the Greeks, but with the Spanish! What a doozy that was. There was no way I could have known. I sent my lone bowman to try to pick off a Med. Inf. The longbowman died. And my Numidian Mercs didn't stand a chance.

The second city was razed from a naval invasion. It was a backline fishing village with a legacy unit (warrior) in it. It was elite and I was keeping it to milk for GLs, but nope. Alexander landed 2 hoplites and a swordman and procded to take my size 6 town and raze it to the ground. A couple of more of those landings on my long undefended coasts and I would be in panic mode.

[dexters]

Just an addition I want to make.

Capture > Build Airport > Airlift a ton of units strategy should be taught to the AI. It's the only way to keep a beachead in the late game and humans, even casual players, know its value and practice it with varying degrees.

And 2 or 3 AI cities with airports airlifting units is enough to keep a challenging invasion going. Anymore would be unbalancing to casuals.
---------------------------

conversely, the counter to this strategy should also be known to the AI. I think that Soren's "broad" rules for the AI may have to be cut back in this case with a simple but direct condition. If Airport =1, bring the artilerry , blow the city apart and hope you hit the airport ().

[Theseus]

I've seen some impressive AI naval performance, but nothing overwhelming.

As opposed to Civ2... I remember some tough tough games, where the only way I could win was to capture the leading AI civ's capitol, thus destroying the SS... I'd send out a truly massive intercontinental invasion, only to see it reduced to shark meat on approach. Nothing like that in Civ3.

[vmxa1]

I have not seen anything exicting. I have, like Catt seen large navies, that I could not handle for a long time.
I fixed what they broke, hit with arties and sunk any stragglers that I could. Eventually, I get bomber and battleships and start to thin them out.
They seem to not stay together in armada's. nor do they invade with a sizable force.

[Kaos XIII]

1. I’ve seen strong AI navies. It also helps that I reduce the cost of ships, to encourage the AI to build more, and they do. I’m not sure they rule the waves, however. The problem is that, there is nothing for the navies to control. Blockades are nearly impossible. One of the reasons navies even exist, is to attack trade, but we can’t do that in this game.

2. Never seen it. In early wars when there aren’t that many units around, sometimes the 2 or 3 that they land, can be effective. But in the modern age, I still see them land 2 or 3.

[Cruddy]

1. I've never seen a real large AI armada - biggest was a 4 carrier/4 battleship force that was in a staggered formatation. It bombed quite a few coastal tiles but once I had fighters in that area it's effectiveness was zero.

2. Biggest invasion I have seen was 25 units dropped off by 4 transports. 17 of those units were tanks. It would have been bad news, because it was on a continent I had recently overran. I wrote would have been because I had had a transport relay going for 4 turns anyway, so I had enough armour on the island to deal with it in 1 turn.

[dexters]

Interesting thoughts all around. I've seen the AI with some rather large armadas as well. But the thing is, it seems to be a hit and miss affair.

On a 1.29f vanilla Civ III game, the Greeks had a large armada of 20 ironclads, 4 frigrate and 4 galleons maybe because it started on a large island. On my recent PTW 1.21f game, they had barely 2 ironclads and a few galleons.

Theseus, if you feel the Civ2 AI navy is more challenging, do you think the AI should be allowed to blatantly cheat to get the desired result? Or do you think there is room to improve the current code to get a more desireable result?

[Dominae]

I'm reasonably confident the AI naval algorithms for non-transport ships looks something like:

1. If there is a transport to escort, meet up with it and escort it.

2. If not, beeline for the nearest enemy border, unless there are enough ships there already.

3. Once at the borders, loop looking for something to do: a) if there are enemy ships within striking distance and the odds are good enough, engage in naval combat; b) if not, enter enemy borders, find an improved tile, and bombard it.

The reason I know this last bit is true is that in the Total War AU the AI kept bombarding tiles within my borders that no city could access (off a little island).

This is simplistic, but works fine if the AI has a production advantage (i.e. can pump out more ships than the human player would reasonably want to). But as any "brute force" method, it runs into problems when there is not enough time and resources. In other words, the fact that the AI is not actually programmed further than this makes it look "stupid" when it only has a few ships in the water.


Dominae

[dexters]

Good stuff Dominae.

I've been mulling over the idea that perhaps Soren et al should give the AI layered strategies.

That is, instead of general rules that govern their behavior at all times, additional strategies will kick in given certain conditionals are met.

For example, if the AI has a small navy (relatively and absolutely) it may move its ships in wide arcs, out of sight and hunt your shipping.

If they are at war, their ships will stay in port. move out in between turns to bombard, and attack your shipping before retreating to port.

A general rule that may prove effective at all times is also for the AI to value wolf pack formations highly and have their ships move around in packs of 2 or 3. And escorts will likely be in this configuration.

Just to go OT a bit here, the idea of a layered AI comes from the fact that as Soren noted in one of the old AI threads, the AI can "fly off the rails" if it has only a few cities. I thought perhaps in those cases, the AI could be given specific instructions. For example, giving it some standard human OCC strategies so that they can still be hard to take down even when reduced to a one city empire somewhere.

Thoughts?

[DaveMcW]

Since this thread is taking an AI wishlist turn, I'd like to add my own.

First, the Airport beachead idea is great, it would be my #1 priority too. A variation would be an airbase made with a captured worker, which avoids the problems with resistance, bankrupt/democracy, and size1/communism.


I am impressed that the PTW AI knows how to use marines + transports, bombers + aircraft carriers, and tactical nukes + nuclear subs. This is a great step forward.

It's awesome when 2 marines take a city and 6 tanks on the same transport unload there and wreak havoc. Unfortunately 90% of the time the marines fail to take the city! I would like to see the AI treat transports as "marine armies", loading them with 100% of the same unit and only attacking when at full strength.

It would also be nice to have the AI bring some defensive units when landing an assult force. Unfortunately the "railroad rape" will still wipe them out unless they land in overwhelming numbers.

And a final wishlist item, teach the AI to soften a city with concentrated air bombardment (or nukes! ) before sending in the ground troops.

[Cruddy]

Originally posted by dexters
Good stuff Dominae.

I've been mulling over the idea that perhaps Soren et al should give the AI layered strategies.

That is, instead of general rules that govern their behavior at all times, additional strategies will kick in given certain conditionals are met.

For example...

My own pet hate of the AI Naval Strat is that it DOESN'T know how to use subs properly.

If you look at Cold War, a lot of money and effort was put into sub picket lines, to keep an eye on the enemy fleet's well away from your borders.

I usually have such a picket line, and it almost never runs into an AI sub.

[dexters]

AI's don't need subs to see your units. It knows where they are when it needs to find them (like attacking_

But I don't think there's a central intelligence that factors this piece of information in when it makes other decisions. It's in many ways, a way to blind the AI of this cheat in some actions, but use it in others to help it make better decision.

I'm currently working on a post regarding the AI's use of naval power and naval landings based on a DEBUG game. It's actually surprisingly simple why the AI can't pull off a decent invasion most of the time. I think it's fixable if Firaxis gives the naval AI a few more lines of code to help it along.

Just a summary of my findings

1) AI invasion forces are highly fickle. They sail back and forth midstream when the target changes. (ie: a city it is aiming for it taken by another unit.) Instead of reinforcing the garrision at the city, it moves on to the next target. I really have to wonder how many landing parties we never see in our real games because of this effect.

The same effect also happens if its target city is taken by an ally. I've seen an AI landing party sail back and both between targets for 5 turns. The AI made peace before it had a chance to land its troops. In this case, the AI's ability to see all units work against it. The happy accidents that humans sometimes get, when they send in fresh troops only to find that they just arrive in time to save the city from a massive counterattack never happens with the AI. The additional forces headed for a target just sail off elsewhere, even when it is prudent (for a human) to reinforce a city anyways.

2) The invasion convoys rarely carry enough troops, and usually, it carries NO defensive units and thus cannot hold cities when counterattacked nor can it protect the attackers the turn after it lands.

3) It likes to spread itself thin. Offensive units will immediately move on to the next target instead of securing the city and waiting for reinforcements.

If 4 Siphani lands and 2 of them takes a city, the damange units will stay in to heal and there will always be at least 1 unit defending. The healthy units will however go on in their own little thing, usually, getting killed attacking or in a counterattack in the next turn.

When enemy units arrive to retake a city there's not enough Siphani left to defend. Instead of 2 damaged Siphanis and 2 healthy ones ready to pop out of a city to cut down the attacker, you have 2 helpless units waiting to be killed.

A lot of it seems to me like insufficient programming. In the search for general subsystems that works for all occasions (not a bad thing I might add), most of the work went into more general unit algorithms and not enough thought was put into how units should behave in the very uniqe situation where it lands an invasion force in foreign soil and must hunker down to defend its position.

The AI can get away with being a little sloppy if its figthing on land because its cities are usually near enough to reinforce a position and there is usually fresh troops streaming in. Naval invasions badly need a new set of if/then statements for the AI just so it can be somewhat competent at it. I don't think anyone is expecting anything more than that.

I've got a lot of crap coming up in the next few days, but I'll try to write something up with screenshots when I find the time. Wednesday is a good time I think.