Impressive AI play

[dexters]

Well, I decided to try and watch a game with the AI going at each other. I generated a random map, standard size (100X100), pangea for 8 players. I created 1 tile mountain in the sea and set that as my starting location. This effectively puts me out of the way for the AI.

I won’t bore you with details of who is where but I will mention that there are two continents. On one continent, the Scandinavians took the top third. The Chinese took the middle third and the Romans took the bottom third. At around 200 BC, the Scandinavians started attacking the Chinese. And impressive it was. Streams of warriors, archers and swordsmen poured down from the mountains, quite literally.

The Chinese response was swift, beating back the first wave of attackers, and a handful of warriors, swordsmen and archers began moving up to engage the nearest Scandinavian city. The AI fixes appear to be working wonderfully, as the AI took the mountain route and effectively slowing the Chinese advance. In effect, the Scandinavians used its own terrain to its advantage.

A few turns after the Scandinavians attacked, the Romans joined in. They moved their Legionary and Archers as well as Legionary/Settler couples into Chinese territory and plopped down cities. The Chinese region bordering the Romans was largely newly settled and the Romans promptly invaded a few 1 pop towns and replaced it with their own towns.

On the other front, the battle had became a stalemate. The Chinese advances were beaten back, but despite their numbers, Scandinavian archers and swordsmen began to meet stiff Chinese resistance beyond the 1 town they took. Chinese horsemen began harassing the offensive units and taking them out before they could do any good. Soon after, the attacking stopped. It’s apparent the Scandinavians made peace.

With one front secured, the Chinese turned around and unleashed hell on the Romans. Horsemen, and archers poured out from Chinese cities and the Roman legionary, spread thin through Rome’s new conquests were picked off by stacks of 3 or 4 horseman/archer combo’s. It was really an amazing attack if I may say so.

The Chinese just kept coming. They had superior units in their horsemen and superior numbers. Typical battles saw the Romans attempt to counterattack the Chinese offensive with Archers. While they were somewhat successful, the damaged offensive units would be promptly taken out by the mobile Chinese horsemen or a nearby archer in the Chinese turn. The Roman offensive never gained the momentum.

Then, the Chinese landed a horseman behind Roman lines, at the very edge of the continent, and promptly took a 1 pop town. They had snuck a galley there. The damaged horseman then fortified itself to heal. A nearby spearman fortified in another 1 pop town didn’t even come out and the Romans simply had no offensive units to send out to attack. Unfortunately for the Chinese rider, his next attempt at the next town ended in its death.

China was advancing quickly. All of the Roman acquisitions from the Chinese were retaken, and 1 or 2 of the newly settled roman towns were wiped out. The Chinese then took Hispalis, a city northwest of Rome in a determined charge from a stack of oh… 3 horseman, and perhaps 2 archers. The 3 defending spearmen didn’t stand a chance.

Roman resistance stiffened after Hispalis’ fall. The Chinese attempted to move its horsemen towards Rome, which was only about 4 tiles away. Surrounded by hills, the defenders had the advantage. Several attempts by horseman/archers to break through the Roman defenses failed. Each charge was answered by a counterattack from the legionaries and archers that pushed the Chinese back to Hispalis. One particular incident had a Roman offensive, with about 3 legionary and 1 archer attempt to retake Hispalis. The attempt failed. After the 3 legionaries were killed trying to take the city, I saw what I can only coin as a “strangely human” behavior. The last archer, even though it is an offensive unit did a defensive thing. He cut the road connecting Hispalis to Rome.

Again Chinese horseman, the act may have been futile, but the Chinese had had enough of trying to take Rome. They concentrated on Neapolis, a city which had been isolated by the all of Hispalis, took the town and soon the war was over.

After that, it was around 700 AD and it’s been quiet since. No other wars and its about 1,300 AD currently. I’ll play some more tomorrow and see what happens. I dread the modern era, since they may find my settler on a hill and kill me. Then it’s game over.

Anyways, for those of you who use the debug mode, do you know if there is any way for me to access the AI’s trade and diplomacy screens? And if there is any way to watch them play without a human player?

[Kuciwalker]

How do you use "debug mode"?

[Fosse]

Dexters,
Sounds fun, and I'd like to think the AI was really using terrain to its advantage... but is there any evidence Scandanavia would have fought differently had those been grasslands it was crossing?

I'd love to hear what went on in the rest of the world.

For a truly impervious "Player Island," try ringing your one tile mountain with coast, then ringing that with a circle of mountains. Since nobody can build cities on those mountains to connect your moat to the ocean, there is nothing anybody can do to get into the waters next to you. Safe haven, even from marines.

[Kuciwalker]

Not nukes though

He said he had himself start on a mountain. So he can't build his city. This means his settler and warrior can be destroyed

[Fosse]

True... but if they can't make contact with him, they will never be able to get a map with him on it... and they won't KNOW to nuke him... make an even BIGGER moat, to prevent contact, if nukes worry you!

Of course... they could always fly a recon mission!

[Kuciwalker]

They could know to nuke him if they nuked the tile he was on, or near it. It's kinda stupid, but nukes reveal the tille they attack and everything in a one-tile radius.

[Nakar Gabab]

Not even a computer is stupid enough to shoot a nuke into the ocean. I think you're cool.

[Kuciwalker]

I was bored once, so I shot half my arsenal into an unexplored patch of ocean, hoping I'd hit something. That's when I found out. Now I use exclusively nukes for exploration j/k of course

[dexters]

Skywalker, Dubug mode is available in the PTW's scenario editor. I just started using it after Catt showed me how. I'm still waiting to see if anyone knows if it is possible for me to go into the AI's diplomacy and tech screens. It would be far more flexivle if we can do that so we can see who is at war, trades, and their tech level.

Sounds fun, and I'd like to think the AI was really using terrain to its advantage... but is there any evidence Scandanavia would have fought differently had those been grasslands it was crossing?

Well I've attached an image of the mountain pass Scandinavia used to invade. Note that they are currently not at war so their troops are just lazying around and not massing anywgere. You can see that it is a line of mountains flanked by jungles. 100% d bonus to Jungle's 25%. The Scandinavian attack followed the mountain in one straight line. It made the Chinese counteroffensive falter because of it.

Would they have went through plains or grassland if it was the only tile available? Probably. But So would a human player. The AI seems pretty good at using terrain to its advantage. What needs work on, is then naval aspect.

I'm not giving my final word on the naval invasion aspect until I see a few more attempts at naval landings, but in the war I watched, both Rome and China attempted their own landings. As I mentioned, China landed a lone horseman with their galley behind the Roman lines It was able to take one Size 1 city but got killed trying to take a second one. But why not 2? A galley can carry 2 units. The Romans landed a spearman on a Chinese island, destroyed an improvement on a hill and just fortified itself there, watiing to be killed or for reinforcements that never did come. Had the landings included a legionary for offense, the Romans might have, through the RNG took the city, since it was a size 1 city, that was on an island and was difficult to reinforce. China couldn't pop rush a unit either.

[badams52]

Originally posted by dexters
Would they have went through plains or grassland if it was the only tile available? Probably. But So would a human player. The AI seems pretty good at using terrain to its advantage. What needs work on, is then naval aspect.

From my experience and use of the pathfinder, I believe when units have a destination in mind, they always choose (a) the fastest path and (b) the path with the most defense if there is a tie for (a). So the AI like to take to the hills, but you can draw him down to the grasslands/plains/desert if you get there first.

I'm not sure if he takes rivers into account when attacking, but when I'm marching on cities, I always attack from the side where I'm not crossing a river, even if it means I take an extra turn to get there.

[dexters]

Additional Thoughts

Well the game ended with a diplomatic victory for the AI russian civ. The AI actually found my settler on the mountaintop and started contacting me. They also declared war on me. Hehe. no attacks though.

There was however one last war. China attacked scandinavia and dragged Japan into on their side.

A few thoughts on the naval aspect of the game.

1) The AI seems capable of decent bombardment and encirclement strategies.

2) After watching the AI play, I have increased respect for their naval capabilities.

The Scandinavians started the war with five Frigates and was unable to beat a small Chinese fleet of ironclads parked off its coast and was practically helpless when Japanese ironclads appeared by their core cities and started to bombard the hell out of them and the surrounding improvements

By the end of the War, the Scandinavians were able to pump out 5 or 6 Ironclads and they scored several key naval victories over the Japanese. They sunk about 5 of the Japanese ironclads, practically decimating the Japanese fleet presence and sunk 1 Galley + 2 units and 1 Galleon + 4 units.

3) The AI's attempts at Naval invasions still need work
It's improved over previous incarnations. The AI however tends to send poorly defended (read: no defensive units) and oftentimes expendable out of date units in these adventures.

Perhaps this is Soren's way of making sure the AI doesn't committ its best troops and suffer a setback at sea that ruins their advantage. I would argue however that unless the human player is the "save/roload" type who replays battles again and again to avoid losing any units, then the AI would infact be better off sending their best units in a naval invasion and risk being caught in a naval firefight with human/AI players.

The worst that could happen is that they die or face a reloader player, in which case, no amount of AI tweaking is going to change anything. The best that could happen is that they land a superior force in a location and throw the other Civ's defenses off balance.

In this 2nd big war, there were in total 3 landings of note. The 1st landing was piss poor by the Scandinavians against the Chinese. They landed ONE medeival infantry behind Chinese lines on an Rialroaded tile no less. It didn't stand a chance.

Soon after, Chinese/Scandinavia made peace. The landings shifted. Japan attempted to land a Galleon next to a Scandinavian city and almost succeeded. They had 2 medeival infantry, 1 samurai and 1 Cavalry and managed to almost take Odense (see screenshot above for location of the city). They took down the last defender to 2 hp. What perplexed me was, there was a galley with 2 more units just off the coast. Instead of coming in to reinforce the Japanese position, it sailed off. I think in such cases, when the AI has parked troops outside a city and is attacking it, the AI should know the defensive status of the city. A human player certainly would know. If they see a damaged unit defending the city, it usually means it is the last unit in a stack or all the other units aren't even worth putting before that unit. Well, so that was the end of that. the Scandinavians counterattacked with their Cavalry and the force was wiped out.

This time, the landings shifted to the Scandinavians. It was actually 2 separate attempts. The first attempt was intercepted and the escort destroyed. After regrouping, the Scandinavians had an impressive fleet. 2 galleons and 2 escorts, a total of 5 offensive units, all Cavalry.

This may have worked about 50 years earlier, but facing an industrialized Japanese city which was just about to turn out its first Infantry unit, the landings were crushed when the Cavalry threw themselves against a well defended size 12 city, totally ignorazing a spearman defended Yokohama just next to it.

There was infact another Scandinavian invasion in the works, this time with 7 cavalry units. The invasion was called off however when a peace treaty was signed.


Recommendations.

Some tweaks that I think would make the AI more competitive against each other and keep human players on their toes more.

1. If the AI absolutely must keep some of their older units because of costs in upgrading or a propensity to not disband them, they should use them like humans would, as shock troops in a land invasion or a sea invasion, but not as the primary invasion force, which they are often used for in naval landings.

2. Better Intelligence Gathering. The AI's intelligence apparatus is totally unknown to me (can anyone shed light). But there is enough free information in the game mechanics that human players have access to that the AI should also know. Like the size of the city they are attacking and the kind of infrastructure they face. Raildroads? or Roads? If the AI HAS the gold, it should investigate the city it is about the attack and find out the forces needed to take it.

Even without investigating cities, simply using the free information would vaslty improve AI naval offensives. If a unroaded tile is available, close to a city, the should land there and set up a defensive perimeter. Landing in a railroaded tile and then proceding to attack a well defended city just doesn't make sense, when a human player would quickly see a spearman defended city next to it and seize on the opportunity to attack that city.

3. Varied naval stacking.

The AI is too predictable. It is always 1 transport + 1 escort. I'm not saying this is a bad configuration since I've used this myself from time to time. But the AI uses this exclusively. The key disadvantage here is that if the invasion force is intercepted at sea and the escort destroyed, the transport usually stops dead in the water waiting for the next escort and sometimes, they get picked off.

A solution would be to stack more units. Why not 2 escrots + 2 transports? Or simply send more escorts out per transport? These decision would naturally be made based on the AI's resources. If the AI has the offensive ships in the general area, it should divert the ships to reinforce the landing force. If it does not, the 1 escort rule may be the only option it has.

The Japanese had the navy to the able to cover their invasion force with at least 2 ironclads. Instead, they went with the 1 escort + transport rule while their other ironclad units where running around the Scandinavian coast bombarding. The only reason the Scandinavians were able to intercept the Japanese invasion fleets and destroy them were beauce of the poor escorting strategy.

The best solution for me would be variety. So the next time I see an invasion force, I would wonder if it is 1 escort or five. And I wouldn't be so confident in sending my lone destroyer to intercept.