Does the AI use creative tactics?

[Master Belisarius]

I go for the first option...
Sorry Rollo, but I'm with Mephisto and Alpha Kodiak: I think that sometimes the AI can do some "intelligent" actions, but really is not smart. In fact, I would prefer that the SE4 AI could cheat something and then, play better.

[Gandalph]

OK, I have completed this test. With all else equal, I put up a minefield on planet B and the AI attacked planet B and blew up, conclusion - AI cannot see mines. Next scenario, I put 5 small WP's on planet B, AI attacked and destroyed planet A. Next scenario, I put 25 fighters on planet B, AI attacked and destroyed planet B. Last scenario, I put 50 fighters on planet B, AI attacked and destroyed Planet A. Between scenarios, I wiped out previous defenses. From this, I conclude that the AI CAN see WP's and fighters and will avoid that planet if it does not think it can win.

[Arkcon]

Originally posted by Gandalph:
[QB] Well, the minister sent the warship back to the planet and scrapped it. QB]

ROFLMAO I didn't really predict this, I've seen it before when I take over an AI that's losing just to see what it's doing, often it will make and break it's new ships immediately. Must be some hard coded algorithim -- it must want have a certain amount of planets producing resources to keep ships. Typical conflicting ministers.

Gandalph -- I gotta say thanks for doing this and the next test. I wouldnt have had the patience, but now that you see a pattern, I'm intrigued. It can see WP and fighters but not mines(we shoulda guessed that one, or it would never bungle into a mindfield -- that would be too obvious). I'm in the mood to try it out myself. Questions remain:

1). Does it always find the right planet, or is there some ramdomness. Or is there some other rule, how nice the planet is, how close a planet is to it's homeworld, what direction the ship was traveling, etc?

2). So it can count fighters in cargo. If it can see the difference between 50 and 20, can it see the diference between 20 and 25, 20 and 21? Could this just be a non-obvious bug -- it should be able to count fighters in orbit, maybe MM did'nt put in a flag to hide this info when they are cargo? Does it care about small medium large shielded armored fighters or would a ton of cockpits and life support fighters keep a planet safe?

3). What can it see of the weapon platforms. What's it more afraid of large full of armor and point defence vs mediums with engine destroyers?

4). Obviusly it can see sats in orbit, but how about cloaked ones?

If I'm bored enought this weekend I'll try some of these

[Gandalph]

I can save you the time with some of these.


1). Does it always find the right planet, or is there some ramdomness. Or is there some other rule, how nice the planet is, how close a planet is to it's homeworld, what direction the ship was traveling, etc?

There were only 2 planets in the system and both were homeworld, large, breathable, identical population numbers, identical facilities, and identical distance from the warp point, which is why, in my previous post, I said "all things equal". The AI randomly chose a planet if no defenses on either. With WP's or fighters, it always chose the planet it could defeat. If it could not defeat either, it seemed to choose the lesser populated, (the differences were minute, ie 4184 to 4056). If there were sufficient defenses it would attack the weaker planet and , if it could not defeat either, it would hold from attacking.

2). So it can count fighters in cargo. If it can see the difference between 50 and 20, can it see the diference between 20 and 25, 20 and 21? Could this just be a non-obvious bug -- it should be able to count fighters in orbit, maybe MM did'nt put in a flag to hide this info when they are cargo? Does it care about small medium large shielded armored fighters or would a ton of cockpits and life support fighters keep a planet safe?

See above observations.

[QB}3). What can it see of the weapon platforms. What's it more afraid of large full of armor and point defence vs mediums with engine destroyers?[/QB]

Again, see above observations.

4). Obviusly it can see sats in orbit, but how about cloaked ones?

I did not test this, but based on my obsevations, if the AI had sufficient sensors to see the Sat's, it would again pick the weaker planet.

[Rollo]

Originally posted by Master Belisarius:
I go for the first option...
Sorry Rollo, but I'm with Mephisto and Alpha Kodiak: I think that sometimes the AI can do some "intelligent" actions, but really is not smart. In fact, I would prefer that the SE4 AI could cheat something and then, play better.

Nothing to feel sorry for . Actually I agree with you. All I said was that the AI can look at things that humans cannot. Whether or not it makes intelligent decisions is a totally different matter.

Rollo

[dogscoff]

nice to hear some stories of challenging AI behaviour. I have my own theory:

Having read our continual *****ing about how predictable the AI is, Aaron snuck an "occasionally do something unexpected completey at random" function into the AI code during a previous patch. Most of the time this will result in the AI doing something new and stupid, and we will disregard it as such, but every now and again, a random action will occur which looks like excellent strategy from the AI.

If there isn't a function like this, there should be one=-)

[geoschmo]

It wouldn't suprise me at all if the AI could see the contents in the cargo of our planets. I would think it's an unintentional thing, could be called a bug. But it might be a bug that we don't want fixed.

Consider some of the other bugs that have been fixed in the past. Populations used to get unhappy at the presence of cloaked enemy ships. And you used to be able to see the contents of an enemy planet yourself by pulling up their planets in the combat simulator. It's very possible that these were fixed only from the players perspective and not from that of the AI.

Geoschmo

[Q]

Excellent observations Gandalph. I always had the feeling the AI could "see" how strong the protection of a colony through weapon platforms and fighters was, but I never could prove it as you did.

[ August 01, 2002, 12:03: Message edited by: Q ]

[Atraikius]

Just to throw a little more to think of in this

I've seen a similar situation in several of my games while working on AI's.

Early in the game, I had a few frigates and a stack of sats defending my side of a warp point, with about a dozen lightly defended planets and a home world in that system and the adjacent one. I was attacked by a fleet of about 15 destroyers which I didn't have a chance of destroying with the forces I had available in the area. Resorting to cheesy tactics and using tactical combat, I attempted to damage or cripple as many of the enemy ships as possible, taking advantage of the AI's tendancy to retreat all damaged ships back to the nearest ship yard for repair. I was able to destroy 7 ships, and damage 6 others to some extent. However, instead of returning the ships for repair, the AI proceded to send the two undamaged ships into the adjacent system, and glassed the one breathable planet in the system other than my homeworld (assumed this was due to that being the only planet colonized other than my homeworld the Last time they had a ship in the system), while the 6 damaged ships (one of which only had 1 engine, 1 DUC 5, and 1 CSM 3) each proceeded to attack separate planets that they were capable of reaching durring the AI's move, glassing all 6. The next turn, the AI sent its three most damaged ships back for repair, while the other three proceeded to glass the remaining planet in the initial system, and two planets with minimal defenses that were within range in the adjacent system (bypassing the homeworld which was the closest planet to the warp point). Meanwhile, the two undamaged ships, instead of glassing any of the other planets in the system, chased down and destroyed the colony ship that I had sent out from my home world durring my turn, followed by a second colony ship the turn after, placing the two undamaged ships 4 systems from thier closest colony. The odd coincidence in this (which I have also seen with drones) is that in the strategy used by the enemy fleet, colony ships were one of the top target type priorities (use target type first was set to false).

My point in all of that is that I am wondering if the strategies used also have an effect on determining the AI's target selection outside of combat.

[jim]

Here is another example that suggests the AI benefits from info advantages:

1) The AI had a force of 45 ships in the square adjacent to a warp point that was a choke point into my empire,

2) I had a fleet of superior force - 40 ships but of much more combat strength than that of the AI - sitting on the warp point serving as a "cover force" for

3) Another 10 ship fleet on the warp point of 5 shipyard ships that were building defense bases and 5 other warships serving as close escort.

The three fleets sat there, without any combat, for several turns. I then selected "attack" for the cover force against the AI force. The result was that the AI force moved into the warp point square, destroyed the 10-ship fleet and was destroyed by my 40-ship fleet next.

Thus, the AI "knew" impossibly that I had ordered my cover force to attack and "impossibly" slipped through it to attack the weaker force.