You can see the stars on the map automatically. If you send your ship near it, it's not uncommon to be accidentally beelining for a map. I've sent colony ships exploring and then got a popup saying they've colonised.



Logic. They don't display the cheating behavior when this cheat is off, as posted by magnumaniac, and as I have seen myself. They'll arrive 2 or 3 turns late into the grid and realise I've just beaten them. However, once you reveal the map, they start cheating.

I'd say that's fairly decent proof that they're seeing the cheat.

EDIT: Yay, it is mumble!

I am not a dev, so yes, I may have been mistaken (rare, but possible when answering questions from memory!).

Actually, there is definite proof that if you use ctrl-u to remove the fog of war that the AI's can 'see' this as well. The proof is that the same turn that you use ctrl-u you will instantly be contacted by and greeted by all AI's (usually this is before you have universal translator). Clearly ctrl-u allows them to see you long before one of their ships stumbles on one of your ships or planets (or vice versa).

I had to think about that for a second, but you're right. If they couldn't see you, you'd be able to contact them, but they wouldn't contact you unless you contacted them first OR scouted your territory.

Maybe you are right that AIs can see the map as well when revealed by cheat code. Only MAYBE. I would not be so bold and say the proof is definite - it is possible to explain the contact by other means. However, I agree that the explanation that you gave is the simplest and therefore has the most force.

However, how would you explain the following? I did the test again. This time, however, I did not reveal the map right away. Instead I waited for 4 turns. On the fifth turn I revealed the map and guess what? Thalan colony ship is happily beelining towards nearest habitable planet and has made good progress already. Evidently, it started its move on turn one. The planet is too far to be within sensors and the flagship has moved to another direction. I did the test on "cakewalk" - so there should be no sensor bonus whatsoever.

I repeated the test three times and got the same result each time except once where the nearest habitable planet was too far away for Thalans to reach. In the occasion the colony ship stayed close to the home system.

How do you explain this behaviour if you think prior knowledge is ruled out?



The example you give does not necessarily rule out prior knowledge. It seems to me that AI knows the location of NEAREST habitable planet. If it happens to be PQ5, it will take it even if there is PQ26 just two moves away. Would it fit your experience or have you had instances where AI takes low quality planet that is further away? (Further away from the direction of origin of the colony ship)



I think it is other way around: it proved difficult to "teach" AI to search effectively. So they gave AI the knowledge of nearest anomaly. To balance things the same knowledge is avaiable to player via autosurvey. The same is true for mining ships.

As regards prior knowledge of resources then this has been confirmed by stardock staff member. To be precise: the person believed that AI has prior knowledge of the location of resoruces because of very old code.

Finally, I tend to believe that AI also has knowledge of the location of colonies (without exploring) for purposes of freighter destination. However, I have to test it more with Korx as an opponent to say with more certainty.

All AI's start with Stellar Cartography so they "know" that there's a star with planets. As I mentioned I send my colonies to explore based on the mini-map. I see no reason the AI can't do the same.


Assuming that the AI scouting is the same as auto-explore, I have to agree that it's pretty bad. As to the why anomalies are known I have no idea but it is knowledge available to both player and AI so it's certainly not an example of an unfair AI advantage.

Your supposition is that the AI operates with knowledge that's not available to the player or that the AI can do things that the player can't (i.e. colonize multiple planets with a single colony ships). This is a natural assumption to make when the AI seems to be able to do things that the player can't seem to do. I know I felt that way when I first started playing the game. Over time I have eventually been able to replicate every behaviour or ability exhibited by the AI and I've been able to exceed AI performance in any aspect of the game that's not directly related to known AI bonuses when playing at high levels.

Certainly one of the main points of this game is that the AI plays the same game as the human. I believe that is a predominately true statement. I suppose there could be a hole or two in the theory, for example, resource locations, but I really do doubt it.

I do not believe that is the case--AFAIK, the AI will build scouts and "manually" send them to stars it sees on its mini-map. It is simply unable to colonize a planet unless it has "seen" it, by whatever means (this may be a less hard rule above intelligent level).

I do recall that there was a bug at one point where the "get a planet I can see" function was returning planets the AI hadn't actually seen, but I'm pretty sure that was fixed (there was a discussion on the forum here about it).

I should have been more specific. In the test, I set habitable planets "rare", stars "abundant" and quadrant size "medium". As a result there were very few planets (about 5) that could be colonized without specific environmental technology. The total number of planets was much-much higher, I think close to 100 would be a good estimate. However, Thalan AI NEVER went for a planet that it could not colonize. For instance, it simply ignored promising clusters of nearby stars that had 4-5 (unhabitable) planets each. Instead, it went STRAIGHT for the nearest planet it could colonize provided it was within range. By straight I mean that I could see the colored line from the ship to the planet in question after revealing the map. OK, maybe the path line appeared at the moment when the map was revealed. However, revealing the map could not have moved the ship towards the planet.

In short, I find it very difficult to believe that AI just gets lucky by randomly choosing a right planet for colonization. Odds are way too low for that. As I wrote I did several tests. Please note also that on one map there was no habitable planet in Thalan range and in this case the colony ship stayed close to the home system.



You are absolutely right. It is very natural to assume that AI has some sort of advantage while in in fact it just effective in what it does. However, I am not able to explain satisfactorly the test result without assuming that (Thalan) AI knows where the first nearest habitable planet is. Frankly, I would be happy if somone could explain the test result that leaves intact the assumption that AI does not have the knowledge.

(BTW, it was not me who claimed to have seen the same colony ship colonizing three worlds. I have assumed many other things, though, that on closer inspection and with more experience have turned out to be very wrong assumptions.)

Just wanted to chime in and say I've seen the AI make some rather incredible guesses with regards to where to send their colony ships. It's strange to turtle up and make first contact with a given species on a gigantic map with some colony ships making beelines for the unclaimed toxic/aquatic/heavy grav/etc worlds in the middle of my empire they just gained access to the tech to colonize.

I'm not sure I'd go as far as to say that the AI is cheating, but if I see a colony ship rushing through my space it's generally a safe bet that I can draw a line from it's course to a habitable planet I missed. This can occur even if I'm boxed in along a wall and the AI has yet to penetrate my empire that far with ships.

Try testing with the humans. Or custom races without SC.

Also I'm still of the mindset that the AI gets the fog revealed when you do.

I don't play with cheats on and teleport my flagship around with ctrl-t instead of revealing the map if I'm testing a mod I made (usually no reason to need the whole map revealed.

Oh. Yes I guess your test does have merit. I guess you also mentioned that this was on cakewalk as well so sensor bonuses wouldn't be coming into play. It seems you have a valid complaint.

Just a clarification: the test result cannot be explained by access to stellar cartography or revealing the map by cheat code. The point is that Thalan ship accurately pinpoints the nearest habitable planet out of dozens of unhabitable planets and starts to move towards it before the map is revealed.

How does it work on lower difficulty levels ("tough" and below)? Does it mean that AI does not know the location of habitable planets before "seeing" them? Or does it mean that AI knows the location of habitable planets but is unable to colonize them without "seeing" them? Obviously, in the latter case the restriction is redundant: AI can send a colony ship straight to the location of a habitable planet, "see" it with colony ship and colonize it.



Yes, it was on "cakewalk".



Please, stardock, revive my faith into the AI by explaining the test result.  

The non-Terran AIs know where planets are via Stellar Cartography, but it does not know whether they are habitable or not until it has actually been near it with a ship and "seen" it. And it is flat out not permitted to colonize a planet it has never seen, AFAIK. If it is doing so, it might be a bug.

When trying to beat the AI to a planet I've uncovered and know the AI has a colony ship headed towards, I as a player can tell when it flips by seeing the AI's logo pop over it if they get there first. AI either can not tell this without having the planet in radar range, or just doesn't change their colony ship to somewhere else once it's on it's way. I think everyone can agree to those points.

The problem is just that the AI seems to have something similar to auto-mine and auto-survey for colony ships. This would also explain why the AI will frequently colonize a low PQ planet and ignore a very high one right next to it until later. Even very early in the game on a galaxy with lots of planets and stars but few habitable planets you can see the AI making incredible guesses on where to send it's colony ships that are very difficult to attribute to scouting given how early in th egame it is. You can see it again later in the game when the AI uncovers an extreme colonization tech and frequently sends colony ships directly towards the inhospitable planets they can colonize in the far back corner of your empire where they haven't yet sent a ship.

If this is happening however - do we stop it?

I'm of the mind we shut it off unless the AI is playing on any level where it gets bonuses over the human player.

This has been a lively discussion, so I thought I would run a test on DA myself. Cruelsader and Tetrasodium seem to be onto something, but I also got results consistant with what Magnumanic posted. It seems like the AI is sending it's colony ships straight for the nearest habitable planets. However, if you colonize the planet a half turn before the AI gets there, it will try to move the colony ship onto the planet and seems to get "confused" for a turn or two. I saw no difference in the AI colonization pattern (save for number of colonizers) between cakewalk and suicidal. I DID notice a difference if you gift the AI a colonizer on those two levels, the cakewalk AI never seemed to take control of my gifted ship, the suicidal AI used it to colonize after a few turns of it sitting idle.

If this is a bug, and if it would only be changed on lower difficulty levels, then it doesn't effect me much. To be fair, try running the test he described. The bigger question seems to me, is this a problem even if not working as originally intended?

Good luck all.

Fixing it so the AI had to guess or scout for planets or make hopeful guesses with colony ships more like players do would pretty much mean the early mad rush to colonize everything you can as fast as possible would change to allow slower growth/spread while building up the infrastructure. Do you take a third of the galaxy with empty planets and deal with the AI's much stronger position, or do you expand at a slower rate while keeping research and social production at decent levels too? It also means that some races to be set with an affinity for expanding quick and others for expanding logically. It probably also would mean that the AI might be more likely to take that PQ28 world instead of the PQ3 one next to it and that they simply divert to another unclaimed or promising uncolonized cluster of planets as soon as you colonize a world they were heading for .

I think it should certainly be fixed at lower difficulty levels. For instance, this thread was started by a player who apparently had problems with AI colony rush.

As regards higher levels then it depends on how it affects the overall efectiveness of AI. If the AI can scout effectively then it should be certainly fixed. As Tetrasodium pointed out it would allow more varied strategies in the beginning of the game. Also, it is lame that AI builds hordes of scout ships just to create the illusion that it is scouting.

BTW, what is the use of scouting currently? AI always seems to go for the nearest anomaly and asteroid field - no use for scouts. It also knows the location of resources and nearest habitable planets - no need for scouting unless scouting enables AI to make more intelligent choices, i.e PQ 26 planet instead of PQ5. AI also appreas to know the location of colonies for the purpose of freighter destination. This leaves what? Only contact with other civs?   )

Speaking of SC, seems that from 2178 to 2225 the humans didn't have enough time to map the galaxy, yet all it takes them is to research a single (and simple) tech...    Maybe this did make sense when you could only play the humans, but does it still? A tech for a single race to research?!

For what it is worth, I have a streamlined version of this test so you can see what is going on.

Set up: Medium map, scattered stars, rare habitable planets, common planets, abundant stars, very fast research, tech trading on. One opponent.

Starting out, just research up through space weapons and build a cargo hull with one particle beam and a bunch of engines. Now, go hunt down the enemy flagship and enemy scouts (if any). You can knock out all the enemy sensor craft in the first few turns if you do this.

Now, research through the sensor line and build about a dozen sensor range 15 cargo hulls, and space them across the galaxy so you can see everything.

Offer a peace treaty with your opponent.

Research up through the range line until you max it out. Now gift all of these techs to the AI (so that all uncolonized planets are in its range).

Don't colonize anything yourself, just sit and watch (and manage your money).

If you are on cakewalk, now is the time to hurry up and wait. You might have to grind through the turn button 200-300 times for the AI to do anything. This is the advantage to testing with the Thalans, they start with a colony ship, and their super hive ability seems to force even the fool AI to do.... something, anything. The disadvantage of testing with Thalans is that sometimes that first colonizer finds a world before you get to them to start watching.

BUT... you know the AI has not managed to reveal most of the map... you destroyed all the scouts. And now all uncolonized planets are in the AIs range. When it finally gets around to colonizing planets, those colonizers will go straight for an uncolonized planet, even if it is across the map. Colonizers will never "scout" a system with no habitable planets, they always go straight for uncolonized worlds.

This test works on any difficulty, and to save your sanity I do not recommend running too many cakewalk tests... the 20 minutes you sit there watching the AI do nothing you will never get back.

Hope that helps.

I've performed a test myself using means not available to the general public.

In my test case, an AI was chosen for observation (Korath). I immediately dismantled its flagship and miner at the game's start, and every non-colony ship it built afterwards before they could escape the home system and see other planets.

The AI built four colony ships in the early turns, most likely intended for use as soon as it found suitable planets. However, after more than 60 turns, having had all attempts by the AI to scout stymied (and the fighters it built dismantled too, for good measure), it had not used any of these and owned only its homeworld and class 4.

The game was set to Tough difficulty.

Has the behavior noted in your tests been observed with any specific AIs? I can test them as well, in case the colonization code is not common to all AIs (I don't know if it is or not). But based on the Korath test, there seems to be nothing amiss.

Yes, I tested with Thalans.

And when testing with Thalans may I suggest that
a) Do the same test as with Korath; if that gives no abnormal behaviour
Just observe Thalan intial colony ship (do not destroy any other ship); if still no abnormal behaviour
c) Try the test as lined out by Wyndstar