Does the AI have preknowledge of the map?

[Kolyana]

I don't believe the AI has knowledge of unrevealed resources under most circumstances. The days of AIs sending settlers halfway around the map to claim unseen resources are over.

This issn't really open to debate, to be honest: THe AI knows of hidden resources. Period. I tested it half a dozen times yesterday under different scenarios, and he went for the resource every time (all other factors being equal).

DISTANCE to that resource plays a part ... so if a resource is 20 squares away and there is a viable city location somewhat closer, the AI will head ot the closer path, but if both points are the same and one has *any* resource, the AI will head to it.

THIS INCLUDES RESOURCES HIDDEN IN FOG OF WAR.

THIS INCLUDES RESOURCES HIDDEN BY LACK OF TECHNOLOGY.

There is no "if's", "but's", ":why's" and "maybe's" about this ... it's fact.

Anyway, I don't think this is a huge issue in practice. Given that the AIs do not any longer send settlers willy-nilly all over in the map in pursuit of resources they can't see, the only advantage they're going to get is by putting cities they were going to found in the area anyway into slightly better spots. And human players can do the same, now that the flaw is known.

I disagree.

What else does the AI know?

Your city placement?

Units inside your city?

Your tech level?

What you're researching?

What you're building?

Where all the hidden oil reserves are (oh, wait, he does know that one ...)

 

[Carewolf]

I could test that. What do you want me to do?

Well, yes please. I don't have any experience with the world editor yet.

 

[The Condor]

I will. Just please tell me what to do (I am only good with very clear and precise directions).

 

[Psyringe]

Does the AI follow exactly the same blue circle advice that the hints give? If so they have relatively little info which isn't available to the human player.

I don't see an easy way to test this. The AI *seems* to do so. However, it does get more information as the player. The player only sees "recommended (circle)" or "not recommended (no circle)". The AI seems to get quantitative information as well. In the test setup, the human player always sees two blue circles at the end of the cross arms, with no one to tell which one is better. The AI always settles in the blue circles adjacent to the "hidden" resource, so it knows which one is better.

Calling Firaxis a bunch of liars is just silly. The vast majority of the AI's map knowledge has been removed, as is very clear from a couple of games. The undiscovered resources being left in is a fairly minor issue which seems to have returned by mistake. This issue has been vastly improved since Civ 3, as the AI can no longer beeline to the best sites on the map from turn 1.

I agree that namecalling doesn't help. Still I'd like to clear this issue up. Btw, do you have links to savegames which prove the vast AI improvements? I experienced some improvements myself (e.g. the naval invasions), but I'd prefer having testable proof over anectdotes from player experience (which usually have unchecked variables in them).

Complaints that the test should have been done by Firaxis seem a little odd however. The AI was programmed not to have knowledge of the map, and if they forgot about undiscovered resources being in the city placement evaluation there would be no reason to assume there was a problem. I doubt the beta testers would have noticed because, simple as the tests in this thread are, there would be no real reason to run them. I doubt they set up a special map to test each rule and element of the game individually. It is far from obvious that the AI still has this ability in a conventional game, as is evident from the fact that it has taken more than two months for it to be proved the issue exists.

My point was that the pre-knowedge of resources is/was an important issue to many fans. Of course neither the designers nor the testers can design specific testbeds for any rule they implement. I wouldn't expect that. However, for issues that are regarded as important, and easy to test, I'd expect them to be tested thoroughly, because we all know that mistakes *can* happen, even when you're sure that you made everything right.

Anyways, debating who's to blame probably won't lead to anything constructive. I just wanted to explain why the argument "it wasn't a lie, it's just a bug" doesn't make *everything* right.

Btw, I agree that currently we can't tell how much real games are influenced by the issue we found. To determine this, we'd have to approximate the test situation to real game situations, i.e. experiment with varying landscapes. We *may* find out that a difference in landscape always overides the effect of resource pre-knowledge, in which case the relevance of our findings for real games would be pretty small. But we can't tell yet.

 

[MrCynical]

What do you mean by resources hidden by fog of war? Fog of war doesn't hide resources. If you mean unexplored terrain, then it hasn't it been shown that the AI doesn't know where resources are in this case? Yes, the AI knows where resources from undiscovered techs are, as this thread has proved, but I don't think anyone is denying that.

I would be interested to see some tests on some of your later suggestions though, particularly in the case of what you're researching . Some of the AI's research decisions seem a little suspicious. For the city placement I think the answer is probably not, since the AI can't see into unexplored terrain. Since the human player can see loads of indicators of the AI's tech level, it's probably only fair they have info on the human player's tech (and I'm not certain they do). Would also be interesting to see if they know what you're building, but again I'm unconvinced on this. I've seen AI's miss modern wonders by one turn when a production boost was possible. Surely they wouldn't do this if they knew what you were building? I get the impression they do have some info on what units are in your cities, even when hidden by fog of war (and outside them as well) since the AI seems to home in on undefended units (nowhere near as badly as in Civ 3 though). Would be nice to see it tested though. Presumably set up a map where you're at war with an AI, and stick a unit half way between two identical cities hidden by fog of war. Have one defended and one not, and see which it goes for. In fact I might try that...

EDIT: In reply to Psyringe, the improvements I referred to were the fact the AI no longer has knowledge of unexplored terrain which I think this test proved? Also, if the AI does have knowledge of undefended cities it no longer beelines for them in the ludicrous way they did in Civ 3. I haven't done a specific test, but I have played a game where my frontline cities were heavily defended, but the next row were completely empty. For the entire 50 turns of the war the AI tried to take the frontline cities, whereas the Civ 3 AI would have just marched past them to the undefended cities hidden in the fog of war. These two issues were a major problem in Civ 3 that have now been fixed. There is still some work to be done, but it's a big improvement.

 

[Kolyana]

Psyringe, don't forget that this whole thing kicked off because of a real-world situation in which the AI grabbed an oil reserve well before it could be seen and the player suspected something was afoot; so this situation DEFINTIVELY effects gameplay.

As for the whole "blue circle" thing ... guys, I can see where you're coming form, but you need to broaden your minds a little. The blue circle (for the player) only ever appears within a few squares of your units position and you only get a couple of them ... best 2 locations within line of sight.

The AI doesn't get circles ... it uses an algorithm that is almost certainly the same that derives the circles you see, but it sees the indididual scores per square, without a doubt. In other words, it sees the mechanism used to determine where those blue circles go, but it gets to see these for all squares.

MORE IMPORTANTLY, it gets to see them in it's own fog of war ... it gets to see recommended spots that are out of range.

Granted, the AI may not be able to map a path to your location ... and this may be the Fog of War Firaxis is talking about, but it does know the numerical score for each square on the map and will settle according to this.

And for those that think that this has no gameplay effect, I would argue that the first time you play the EARTH map and the AI knows the best spots in the new World to settle from turn 1, there is a cludgey situation here that just doesn't fly.

And this applies to every map.

 

[Kolyana]

What do you mean by resources hidden by fog of war? Fog of war doesn't hide resources. If you mean unexplored terrain, then it hasn't it been shown that the AI doesn't know where resources are in this case? Yes, the AI knows where resources from undiscovered techs are, as this thread has proved, but I don't think anyone is denying that.

I would be interested to see some tests on some of your later suggestions though, particularly in the case of what you're researching . Some of the AI's research decisions seem a little suspicious. For the city placement I think the answer is probably not, since the AI can't see into unexplored terrain. Since the human player can see loads of indicators of the AI's tech level, it's probably only fair they have info on the human player's tech (and I'm not certain they do). Would also be interesting to see if they know what you're building, but again I'm unconvinced on this. I've seen AI's miss modern wonders by one turn when a production boost was possible. Surely they wouldn't do this if they knew what you were building? I get the impression they do have some info on what units are in your cities, even when hidden by fog of war (and outside them as well) since the AI seems to home in on undefended units (nowhere near as badly as in Civ 3 though). Would be nice to see it tested though. Presumably set up a map where you're at war with an AI, and stick a unit half way between two identical cities hidden by fog of war. Have one defended and one not, and see which it goes for. In fact I might try that...

Already proven.

I built the regular test map and placed DYE in a Fog of War (unexplored terrain) for the AI (I went to the map builder and made sure that they had no visibility to it). The AI went straight for it.

The AI knwos resources are there. PERIOD. It doesn't matter if they are hidden by lack of tech or unexplored terrain, I tested BOTH. I tested a multitude of scenarios and resources, some hidden by distance, some by fog of war, some by tech ... all were known to the AI.

For these tests I always made sure that the terrain was unexplored for the AI and watched on a move-by-move basis as he would go straight for the resource with no detour; moving into unknown terrain, but a known bonus.

 

[CivScientist]

I have noticed in past games of mine that the AI will go out of its way to conquer my cities that have important resources (such as oil) long before anyone know those resources are there. They'll even seemingly defy logic to take those cities.

One game I was playing, 3 of the most aggressive AIs went after this dinky yet reasonably protected desert city. The strange thing was that these civs sent their armies past much nicer cities with equal protection. It turned out there was oil with that city's radius.

Now, naturally, this single game is not a good test by far. However, I have noticed I can almost predict which of my cities will be important strategically later in the game (due to resources) by which ones are sought after by enemy civs. I would be interested in a variation of the original experiment where, instead of having the AI send a settler, he sends a powerful (but small) army to overtake one of your 4 cities on the end of each peninsula. Only one city has the oil.

The main reason I bring this up, however, is to point out that this extra information is significant. Although the computer only knows a little more than the blue circled tiles the human player can see, that little bit does give a significant and unfair advantage (at least in the eyes of the players) to the AI. More, a savvy player can exploit this like I kind of have. You want to know if your border cities in the tundra or desert are worth anything? Just see where the computer attacks and hold onto that city like the dickens.

 

[The Condor]

Sorry people if I got a little crazy with the name calling.

 

[Kolyana]

I think they mean that it is hidden by techs that haven't been researched yet. Also, sorry people if I got a little crazy with the name calling.

Actually Condor, both ... resources in unexplored terrain and resources hidden by tech; the AI knows about them both.

I would say that it is using (as was pointed out) the algorithm for city placement and I would suspect that each square is ranked or scored ... it's merely following a better score for that square, but the point is this effectively means that the AI knows the best squares regardless of explroed/unexplored terrain right from the get go.

 

[onedreamer]

I don't believe the AI has knowledge of unrevealed resources under most circumstances. The days of AIs sending settlers halfway around the map to claim unseen resources are over. In one of my first games I wound up with a whole swath of tundra and ice to my south that nobody bothered to settle the whole game. When I finally got to see oil, lo and behold, there were two sources down there. The first AI to show up by the oil with a settler was my nearest neighbor, about five turns after I'd sold him the tech to see it.

I wouldn't expect such a clueless post from someone of GOTM staff. First off it's been proven beyond doubts in this thread that the AI has pre-knowledge of resource placements, in case there was a need to prove it, which I didn't think, but it seems now clear to me that even with a game certain people will never accept reality unless it's slammed on their face.
If all you can come up with to the debate is "but the AI has this pre-knowledge only when settling cities", well my obvious answer is when else do you ever need to know where are resource placed before you can see them ? The answer is only "when settling".
Second, it's evident that the main difference between Civ3 settling and Civ4 settling is that now careless and redundant settling is penalized. The AI thankfully doesn't seem so stupid to settle tundra with oil if it can't use it yet and if it has already a big enough empire to maintain.
I want to close stating that personal impressions do not make statistics. If you want to debate against proofs, bring proofs along and not your impressions.

 

[The Condor]

Hi, uh please use whatshisface's first test and not mine cuz it had an error in the testing so it is unreliable on the second page (settlernobletest). The rest are reliable however. Just wanted to get that across.

 

[onedreamer]

Complaints that the test should have been done by Firaxis seem a little odd however. The AI was programmed not to have knowledge of the map, and if they forgot about undiscovered resources being in the city placement evaluation there would be no reason to assume there was a problem. I doubt the beta testers would have noticed because, simple as the tests in this thread are, there would be no real reason to run them.

Dear MrCynical, you evidently have no clues whatsoever on what a (real) beta test is. Do a search on google, read some articles somewhere, ask someone compentent, and then come back to discuss about something you have at least a little knowledge of.

 

[Carewolf]

I will. Just please tell me what to do (I am only good with very clear and precise directions).

I think it can be tested like this:

1: Place AI destroyer in the ocean on a random map and see where it moves.
2: Now place Human transport with settler aboard 4 squares away (outside view)
3: See if the destroyer goes for it.

For consistency it should be tested with a barbarian unit as well; and with a a settler and AI/Barbarian units on land as well.

 

[The Condor]

O.K. Thank you for your clarity. I get the first part but what do you mean with "and with a settler and AI/barbarian units on land as well"?

Do you also want me to place a settler randomly on a map and see if the AI goes for it cuz I can do that then. Just say yes and I will do it or no but change your statement to a more logical statement (no offense) so I can do it o.k.?

 

[jayeffaar]

1. If this is a bug (and not planned behavior), then - and I do this for the first time, I have always defended the testers in many previous posts - I wonder how it could slip through. The claim that the AI doesn't have pre-knowledge of resources is and was an important issue for many people, and Firaxis knew that. As you said, they have been quite vehement about map pre-knowledge being gone, and I'd expect an issue with so much importance to many fans to be tested thoroughly.

To be fair to the testers, this isn't something that seemed likely to happen accidentally (even if in the end, it may have) and if the devs told the testers that the AI wasn't doing that anymore, I can't think of a reason for the testers to specifically design tests for that. They had no reason to actively look out for that behaviour.

 

[Kolyana]

You know, I remember a game not so long ago in which the AI settled on one side of a sea-strait that was much less hospitable than the other. I remember thinking "WHUH? Why did he settle in the desert when such lush grass was on the other side?"

I bet it was a hidden resource.

 

[Psyringe]

EDIT: In reply to Psyringe, the improvements I referred to were the fact the AI no longer has knowledge of unexplored terrain which I think this test proved? Also, if the AI does have knowledge of undefended cities it no longer beelines for them in the ludicrous way they did in Civ 3. I haven't done a specific test, but I have played a game where my frontline cities were heavily defended, but the next row were completely empty. For the entire 50 turns of the war the AI tried to take the frontline cities, whereas the Civ 3 AI would have just marched past them to the undefended cities hidden in the fog of war. These two issues were a major problem in Civ 3 that have now been fixed. There is still some work to be done, but it's a big improvement.

Mh, the "unexlored terrain" issue is a little complicated. Condor's first test (the one with the AI army and four cities around it, one of them undefended) shows that the AI cannot see unexplored terrain under these circumstances. However, Kolyana's tests (and my own tests too) show that the AI can see unexplored terrain when deciding where to settle. Btw, I think this can be seen as strengthening the hypothesis that the oberved behaviour is a bug (rather than deliberately programmed), caused by the AI accessing map evaluation routines which weren't made for it, although it's not a proof of course.

 

[MrCynical]

@Onedreamer: No, I'm never done beta testing and have only read one or two articles on the subject. Have you ever done beta testing? If so, then when presented with Civ 4 for that purpose, would you at any point have decided to do the the various tests that have been carried out in this thread? Yes, my knowledge of the subject is limited, but the basic principle seems to be to try and break anything possible. I'm not sure this extends to creating maps for every little thing. This thread only appeared after a month or so of suspicions from the anecdotal evidence of the whole forum. As a result a specific issue which seemed dubious was focused on, and proved to exist. If you are a beta tester and feel you would have carried out these specific tests, then my apologies and feel free to enlighten me on good beta testing practice. If you are not a beta tester, then at least enlighten me on your clearly superior knowledge of the subject, and maybe tell me the source of it, so I can learn more.

@Kolyana: In some of the tests you seemed to be saying that the AI had a longer field of view into unexplored land. Is there then a distance at which resources are invisible even for the AI in unexplored terrain? If so it might account for why it isn't as noticeable as in Civ 3, or is it simply that the AI isn't willing to hike across the entire map any more even if it does know a resource is there? Also do they know where better terrain is, as well as specific resources? Should be an easy test with one arm of the X having some grassland at the end, and the others with desert. I couldn't find one in this thread though that didn't have resources involved as well.

 

[The Condor]

Hi, Carewolf. I tried the test by putting on a island with eight paths that span out five squares each one settler and gave the AI a warrior (and myself a spy to watch it) and the AI did not instantly go for the settler (it has still been bumbling around for fifteenish turns).

Also, I tried the test where I put a settler in a transport five spaces to the right of a enemy destroyer and it took the AI three turns to find and destroy (four turns total cuz I didn't declare war on the first turn) it (actually it went right past it twice during war). Also I would like to note all my tests are done on noble.

So in conclusion the AI can't see every unit while warring in the sea nor on land.

Here is the save:


Also I got some pics:

Here is the test beginning with the destroyer. It ended three turns later with the destroyer blowing up the transport.

Here is the test with the warrior and the settler. It is still bumbling around.