Different playstyles and how game mechanics & exploits affect your preferred playstyle and enjoyment

[Lanzelot]

I think, "full map knowledge" can be taken as a fact: the Civ3 lead programmer, Soren Johnson, confirmed it in a public lecture he gave about game development:

(And he was the one who wrote that code, so he should know... )

The open question is the one about the path finding algorithm. Can the AI send a unit/settler to some point on the map it knows exists, before it has charted a path to that point?
I can't tell for sure, but I think it can, for two reasons:

• From a programming point of view, it would be very difficult indeed, to prevent the AI of taking advantage of its full map knowledge. We know, that the AI has full map knowledge. So in order to prevent the AI from just sending a settler to that uranium island it knows exists "somewhere in the dark", the game would need to keep two maps in memory for each AI: the one we know about (with the complete knowledge), and another one that keeps track of what the AI has already charted at any given point in time. But as I understood Soren in the above talk, one of the points of giving the AI full map knowledge in the first place was to avoid the hassle of having to keep track of which tiles each AI has "discovered", and which not...
• At one point Soren, in the talk I linked above, says something like "the AI explores only for show, to make its play appear reasonable towards the human player. It doesn't need to, because it already knows everything".

So I think we need to distinguish two different mechanisms:

1. Sending units (in particular settlers) to "yet unknown" territory. As I outlined above, I think the AI is capable of doing this.
2. Establishing trade routes. (This is what Bamspeedy talks above.) This is a different game mechanic, and as Bamspeedy outlines, for this the AI needs to have a complete route (via roads or coastal tiles) from their capital to ours. Here the game mechanic is the same for both, the AI and the human player.

 

[justanick]

So in order to prevent the AI from just sending a settler to that uranium island it knows exists "somewhere in the dark", the game would need to keep two maps in memory for each AI: the one we know about (with the complete knowledge), and another one that keeps track of what the AI has already charted at any given point in time.

Since this map already does exist this does not seem that big of a hassle.

 

[Fergei]

Start a game in debug mode and enable full map visibility. The Expansionist AI will disband its scouts immediately.

Start a game as normal. The Expansionist AI will (if I remember right) use its scouts to explore.

This may make you think the AI explores only based on the map visible to it. But the programmer's insight trumps this, and presumably the scout disbandment is simply a cosmetic mechanism to ensure the AI don't make scouts and explorers after the world is fully revealed.

Exploration with non-Scout units (excluding settlers) would therefore only be legit in terms of them being coded to locate goody huts and encounter other Civs, so I don't think that technically counts as exploration "for show". Unless the AI spanning the continent taking goody huts and bumping into other Civs is completely coincidental, which I can't rule out - but seems unlikely. The game does a very good job of simulating intelligence and variation.

 

[r16]

for all its worth ı have read it somewhere on this forum that the Al will supposedly stop building ships whsn it has the whole map . Yet of course in Age of lmperialism they are apparently building new things as tech becomes available .

 

[Fergei]

Going back to the excellent presentation fro Soren Johnson above (where he speaks theory for the first 25mins then is more Civ specific) I think there is a clear insight into whether the AI will take your 'go-to' moves into account in order to get a settler to a location before your settler gets there via 'go to'. I suspect Soren will not have permitted this.

In his presentation he makes it clear that one of the things he didn't like in the first Civ, was that the AI would base calculations on whether they related to an AI or a human. He also states in the Civs he did the AI for (III & IV) his principle was for the AI to not necessarily take into account whether its calculation involved a human or rival AI (although trading techs appears to have been an exception to that rule). He also talks about the memory demands of AI unit movements and that he kept it fairly simple. So I suspect he wouldn't have added an extra lair of AI calculations to differentiate if a human or AI is moving the units.

It is not a definitive answer but I suspect that the AI are not coded to beat your 'go to' settler in a race to a new city spot. I think it will more likely be that if you as a human consider a location is a good spot for a new city, the AI will be sufficiently well coded to also consider that area as a priority and race to it (perhaps instead of building a rubbish city in a more nearby desert). With that said, the game is well enough coded to make me paranoid enough to make my settler 'go to' a space near my intended city spot, rather than the intended spot itself. At least that way I won't be turning the air blue when the AI 'steals' the spot.

 

[Buttercup]

The thing about what the AI can and cannot see is it's quite a complicated problem. But because it's complicated I don't think we should assume it's deviating from it's intended parameters of primarily providing the player an experience first and foremost.

For example, there must be some calculation of what the AI has personally 'discovered' otherwise it wouldn't know how to value maps that you sell to it. And this is a very important aspect of the game from a player affecting perspective. If you reveal so much map before the AI then they will give you X amount of gold for that trade. Once you have sold them the entire map they no longer offer you much if any gold, or whatever you're trading. So there MUST be a separate track of individual map reveal.

However, at the same time, the AI already knows all the map.

So I would suggest the code separates different factors as being revealed and non-revealed solely for the purposes of individual game mechanics, not as a one-size-fits-all approach. It will record a separate map for the purposes of map trading, for example, but not for object hunting.

Here's some more evidential screenshots:

Picture 1:

Before my capital has even had it's first expansion, AKA within the first 10 turns, an Arabian Scout arrives at MY border. Not only that, but the first goody hut (that I know about) it finds NEAR ME provides it with a Settler. All in what can ONLY be described as a BEE-LINE manner of approach:



And just to confirm that the Arabian city was founded that quickly, we can see that my first boat out, I always start with a boat on such maps) confirms that this goody hut MUST have been popped within the first 20 turns, the Scout was coming back from the desert above the Arabian city when my boat went round there:



And just to confirm that the AI had INNUMERABLE scouting and exploration options at the start of the game, we can see that this island is indeed quite big and actually has FOUR civs on it including myself, meaning that the statistical likelihood that the AI was RANDOMLY exploring is VASTLY diminished. The nature of the exploration of the Scout and the end result of that exploration and the way it specifically affected the human player puts the emphasis of any RNG squarely into the UNLIKELY bucket of statistical improbability:



Which would confirm what I postulated above about the AI having different maps for different game mechanics criteria.

 

[jarred!]

You can investigate Medina once you have Writing to find the exact date it was founded. Do you know if there was there another hut in the desert on your peninsula the Arabian scout might have popped?

Edit, and just to be clear, you believe every AI is beelining to you every game because they know where the human spawned?

 

[Buttercup]

And there you are. Gotcha.

I know exactly when Medina was founded because I popped a hut when I settled my capital, and it gave me maps of the region.

Now try harder to look for reasons not to look at the evidence, huh?

 

[Buttercup]

Edit, and just to be clear, you believe every AI is beelining to you every game because they know where the human spawned?

And regarding this nugget of joy specifically, I stated exactly what my position was in the post you quoted, none of which contains any of the gibberish you write here.

I have no doubt that a civ on a far away island with no water access to me and prior to learning boats is most definitely NOT bee-lining me, that would be ABSURD.

For heaven's sake, if you can't comprehend even the most basic concepts what's the point in any of us ever showing anything?

No, not EVERY ai is bee-lining me LMAO.

Arabia, however, IS. Because that's the one that CAN. Duh.

 

[Fergei]

And regarding this nugget of joy specifically, I stated exactly what my position was in the post you quoted, none of which contains any of the gibberish you write here.

I have no doubt that a civ on a far away island with no water access to me and prior to learning boats is most definitely NOT bee-lining me, that would be ABSURD.

For heaven's sake, if you can't comprehend even the most basic concepts what's the point in any of us ever showing anything?

No, not EVERY ai is bee-lining me LMAO.

Arabia, however, IS. Because that's the one that CAN. Duh.

You are postulating things that are easy to test. Create a map and save it (including fixed starting positions for Civ numbers) before you play. If you spot something in the early game that raises your suspicions, restart the map with the same Civs and starting locations, but substitute yourself out for an AI player.

If you repeat the early game several times, in this instance I'd be very confident the AI scout would behave in exactly the same manner even if your own Civ was AI controlled. There is no indication (other than with tech trading) that the guy who wrote the code went out of his way to code the AI to care whether rival Civs and units were AI or human controlled.

And Buttercup, your above post falls very much into the category of what many a reasonable individual may objectively classify as 'vitriol'. That is:

"bitter criticism or malice"

 

[Buttercup]

Yes, they are easy to test, you sit down and play the game.

Yes, I have repeated the early game hundreds of times, that's why I postulate what I postulate.

If you haven't noticed I've already presented tons of screenies.

It's funny how 'vitriol' just happens to be "whatever I disagree with", huh?

Perhaps you'd like to explain how RNG works for us, so that we can understand the difference between RNG and BEE-LINING. Perhaps you could do the math and tell us how many times the AI needs to roll a 1/8 every time it decides to move its Scout in order to walk in a straight line for 6 turns and so provide the odds of the AI Scout ever walking in a straight line for 6 turns, you know, just so we can understand how much RNG was involved in the AI Scout bee-lining the human player RANDOMLY.

 

[Fergei]

With respect, I think youd benefit from doing some research into 'scientific rigour' before drawing conclusions that lead you to condemn the conclusions of others. Playing the game repeatedly as a player without making direct like for like comparison with the exact same scenario that is AI only makes it essentially impossible to draw firm conclusions about whether or not the AI treats players different from fellow AI. This is in part because the game is intentionally so masterful at concealing that because the game designer is on the record as saying he felt it was essential players didn't feel cheated like they did in Civ 1 when it was hard coded that all AI Civs will become hostile to the human if the human is ahead at 1900AD.

 

[Buttercup]

With respect, your post contributes zero to the debate other than an attempt at character assassination.

Since you have done zero research yourself, why do you consider yourself an authority on which 'other people' to 'believe'?

Herd instinct?

 

[Lanzelot]

Please remain civil all of you, or I'll have to put on the "moderator hat"... (Which I try to avoid as much as possible.

Let's stick to the topic. Which is a very interesting one indeed. And it shows that as long as we can't have a definite proof of the AI behaving or not behaving in a certain way (like a peek at the source code that codes that behavior), different people will interpret the evidence we can observe, in a different way.

For example, Buttercup, let's take your "Picture 1" from post #106 above. You say it proofs that the Arabian scout beelined towards you and the goody hut. For me it looks more like it chose a quite reasonable path (the isthmus at location "3" doesn't give the scout much of a choice, if it doesn't want to turn around back into Arabian territory, which is already "known" to the AI, so why should it go back?) And then it just stumbled into you, turned around (6-to-7) and then found the goody hut.

If I would have known about the goody hut and that there's a settler waiting for me, I would have chosen a different path, which would arrive at the goody hut two turns faster:

If it were really "top priority" for the AI to find that hut and pop the settler, this is what I would have expected in terms of a beeline. The red line on the other hand looks more like a "half-random" path mixed with some "reasonable choices" that can be expected from the AI:

• scout decides to go south (the map looks like a narrow stretch of land, so there's probably a 50-50 chance as to whether the AI goes south (towards you) or north (towards the Aztecs). East and west appear to be blocked by water.)
• it hits the isthmus and decides to follow it (instead of turning back into already known territory, which would not be reasonable)
• it runs into Amsterdam and decides to turn around (instead of trespassing into Dutch territory. In the early phase, when relations are still good, I have often observed that the AI refrains from stepping into my territory, especially with a scout that doesn't have combat value. And on turn 6, Amsterdam's borders had not yet expanded, so at "red 6" the scout was directly at your border, not yet inside.)
• finally, when reaching "red 8", it sees the hut and goes towards it

The above explains the scout's route even better than the beeline theory, and by "Occam's Razor", I rather believe that one than the beeline theory.

 

[vmxa]

"Before my capital has even had it's first expansion, AKA within the first 10 turns, an Arabian Scout arrives at MY border. Not only that, but the first goody hut (that I know about) it finds NEAR ME provides it with a Settler. All in what can ONLY be described as a BEE-LINE manner of approach:"

This is the second time I noticed you ascribe an event as it can only be interpreted in one way. Though I would say that is not self evident in this case.

You may be right, but you do not have data at this point to be sure. I have played on huge maps years ago at the lowest level to test out something, that I do not recall any longer. One of things I had happened and I have seen more than once, is that I could find a hut pretty close to an an AI town. They had not even popped, by the time I got to the other side of the land. I have seen huts not popped a number of times.

I cannot say that proves anything. Well I think I can say it proves the AI has issues, lots of them.

That I am still waiting to see something that is irrefutable is in no way meant to be an attack. I am actually fine with it, if you can come up with the smoking gun. Then the issue can be put to bed.

Anyway to your test, I would suggest the first 6 steps where about the only reasonable path that could be chosen.

Edit:
I guess I should have read all the other post and seen that Lanzelot has said it betterf than I.

 

[Buttercup]

For example, Buttercup, let's take your "Picture 1" from post #106 above. You say it proofs that the Arabian scout beelined towards you and the goody hut. For me it looks more like it chose a quite reasonable path (the isthmus at location "3" doesn't give the scout much of a choice, if it doesn't want to turn around back into Arabian territory, which is already "known" to the AI, so why should it go back?) And then it just stumbled into you, turned around (6-to-7) and then found the goody hut.

In order to get to the point of being able to be "stumbled upon" you already have to imply a whole raft of factors that you don't address:

1. How does the AI decide what is a "reasonable" path. The AI doesn't 'reason' in human sense. You use the phrase reasonable because you could imagine yourself doing the same, but the AI would not have that kind of reasoning, only programming. You have said yourself in your previous post that the AI is programmed to act in a certain way in order to present to the player the illusion that scouting is taking place. So:

a) The AI presents itself to the human player.

b) The AI doesn't move randomly. If it moved randomly then it couldn't walk in a straight line, the odds of that happening via RNG are remote. This suggests the AI uses a 'go to' order when it moves its Scout rather than making an individual new RNG each time it moves a square. If it uses a go-to order, it is highly likely it is using map knowledge to do so, otherwise any go-to order would only be 2 squares long, again, massively shortening the odds of it ever walking in a straight line.

2. Scouting has three main functions - Revealing land, meeting other players, opening goody huts. It is remarkable that the civ AI does not care about revealing land around its own capital before charging off into the sunset.

If I would have known about the goody hut and that there's a settler waiting for me, I would have chosen a different path, which would arrive at the goody hut two turns faster:

What might be contained within the goody hut is quite possibly a distraction here. It's certainly relatively irrelevant to the discussion of map knowledge.

It's more a matter of priority. Does the (nearby) AI prioritise meeting other players and if it does so does it do it as I postulate in that it REALLY LIKES finding the human player. I can't imagine the AI behaving like yourself, I believe the most common assumption about the AI in all other discussions is that it invariably DOES NOT act in a way that a human would, so why would you use "what I would have done if XYZ" as a means to be blind to what's being shown?

The whole goody hut mention could well be a distraction, perhaps I shouldn't have mentioned it? Perhaps goody huts operate on a "when within 2 squares" parameter and it has no pre-designed objective to grab them.

But then again, maybe the program is nuanced here again. Perhaps the objective is to insight the human player into an 'action choice'. Hey there human player, if you don't kill me I'm going to go and grab all your goody huts haha. That is certainly an option that makes sense from a developers perspective in ensuring lots of early game human drama.

If it were really "top priority" for the AI to find that hut and pop the settler, this is what I would have expected in terms of a beeline. The red line on the other hand looks more like a "half-random" path mixed with some "reasonable choices" that can be expected from the AI:

• scout decides to go south (the map looks like a narrow stretch of land, so there's probably a 50-50 chance as to whether the AI goes south (towards you) or north (towards the Aztecs). East and west appear to be blocked by water.)
• it hits the isthmus and decides to follow it (instead of turning back into already known territory, which would not be reasonable)
• it runs into Amsterdam and decides to turn around (instead of trespassing into Dutch territory. In the early phase, when relations are still good, I have often observed that the AI refrains from stepping into my territory, especially with a scout that doesn't have combat value. And on turn 6, Amsterdam's borders had not yet expanded, so at "red 6" the scout was directly at your border, not yet inside.)
• finally, when reaching "red 8", it sees the hut and goes towards it

The above explains the scout's route even better than the beeline theory, and by "Occam's Razor", I rather believe that one than the beeline theory.

I didn't imply the top priority was popping the hut, that is something that responders have invented. The post was quite clearly to show evidence of how the AI moves is not RNG and that it prioritises 'activity' with the human player.

It is remarkable that, for this topic, you feel that the AI must be acting 'reasonably' when for almost every other topic regarding the AI, the common consensus is that it does not act reasonably.

North would take it too the Ottomans, not overly far away. the width of the island would not be known to any players at this stage of the game. Even taking one or two turns to establish the extent of the width of the island would radically alter the scenario, it is not 50/50, not by a long shot, that is a deliberately belittling use of math, lol.

I'm not sure why you suggest the area around the Arabian capital is 'already known territory', seems like a really bizarre leap of logic. Surely a circular route around its own capital's borders would be the ONLY way for it to get it's own territory to the sate of being 'already known'.

Before the capital expands the Settler is not restricted in any way by borders, from a 'reasonability' perspective, the logical next move would be onto the mountain, to reveal more map for your buck. But then we could use this screenshot as evidence the AI doesn't care about prioritising map reveals, yes?

 

[vorlon_mi]

Consider the oft-seen example of how the AI moves its ships when conducting a naval invasion -- one step at a time. The whole stack, escorts and transport ship, moves one tile and then the next and then the next. I believe that the AI pathfinding algorithm is stepwise, not a long go-to.
In this case, where we seek to explain the path of the AI scout, its path-finding algorithm looks at terrain movement costs. Moving south follows the terrain onto flat land; moving north towards the Aztecs would go through marsh and jungle. That's how I interpret the "reasonable decision" comments. The AI prioritizes movement towards flat land when exploring. Rarely have I seen an AI unit double back on itself *while exploring*. The AI units turn around when threatened, but they don't take a circular route around their own territory. The Ai units exploring go outward like the spokes of a wheel, until they reach a barrier or are attacked.

We're left with incomplete data about the *goal* of this exploration. Buttercup asserts that the goal is to interact with the human player. I feel that goal is secondary, based on the tone of the comments in the Soren video regarding his goals for having fewer parts of the AI logic known whether the player is human or another AI. This whole discussion revolves around discerning the underlying code -- without seeing it! -- based on observations of the results of the code.

Confirmation bias is real; I often remember stories, data points, and experiences that confirm what I already believe to be true. It would be interesting to look at the code that Soren released for Civ4 Ai to see how often the logic actually tests for a human player. Obviously Civ3 is not the same as Civ4, but the designer and design philosophy would be very similar.

 

[Fergei]

With respect, your post contributes zero to the debate other than an attempt at character assassination.

Since you have done zero research yourself, why do you consider yourself an authority on which 'other people' to 'believe'?

Herd instinct?

I'm not sure what grounds you have to say I've done zero research. I've spent many tens of hours testing game balance on the same Earth map, with full map vision, with the same Civs when testing modifications in government balance, Civ attributes and unit strength. Usually AI only, but sometimes with me as a player. With the exception of barbarian camp locations, this recreates identical game conditions that enable like for like comparison to isolate and analyse the impact of a single tweak to settings. It is from this process that firmer conclusions can then be drawn.

At no point during those tens of hours have I ever detected any suggestion that an AI paths its units differently based on whether its neighbour is an AI or human. Things like the AI attacking units loaded on a boat attacking the least well defended opposing coastal city do not take into account if the enemy is AI or a player. Again, you can only establish this in debug mode with the full map revealed.

Further research is watching the video posted above where we hear directly from the head coder of the AI that making different rulesets for behaviour versus humans takes additional time and that broadly they weren't in favour of it. Your position appears to be that

A) the coder Soren has intentionally misled listeners on that matter for reasons that are unknown
OR
B) that there is a misunderstanding in the above video and that actually, during the intense time pressures of making a game, Soren has actually diverted time and resource to coding specific behavioural rules for AI unit movement specifically in relation to human players even though he considers such coding as contrary to his intention for the game
OR
C) some random dude spent days and weeks behind Soren's back, coding this parallel AI unit movement code when near human players, got play testers to check it out behind the game designer's back and somehow got it inserted into the final release and two expansion packs without the lead coder noticing.

Whereas it appears to be the position of every other poster and the lead coder is that you are:

D) Incorrect

It is up to us all as individuals to make up our own minds as to which of the four possibilities above are most likely.

If you consider research as playing a series of one off games and submitting AI behaviour to an eye test through the partial vision of fog of war. Then no, I haven't conducted that particular stream of 'research'. Primarily because it would be a massively inferior method of identifying the impact of changes to game settings that would leave the player unable to draw conclusions with any degree of confidence about many aspects of the game when compared to using the debug option that comes with the game.

 

[Lanzelot]

Buttercup, on most points I'm completely with you:

The AI doesn't move randomly. If it moved randomly then it couldn't walk in a straight line, the odds of that happening via RNG are remote. This suggests the AI uses a 'go to' order when it moves its Scout rather than making an individual new RNG each time it moves a square.

Completely agree here. In fact, this is what I tried to outline in my previous post: what your screenshot reveals about the algorithm the AI uses for moving its Scout. Your suggestion is that it is coded to go straight for the human player, if possible. My suggestion is basically: it picks a direction (perhaps influenced by the RNG, perhaps influenced by the terrain - with terrain allowing 2 moves preferred over terrain that would allow only 1 move, as vorlon_mi suggested) and then follows that direction until it hits an "obstacle" or sees something that either attracts it (goody hut, foreign border) or frightens it (barb camp, enemy unit). Then a new decision is made, excluding the direction where it came from.
That theory would also explain the observed path:

1. From the starting square, W is chosen. (0-1)
2. It hits the coast, so can no longer keep going W. A new decision is made and it now goes S. (1-3)
3. Now it hits the isthmus and changes to SW until it finds your border (3-6)
4. Again it changes direction and now goes NNW (6-8) until it "sees" the goody hut.

Scouting has three main functions - Revealing land, meeting other players, opening goody huts. It is remarkable that the civ AI does not care about revealing land around its own capital before charging off into the sunset.

I'm not sure why you suggest the area around the Arabian capital is 'already known territory', seems like a really bizarre leap of logic. Surely a circular route around its own capital's borders would be the ONLY way for it to get it's own territory to the sate of being 'already known'.

The scout is not the AI's only unit. The AI could be programmed to use the fast-moving scouts to discover far away targets quickly (before barbs start roaming around) and leave the close-range reconnaissance to the "local" units. (Especially if NoAIPatrol=0 is set, one can observe, how the surplus units keep hurrying back and forth constantly inside the AI's borders...) If the second or third warrior discovers the vicinity, it is still early enough to get the necessary close-range knowledge for deciding about the best first settling locations. (If the AI would really need that knowledge... The algorithm that determines settling locations, could simply use the "total map knowledge" we know the AI has. This would be supported by the observation that many have made, that the AI settles apparently worthless tundra spots in the ancient/middle ages, and then surely enough it turns out that oil, rubber, uranium etc. pops up there two ages later...)

This theory (AI using scouts for far away reconnaissance and its remaining units for close-range) coincides with vorlorn_mi's observation above, that it seems the AI is sending out its scouts on paths that look like the spokes of a wheel.

I can't imagine the AI behaving like yourself, I believe the most common assumption about the AI in all other discussions is that it invariably DOES NOT act in a way that a human would, so why would you use "what I would have done if XYZ" as a means to be blind to what's being shown?

That's a valid point here, yes, mistake on my side. When trying to uncover, how the AI is programmed, I should not take into consideration what "makes sense" from a strategic point of view, or "what I would do" in such a situation. Only the facts that can be seen in the game should be taken as a basis for our reasoning. (And perhaps the statements of the developers, like Soren's talk or other things that have been confirmed by one of the devs of the game who posted here in the Civ3 forum in 2001-2003.)
So the "scientific approach" to tackle this question would be as follows:

• Create a completely symmetric map (in order to rule out the terrain having an influence on the AI's actions, e.g. if one direction is jungle/marsh/mountains, while the other direction has mostly open land, the AI might prefer to always send the scout into the direction of the open land).
• The map should look like a wheel or a cross with on expansionist AI in the center and 3 AIs plus the human player on the four spokes.
• Turn off the "preserve random seed" flag to make sure we get fresh random numbers each time we play. (If the random numbers are always the same, the AI's decisions might also always be the same, even if influenced by the RNG...)
• Then play this map 100 times.

If the AI would really be programmed to always beeline for the human player, we should see the scout appearing on our borders first in all 100 cases.
If we see the scout appearing in only 25 cases (plus/minus 2-3), the decision is made by the RNG.
The third possible outcome could be that we see the scout appearing in a significantly higher or lower number of cases than 25. E.g. 40 or only 10. Then we would have a "weighed" RNG, implementing a "preference" (or reluctance) to go for the human player first.

I test like this would be much better suited to uncover the truth than any anecdotal evidence. As vorlorn_mi correctly said, confirmation bias is real, the human mind is very good at "cheating itself" and remembering only those events that confirm its own opinion, while silently forgetting those that contradict it.

 

[Lanzelot]

PS: the above test should also be repeated a couple of times with a different "nationality distribution" on the 4 spokes. Because if the AI has let's say a hard-coded preference to try meeting Rome first, if Rome is in the game and reachable, and the human player happens to be Rome, it may look like the AI is programmed to meet the human player fist, while in fact it isn't...
If the test is repeated with the human player taking all 4 spoke civs in turn, we will rule out any such factors.