I noticed something interesting along these lines in an observer (debug as Japan on its own island, Marla Singer map) game the other night. Russia founded a city very close to China very early on, which must have been via a hut. It's 41 tiles from Moscow, and was founded in 2390 BC, which is turn 34 of the game, so it's impossible that their first Settler traveled all that distance to found it.
Here's the city:
(It's worth noting that due to being a Vanilla conversion and radar towers being enabled, most AIs used their starter worker on a radar tower, and thus AI expansion was pretty slow; many AIs hadn't founded their second city by 2390 BC)
Here's the optimal pathing route from Moscow to Novgorod, in Clean Map mode to hide irrelevant details:
Turn 21 = 2950 BC, so the scout didn't make a beeline directly for that hut. Of course, not knowing this was going to happen, I wasn't watching the scout super-closely to see which route it did take.
It generally supports the "AI likes to explore in a general direction, influenced by various terrain factors", but is again one data point. Still, it shows that the AI will sometimes end up with goody hut cities quite close to other capitals, include AI capitals. Given that I was not land accessible on this map (in 800 AD, no AI has made contact with me yet), it supports the results being luck of the draw. Though again, one data point.
Lanzelot is right that the way the AI is programmed and the way we think "makes sense" likely don't line up, especially as we have 20 years of experience playing Civ and talking about strategies that even the humans of 2001 wouldn't have known about. Still, I'm sure there were various reasons that Soren programmed the AI the way it did. Scouting the immediate area is something we know helps figure out where to settle. But scouting far away in a general direction can help lead to contact with other civilizations, and thus trade. I agree with the statements that the direction does not appear to be completely random each tile. Having programmed an exploratory AI for the C7 project that does exactly that, the units tend not to make it very far from home.
I like the proposal of a scientific approach to studying it. And it could be used for a lot of questions. Will the AI connect a luxury or strategic resource, of equal yield, first? Does it matter if you make the unit that the strategic resource enables more powerful? How much to various factors affect AI tech selection? Etc. In some cases there's community knowledge based on an accumulation of anecdotes (e.g. the AI seems to like techs that enable new governments), but if someone had the patience and motivation to do such studies, a good amount more could be learned. Of course, it would also be a tedious process, which is why it's probably unlikely to be done.
Here's the city:
(It's worth noting that due to being a Vanilla conversion and radar towers being enabled, most AIs used their starter worker on a radar tower, and thus AI expansion was pretty slow; many AIs hadn't founded their second city by 2390 BC)
Here's the optimal pathing route from Moscow to Novgorod, in Clean Map mode to hide irrelevant details:
Spoiler Wide screenshot :
Turn 21 = 2950 BC, so the scout didn't make a beeline directly for that hut. Of course, not knowing this was going to happen, I wasn't watching the scout super-closely to see which route it did take.
It generally supports the "AI likes to explore in a general direction, influenced by various terrain factors", but is again one data point. Still, it shows that the AI will sometimes end up with goody hut cities quite close to other capitals, include AI capitals. Given that I was not land accessible on this map (in 800 AD, no AI has made contact with me yet), it supports the results being luck of the draw. Though again, one data point.
Lanzelot is right that the way the AI is programmed and the way we think "makes sense" likely don't line up, especially as we have 20 years of experience playing Civ and talking about strategies that even the humans of 2001 wouldn't have known about. Still, I'm sure there were various reasons that Soren programmed the AI the way it did. Scouting the immediate area is something we know helps figure out where to settle. But scouting far away in a general direction can help lead to contact with other civilizations, and thus trade. I agree with the statements that the direction does not appear to be completely random each tile. Having programmed an exploratory AI for the C7 project that does exactly that, the units tend not to make it very far from home.
I like the proposal of a scientific approach to studying it. And it could be used for a lot of questions. Will the AI connect a luxury or strategic resource, of equal yield, first? Does it matter if you make the unit that the strategic resource enables more powerful? How much to various factors affect AI tech selection? Etc. In some cases there's community knowledge based on an accumulation of anecdotes (e.g. the AI seems to like techs that enable new governments), but if someone had the patience and motivation to do such studies, a good amount more could be learned. Of course, it would also be a tedious process, which is why it's probably unlikely to be done.