Well, to say the least, I guess I can classify the reaction to that suggestion as 'negative'
(did I mention the word 'option' in there). While I do think this has more potential than the general consensus here has given it(and I don't think what I generally had in mind was well understood), I do also think it's also a minor point that would only bring a small enhancement to a small percentage of players. Axed, and... moving on to more important things.
Right now I'm taking a close look at AI tech decisions, seeing what techs are over/under valued, what likely orders the techs are being researched in (taking flavor values into account for different leaders, as well as things like economic conditions and warplans/current wars, etc.). I've written up code to give me useful feedback on the current valuation of all possible tech choices by any given AI on the given turn, and output it into the help text area so I can do this more efficiently.
The main thing I'm looking for is techs that
should be considered valuable that aren't (again, considering the civ in question and various conditions) and those it probably
shouldn't consider valuable but yet it is. Of course the ultimate goal is decently situationally intelligent choices. If there is no immediate threat (and they aren't planning on making themselves a threat to someone else), they should be making sure they are seeking techs to grow their economic and science output, while making sure at a minimum their defensive units are up to snuff. If they are planning or in a war they should be valuing more powerful units higher. If there is a religion available to be founded that fits the civ (I'll be giving religions 'flavor' values like techs have) it should be as willing as a human to try to take a shot at beelining it. It's a real balancing act between smart decisions and unpredicability.
Fortunately the things I'm aiming for aren't anything new, it's how it's already coded (in principle). So what I'm really doing is not re-writing the code, but finding things that aren't
implemented well and finding better ways to value them. An example is the improved well and mine improvements. Around the time that Planetary ecology comes within three 'path steps' of being researchable (the maximum number of techs the AI is programmed to look ahead), it was uniformly becoming one of the top valued techs to shoot for by pretty much everyone. While it is a decent choice, it's not that much better than the other techs it was competing and winning against.
The primary reason is the deep well. The human knows that while the deep well gives x additional water in a tile, it's really only 1 more water than their current shallow wells give-not
that big of a deal. The AI was looking at it without this perspective, instead seeing it as
really good because it gives all this extra water!(which the AI rightfully values more than other improvement yields). To correct for this overvaluation, I added a tag for improvements to indicate if it is an improved version of an other improvement essentially, and the AI now looks at the deep well in terms of what it adds
on top of a shallow well improvement, just like a human. This same tag is also used for the improved mines.