OK, I'm re-entering the discussion. I had an idea to bridge the gap: an idea to try to implement what Luca keeps talking about (over... and over... and over) using game mechanics that are similar to what we've seen in Civ games before. It doesn't give him everything he wishes for, but perhaps it's worth talking about.
Disclaimer 1: I personally would not find this proposed game much fun. I'm offering it as a possibility. Judge for yourself whether *you* would find it fun, in a 4X game.
Disclaimer 2: Earlier in my career, I wrote computer programs to solve problems in my job. Even earlier, I studied engineering at university, where I learned about approximations, building models, and the compromises and tradeoffs which MUST happen in every project.
First, the old ideas I would like to reuse:
- In Civ3, individual pop points (citizens) in a city were tracked with certain attributes. What nationality they were, whether they were happy/content/unhappy/rebelling, whether they were a specialist or not.
- In Civ4, the tiles on the map were tracked with certain attributes,including a nationality or mixture of nationalities. Ownership of the tile might shift due to cultural pressure.
- In Civ4, production of Great People was influenced by the presence and identity of specialists in a city, as well as the construction of certain buildings or wonders. The probability of producing a particular GP varied. Indeed, players were frequently surprised when a Great X appeared when they were expecting/hoping it would be a Great Y.
- In Civ5, as Krajzen wrote (way back in post 212), the sentiment of the the people for one ideology can be influenced by the actions of neighbors who follow another ideology.
My proposal is to model the beliefs of the pop points in the civilization to include their religion, their nationality, and some approximation of their class (field worker/merchant/landowner/ruler) and eductation. To work towards Luca's vision of a next gen AI, the net value of each pop point could be an average of sub groups. If each pop point "represents" 1000 individuals, give them non-integer / floating point numbers to represent those attributes. Their happiness, or conversely their desire for change, could reflect a number of internal and external factors. For example, building something like a Commercial Hub with multiple buildings could lead to more merchants in a city; more tendencies towards free market civics; less favorable to heavy taxation. One can think of similar examples if the player builds many mines and something like an Industrial Zone with its buildings.
Here's my new idea: the rebellion index. Think of the per-city happiness in Civ3 (where sometimes riots/disorder happened) or the per-city happiness and healthiness in Civ4 (where rebellion or sickness could occur). Now INVERT that -- the individual pop points (or the finer grained averages) will start getting grumpy over time. They will eventually ask for change in a civic, policy, or government attribute. The tendency for *which* change they will request will depend on a blend of factors (Civ4 GP, Civ4 tiles) and may include some suprises. If the player has a whole bunch of factory workers in the Industrial Age, with insufficient representation, they may agitate for some form of socialism or collectivism. If the player has engaged in many miiitary operations, not always succeeding, the population may agitate away from fascism. If the player has many foreign citizens who follow a different religion, the population may demand religious reform.
Lastly, the grumpiness / anger / possiblity of rebellion increases as the city size grows larger.
The player would need to respond to these challenges by either: changing the policies/civics, constructing certain buildings or improvements, adopting a different religion (as in Civ4), or diplomatic measures -- signing treaties, trade agreements, or alliances. The challenges would vary be era and the evolution of the civ. The numbers that go into the calcuation of the rebellion index would change with era, with techs researched, with cultural devices adopted. Depending on how they're balanced, more than one source of discontent could arise at the same time.
Approximations: No, it's not true machine learning. It's not modern AI. It's a potential method of abstracting culture, ideologies, and attitudes which is more multivariate than we've had before. Numerical calculations are still pretty fast, so the impact on turn times should be tolerable. Sorry, Luca, I can't think of a way to get *all* of what you asked for. But it's a path to get tr