Civ 5, with the add-ons, is one of my favourite games of all times. It's really well-balanced and thoroughly thought through. Everything is just in place, game mechanics, city management, unit system, science tree, social policies, trade, etc. I do turn off espionage, though. And the AI leaves room for improvement.<p>Another thing that I also really like about the game is that you can get away with very little micro-management. However, some some micro-management gives you slight advantage. I think they struck a nice balance here.<p>And beating deity was great, it almost felt like playing chess. Sure, you have to have a bit of luck with your starting position. And you need some luck later-on when it comes to the distribution of certain resources. But apart from that, it's in your hands and everything you do has to be near-perfect in order for you to succeed.
I love playing these kind of "strategy" games (Civ, Total War, to a lesser extent DotA 2, etc). I don't have time anymore, as I am a startup founder, and I simply can't play these for just 1-2 hours a week. I'll play again in a few years :)<p>What typically annoys me is the AI in these games: never challenging enough, or challenging in the right way. Years and years, and the only hope is that the AI team at Google will figure out a way to apply their ML/DL AI to many games at once. (they did it for DotA 2, with remarkable results).
I recommend watching true start location earth all AI play through. It's interesting to watch how America is divided if Europe wasn't technologically ahead.
Now give these players a side channel to communicate and you have a solid research paper. An example here: <a href="https://arxiv.org/abs/1704.06960" rel="nofollow">https://arxiv.org/abs/1704.06960</a><p>Plus all of their code is on github
I was surprised at how much time they spent building up their cities, technologies, religions etc., rather than just doing each others head in.<p>There's this open-source game, Battle for Wesnoth, which is sort of like Civilization without the civilization part, so it's pretty much just hexgrid- and turn-based strategy warfare (and that part much more fleshed out). Would be interesting to do the same with that.
In Civ 5, There are so many things to choose from that give tiny 10% or +1 bonuses here and there. Unit promotions, religious bonuses, civics, buildings. I get bonus hunting fatigue.<p>At least in Civ 1, even if there was only a handful of different forms of government, they had massively different effects. I've played every Civ game except four and six.<p>Certain elements in the map in Civ 1 gave good bonuses.
Now everything's watered down. I feel it's a cop out way to balance something by making everything have a tiny effect.<p>The automation also leads to you playing by reacting to popups, which is not fun. Oh, this city state now wants this random thing. Your game totally hangs on the ability to get them on your side.<p>It's fun to make plans and try to follow them through. It's fun (and harrowing) to make interesting decisions. You can't do it with just such random elements. It's not fun calculating cumulative bonuses. And it's also not fun even if it's automated. Have less gameplay elements!<p>Civ V is a game for coupon hunters and bean counters.
The result has a lot to do with:<p>- what Civ is next to each other: each civ has unique units, buildings, etc. this can be decisive during an engagement.<p>- resources<p>- what alliances form<p>There is also another factor in play here: you will see that the last civilizations to lose are the ones that secured the corners. This is because they have a smaller border to protect. In this sense it's similar to game of Go, were openings often start with corner enclosures.
This might be OOT, but I'm just curious how do you train AI on video games? Do you connect AI to internal API of the game or you let AI parse image data the from the screen and use external input (like keyboard or mouse) in order to play the game?
I'd be interested in some of the math here - e.g. if you wanted to compare civ fitness - would you need to test every iteration of initial starting conditions? And then I wonder how that changes if the map is a hexagonal sphere...