The problem with games like Democracy 3, or indeed Gratuitous Space Battles, is that they are VERY hard to balance. Balancing a game is exceedingly hard, even if you agree, like me, that ‘some’ inbalance is actually what can make a game fun. (If all choices are equally effective, why pretend there is strategy to the game?). DFor big AAA studios with huge testing teams, pre-release balance is easier. For one or two man indie studios, its almost impossible to do in house.
The common solution is early-access / pre-orders with beat access, which get a lot of people to play the game. However this assumes that
a) Those players are representative in terms of skill, time, and play style to the wider public
b) You get decent feedback from all those players, not just the hardcore and
c) The feedback from the players is impartial and can be relied upon.
With a game like Democracy 3, c) is a real issue. A lot of people complain that capitalists are hard to please, but how do we know if that’s true, or if its feedback from socialist players? Anyway, as a geek, and a coder, and someone who embraces complexity, an alternative strategy of letting the game balance itself obviously appeals. So clearly, getting the game to play itself constantly and discover, then report, and maybe even auto-tweak and adjust its own rules, is the ultimate goal.
This would be an ideal ‘summer vacation’ project for me.
Actually, the bit most people are probably wary of (coding the bit where the game comes up with a strategy, plays the game and developers improvements to its play style based on the results), isn’t the bit I’m worried about coding. That bit wont be too bad (i’m not aiming for alphago levels of skill here), the bit that makes me roll my eyes and go ‘Can I be arsed?’ is the mechanics of introducing a wrapper to the game that lets it play itself. The problem here is just doing a nice, clean job of introducing such a system without breaking the game or introducing anomalies.
This is where Alphago is AMAZING, because it simply uses visual input. Frankly that ain’t gonna happen, so I’d have to embed code in there…or would I? an ideal system would be hands off, a new ‘GeneticD3.exe’ that launches D3, non invasively, makes player decisions, then handles everything without touching any D3 code directly… Surely that cannot be done?
Mwahahahaha. It possibly can…almost, because D3 is already set up to autosave each turn. And because its turn based, all the ‘decisions’ happen at once. So all I really need to do is to write code that parses the D3 save game (handily already xml), makes decisions as to what to do, runs those decisions through the D3 engine (that bit gets tricky), and then dumps out the next turns save game, repeat ad finitum. To do some rough maths, lets say saving and loading D3 takes about 2 seconds (I rule at optimizing), and we give a generous 2 seconds to make a turns decisions, thats 4 seconds a turn, 64 seconds per term in office, or a 4 term complete game in 256 seconds, so roughly 14 games per hour, or 300 games a day. Thats actually remarkable slow. If AlphaGo needed 100,000 games to learn, I’d need the best part of a year. Still, assume some cunning multithreading and optimizing (I don’t ever need to touch the disk really, all save/loads could be trapped in a RAM buffer), I could probably quadruple the speed, so 1,200 games a day?
I am seriously thinking about it. It would definitely be fun.