Paradox AI

[Preacher]

I am speaking from ignorance, my dear fellows.

Ever since I have been introduced to Paradox games, or any strategic game, really, there has always been massive complaints about the stupidity of AI, and I agree. I desire a more human AI so when I reach a certain skill level I don't rofl stomp everyone.

But if these complaints have been so on-going and in much quantity, wouldn't a company do something about it to validate/reinforce their reputation, to show their customers they care, and to attract more customers by being a company that listens?

A stray thought struck me in the shower the other day: what if the devs of all these strategic games can't do anything about the AI? Wouldn't that make sense they are essentially "ignoring" complaints about the AI? Wouldn't it be bad business practice otherwise? Perhaps this is the epitome of AI in our current time without massive amounts of coding/programming?

I know nothing about programming except the small MATLAB class I took in college. How hard is it to develop competent AI (no cheats, i mean actual intelligence)? Can it be developed past what we have right now in our current time?

Those are my shower thoughts. Again, I'm no expert. Tell me what you guys think :P

God bless

[Marquoz]

It's both a CPU cycles limit and a developer time limit. The AI for a game like EU4 must attempt to find a good path for literally hundreds of nations simultaneously while adjusting to changing conditions. That's an enormous challenge, and without a huge programming staff and super-computer level hardware, there are (as you correctly pointed out) real limits to what can be accomplished.

Look at a game like chess, for example. AI can beat the best human players, but chess is a two player game with no random element and a very limited board and number of pieces. And winning against grandmaster level opponents still takes a supercomputer.

So yeah, don't expect too much from the AI in strategy games. Learn its limits and you're on the way to being a consistent winner.

[bri]

Yeah, it takes a supercomputer ai to win at fairly simple games like go, chess, and checkers. Trying to design one that will provide a top level challenge in strategy games with a lot more variables to consider is really not feasible. This is especially true when you start talking about "real time" games, at least with turn based games you can allow the ai to run a lot longer and thus make better decisions as long as you don't mind the "OMFG these turn processing times are ridiculous" complaints...

[Raider Deci]

This essentially still a 2013-game.

And as much as deepminds alphastar-thing for starcraft 2 is interesting (that AI that reach grandmaster rank in multiplayer on the european server) its something that takes alot of time, money and resources to focus on.
Who knows if that can be applied for a paradox game in the future, im certainly not knowledgeble to answer that

[CyPunk]

It is an interesting question, so allow me to ramble on for those with the patience. For something like EUIV I would venture developing a better AI that actually uses machine or deep learning would just be a question of resources and time, mainly money for the personnel and compute. The algorithm itself would probably not need to be something super innovative. The aforementioned Alphastar solved many of the big issues about missing information (fow, HUGE issue) and tactical action vs. strategic planning.

However, this would just be for show, since the hardware to run the developed AI would be beyond most people's personal computers... but not FAR beyond. The main computationally expensive task is training the algorithm, which they could do on a nice dedicated machine or two ..or 100. Running that on another computer would be less computationally expensive. Yet to again qualify myself, it would not be as computationally cheap as say running an AI app on your smartphone, because the type of problem presented by wining a strategy game would require continuous learning, so it could handle the new environment. So Paradox would develop a reinforcement learning algorithm, that can be the basis for the AI on your machine (that is it would use transfer learning) but that AI would still have to learn more on your game given it would always be different.

You can already run multi-agent simulations with hundreds of agents using reinforcement learning on a circa 2000 euro home computer, albeit their tasks remain fairly simple... as far as I know. I am no expert, more someone who tinkers, but I expect the first thing we will see are mods with some AI ability for specific tasks - just as we see in the real world: no general purpose AI, just task specific. Even here though I would say you would need a decent machine to run them regarding RAM, a multi-core processor, and a RTX GPU (these are perfect for training AIs, for those who do not know). Two GPUs would be better - one to train, one to run all the game itself on.

Maybe next time I am need a project... hell, I am unemployed anyhow :-P

If you want to watch a decent coder tinker with some computer vision on a totally unrelated game (CyberPunk 2077) check out:

https://youtu.be/dUU6ZsJlZKQ

[CyPunk]

Thinking a bit more on it, if one actually wanted to implement an AI update in EUIV - as opposed to making a new one for EUV - it would probably be easier to make it a separate "player" programme that joins in via multiplayer. But I may think this because I know nothing about modding a game.

[Ashling]

They have improved the ai. The most recent change I think was allowing them to carpet siege and forcing them to do something else if they would otherwise do nothing/walk into a disadvantaged situation.

But realistically, you don't really want human-like ai, because humans play this game with an insane level of sternness. To win a war against a human you must only play the optimal way which obviously stifles creativity. And fun. Plus it's non-welcoming to new players who don't already know the optimal way to play and they'd just end up losing war, after war.

There's definitely a balance you can make, but at the end of the day EU4 adds enough variety that people of high and low skill levels enjoy the game whether that be reconquering Rome as Byzantium before a certain date or playing as Portugal and beating up nations several mil techs behind you.

[tonypa]

Personally I think PDX AI is fine. It needs to be fast enough to control thousands of units each step and smart enough to be mostly competent. Sure, it has bugs and it makes mistakes and sometimes does stupid things, but so do humans and these errors make the game more interesting.

All these fancy new words like "machine learning" and such, these are nice, but not required. Capable AI existed in specicif games decades ago, not the universal AI able to solve everything, but coded for specicif task. In the end, the code just needs to be able to run the game and the game needs to be fun. It does not have to be detailed simulation of everything.

[Rumi]

dude don't make the a.i. good or i'll never be able to have fun with the game