HOI4 AI status

[SteelVolt]

Nice thread with good points. I am happy to see some attention on the AI
While I am not willing to go into details about the state of the AI right now, rest assured the work on it continues and I hope to spend a lot of time even beyond release to refine it.

 

[SteelVolt]

The best Ai is simply to copy human strategies. I've always liked Forza's Ai system where it mines player data in order to present a Ai strategy in a race. Totally different. But I generally think the best route is to learn and preform how a player does. Instead of having it think on it's own. For example one of the biggest issues is AI thinks 1 2. Not 1 2 3 4 5 6 7. Learning from Chess computer programs might be useful. I just know when fighting against the Ai it's easy to do tons of tactics that the Ai never defends. Tons of them. I hope it's changed. But I'm skeptical. I haven't seen somebody mirco everything versus the Ai on its own.

As awesome as that would be, it simply is not technically feasible for a game like HoI. As a comparison, consider the recent successes of googles deep neural network tech defeating champions at Go.
Go is significantly more complex than chess because of the much larger number of possible moves per turn. While chess AI gained success by calculating future possible board configurations and attempting to steer the game in a beneficial direction through this search tree, Go just has too many possible board configurations to do that (significantly more than the number of atoms in the known universe).
Google essentially used a supercomputer and a ridiculous ammount of training data (aside from letting it play against itself) to accomplish this feat.
While Go has 361 positions HoI 4 has over 13000 provinces. The number of possible states of each one of the 361 positions in Go is 3. The number of possible states of a province in HoI4 is hard to quantify with a discreet number. And apart from provinces there are multiple off map systems and partially off map systems.

On top of that Go is a turn based game where the AI can take time to process each move. HoI4 is pausable real time. Even if you consider each hour a turn, there are still expectations on the AI not taking several seconds, let alone minutes, to process what it is doing (which can be any number of things as part of the same tick.)

The math on this just explodes.

Frankly, as I write this I am amazed at the level we have managed to get to already with the HoI4 AI.

Trust me; when or if the time comes that this is a viable approach I will enjoy looking into it even if I have to use my free time to do it

 

[Wizzington]

Suggesting that we use recorded learning for a complex strategy game AI because it works in chess/racing is basically like suggesting that you solve a broken spine with massage because it works for back aches. The scope of the problem is simply so far beyond the solution that the solution ceases being applicable.

 

[Groogy]

If I'm not mistaken. Killer Instinct has learing AI aswell. Learning frame traps& combos. Mimicing most player movement,etc..

Edit: Is there really no way for AI to learn small things. At least production order? or most common Teching or generic National Focus?.

First techniques used in Fighting games, FPS's or other genre's are not really applicable in Strategy games and they have a completely different focus on the problems they try to solve. Want to point out there is also a vast difference in purpose between the AI Google made and an AI in a game. The AlphaGo is supposed to beat the human, the Game AI is not. The AI in games is there to give the player entertainment. That is the one thing that holds true between all genre's even if they work on different problems.

And the AI in Killer Instinct is not a learning AI, it is an AI that records your movement and replay them, I was to their lecture on GDC and if we would do the same thing for let's say Crusader Kings II from only four hours of gameplay we would have recorded a database that would start reaching several GB on the players computer. Killer Instinct have very few potential input to record making it a viable option.

A learning AI is usually attribute to an AI that can itself decide if it is making a better job than the previous time it tried to play and as such can self improve. For instance in racing games a learning AI will use the time to finish as it's measurement for success, what in our games should the AI use for learning? World Conquest? How do we even measure that success in a statistical way for the AI? Also do the player actually want an AI that beats it at every turn? Not really, they want an AI that keeps them entertained and an AI they can beat. An AI that they can't crack is just going to make players rage quit and call the AI cheater even if we don't give it cheats.

I have some things I want to try out in the future that would make the AI more competitively against the player, question though is will it be fun to play with? People don't get mad at beating the AI, actually it is the opposite. But if the AI is hilariously broken then they don't feel like they've beaten it.

 

[Groogy]

But if you played a human that beat you fairly (non-gamey approach, just optimal gameplay and great decisions while keeping it logically sound and plausible) would you rage quit and call him a cheater?

This happens in every single multiplayer game where you have 18-32 players in it. So yeah most people would. Believe it or not, but when people lose they stop being rational.

 

[Groogy]

Is the Battle Plan system helping the AI at all. Like will they use it?

Yes the AI plays with the Battle Plan system so you can actually see your ally AI paint plans on the map.
SteelVolt has been the main guy working on that and it was a huge problem to solve and wasn't easy.