I think that AI wait time is related to how long your turn takes. 10 seconds or so is totally reasonable if your turn takes at least a few couple minutes.
In your civ setup, how long do your turns generally take? Are you kicking them out in under a minute?
Yep, but I usually know what I am doing several moves ahead so I am pretty much always waiting for the computer to finish it’s predictable behaviour. There’s not going to be many surprises in most turn based strategy games once you see where your enemy is clearly.
You guys have me craving some RTS-ness. The last one I played was Crusader Kings 2. I think I put in about 5 hours and only scratched the tiny tip of the starting surface beyond the instructional content.
I think this is probably important to make the game feel like the player is playing against another real person, although I don’t ever remember playing a single player campaign wanting this feeling. Personally when playing against AI I’m interested in the “template strategy” type the AI is going for and how well it might be attempting to mask it.
Beyond having to crunch the options and information gathered which I guess takes the most time, wouldn’t a full up master AI whoop all of us fleshlings anyway without effort? That is to say - AI can be designed/created to win every round, every turn, every choice, and the rest of the game allows it because it maxes out the damage and resources and results of those choices.
So adding ‘gates’ that meter the AI to possibly choose a less optimal path, to maybe make a less ‘smart’ decision, to second guess itself and it’s template strategy, (aggressive - war), wouldn’t these gates to review each choice and option slow the turn down quite a bit?
Note: Gate is a term from this artist to describe metering the flow of info.
Template is used as a term by this artist to describe a basic postures of play, aggressive, passive, passive-aggressive, tech win, race to the stars, etc, etc.
The thing is though that when you introduce limitations that are not completely logical or straightforward, it becomes difficult for an AI because the intuitively optimal path can be actually worse than one which is practically optimal (or high risk/high reward). For example making suicidal attacks that expose an opponent’s weakness which you know that you can take advantage of in a small window of opportunity later on, or something like that. In strategy games I’ve found that AI that is trying to make good short-term decisions can be very exploitable, and good planning seems to be hard to come by.
I think that if you set up an AI to behave like a real player would, with pre-programmed limitations and procedures, it becomes more of a challenge to prevent a clash between its ‘brain’ and its ‘body’.
In addition, RTS AI typically has access to information the player doesn’t have access to. Stuff like no fog of war, knowing where the player is.
This (Blizzard and DeepMind turn StarCraft II into an AI research lab | TechCrunch) might be an interesting development which makes AI more like a real player.
I honestly think ‘playing like a human’ is the wrong way to approach an RTS AI.
One of the best RTS AI experiences I’ve had was playing AI Wars. The developers shamelessly cheat with the AI. It doesn’t even play by the same rule set as the human players. The basic idea behind the game design was to only give the AI tasks it was good at, and the same for the human player. Tasks that humans were bad at were removed from the human rule set. And the same for tasks the AI was bad at.
For example RTS AI tend to be notoriously bad at managing and defending a resource gathering systems. Its often possible to win by simply selectively targeting the AIs spice harvester, peasants, metal trucks, SCVs, engineers ect. Or you can destroy the enemies gold mine, oil rig, tiberium field and so on. Or you can just wait until they run out of resources. Its a valid strategy to win, but it often cheapens the whole experience. AI Wars gets around this by simply not requiring the AI player to collect resources.
Depending on the game, having a very robotic AI can actually add strategy to a game, since the AI becomes predictable and you can strategize about how to beat it. Although that works best for games with a simple/abstract presentation.
The AI that I find most annoying is always diplomatic AI. The diplomacy AI systems are miserable in any game that has diplomacy.
Other than that - it’s mostly just really stupid things that bother you - like if the ai sends it’s guys one at a time into obvious suicide or gets stuck.
While I don’t think that AI Wars is a good example here, I completely agree with this bit. When I talk about feeling as if I’m playing a “real” opponent I’m talking about presentation, not computation or algorithmic approach.
When I play a turn-based game with another human I can see them doing stuff. I can’t necessarily see the results, but there’s some visible activity, and I can see their movements on whatever shared board there is. It’s a part of the experience, and to me that’s an important part of playing a strategy game.
If my opponent was just some instant changes that occurred when I clicked “End Turn” I’d feel that it was trivialising or cheapening the overall experience. For stuff that I’m meant to be able to see (eg: when they move a visible army on a board) I also want that clearly presented - not just one of a list of changes, or something I have to look at a board and play “spot the difference” based on my memory from last time.
As to how they make the decision of what move(s) to make, that’s a completely different (and somewhat difficult) problem to tackle.
Could the depth or complexity of the game be a major factor?
For instance the more moves or actions available to a player the deeper and more complex the possible moves.
So for games like Total War if they map the game down to individual soldiers and not just a dice roll a squad/troop/battalion then it would be more complex for the AI to try different moves. (would it also depend on the D of the dice).
Or for civ style games that probably have more complex simulations under the hood representing the countries and people and the impacts of various moves combined with more possible actions per turn.
So if you made a TBSG and wanted a fast AI you would want it simpler with less depth of simulation and limited moves/actions.
Although you would also want to limit the number of Units as even with limited moves actions lots of Units compound the complexity.
Probably depends on what one’s going for. For fun to play though I imagine you’re right, considering other things I’ve heard about game AI. http://askagamedev.tumblr.com/post/76972636953/game-development-myths-players-want-smart