Axioms
Arcane
- Joined
- Jul 11, 2019
- Messages
- 1,516
This is a revised version of a post I wrote on my old blog, on blogspot lol, back in March 2015. It discusses one of the most popular complaints about 4x games and other similar games. One person who read the post said that: “It makes it clear that he's completely insane. And not in an 'oh you, your so crazy' kind of way; in a 'I had to destroy all the funiture because the Russians were listening through my couch!' kind of way. He also has the writing style of an 18th century rationalist philosopher.” Joke’s on him I haven’t sat on an actual couch since like 2012.
In any case this post is about the perennial problem of strategy design, mostly regarding 4X or map painter games, which is good old snowballing and the limitations of AI in strategy games. The first issue is the difficulty of the AI. If you get tired of a specific topic I do recommend skipping to the next header. Especially if that section is just review of stuff you’ve already thought about.
Players Will Always Beat The AI
A lot of ink has been spilled on the debate over “cheating AIs”. That is AIs that get the strategy version of a bullet sponge buff. Tons of resources, units summoned out of nowhere, perfect knowledge of the entire map. In Total War the tactical AI gets large morale boosts, or it did back in the day. I don’t recommend cheating AI as a strategy but it seems a little hypocritical for time warping omniscient deities in a vast interconnected network of fellow deities to complain about cheating.
The human player has a monstrous pile of advantages over the game AI, even a cheating one. I considered putting the infamous meme of a massive pile of stuff outside an undersea pineapple here to illustrate the difference between the player and the AI. Players can rewind the game an arbitrary amount of time. They can start an entire campaign over with tons of general and specific knowledge of what will happen retained. They can go on the internet and look up guides by the best of in some cases hundreds of thousands of people who play a given game.
Whether turn based or in real time with pause the player has an infinite amount of time, if they so choose, to devise a plan and in the case of most games the same amount of time to execute their decisions while the game AI has to compete with all the other game factions for the limited time between turns that is highly limited because designers don’t want players to get bored waiting. Which is fine and basically unavoidable but you do have to consider it when you complain about or merely discuss how to improve the AI experience in strategy games.
What Does Good AI Even Mean?
Since you and I are timebending higher dimensional gods, I can state here that before you consider the details of making an AI good you probably want to consider what you even want to make the AI good at without, it being contradictory. Even within the same game, much less subgenre or among strategy games over all, different people don’t evaluate the AI the same way. Some players want it to try to crush them the same was a player in a competitive multiplayer game would. Others want it to challenge them but ultimately lose, and a third group plays strategy games more to explore than to dominate.
I like to min-max and speedrun strategy AI. I’ve beaten Star Dynasties in 3 turns with only a little savescumming. But I personally am mostly in the 3rd group. I don’t like boardgamey strategy and I don’t care too much about the often dreaded “complexity”. I want a game that creates an interesting world with surprising emergent events and I want to feel like an archmage or a warlord of a sneaky stabby boy or a grey eminence. I want verisimilitude in the game.
AI designed to create cool stories and AI designed to replicate a human board game player are pretty hard to reconcile. I won’t say no one could figure it out but I sure can’t. Paradox has something of a conflict in their product because of this issue. It has an additional hurdle of historical hardliners as well. Firaxis somewhat avoids this conflict by not marketing their games as what they are not which other companies fail at. Their AI is still a bit shaky, though.
Antifreeze Mechanics
We finally arrive at our ostensible topic. How do we stop snowballing and blobbing? This is almost as fraught as the previous issues. Do we want to? I enjoy a good blob, especially in games designed for it. Well some games probably do want to stop it. Axioms itself restrains snowballing and blobbing through verisimilitude related mechanics. On the intended map size it would be pretty hard to blob in any case because I want to have 40000 provinces. None of my design is focused on snowballing or blobbing directly.
If we did want to reduce it what should we do? Well lots of things have been tried. Both hard and soft caps on expansion and on stability of large territory. Most boardgamey Civ style games use corruption or happiness or bureaucracy or administration or some other word that papers over a pretty simple equation. More simulationist games are similar but not as detailed as Axioms and try to make at least somewhat organic limiters. Paradox games tend to use softer but not simulation based methods like steadily increasing research costs or caps on military units. Some of them have global penalties for expanding like infamy in Victoria 2 or aggressive expansion in EU4.
In the end the AI can’t truly stand up to a player steeped in the knowledge of the speedrun and boundary pushing communities. Can you win Medieval 2: Total War in 14 turns with just Egyptian peasants? With a few key exploits yes. Can you conquer the whole world in Europa Universalis 4 with Ryukyu or Ulm? With some exploits and incredibly intense micro, sure. A common idea these days is that the game should just give you a win screen once you are obviously dominant, especially in Civ style games, no need to keep going after you have 51% of the world. The problem is you generally actually become unstoppable well before the game code can really effectively detect it.
Player Limitations
In the original plan for Axioms there was a turn timer, which rolled over extra time, such that you were limited in a way inspired by the limits of real world leaders. You didn’t have time to micro every little thing. I eventually decided this was mostly unfeasible. Hard to code properly, pretty easy to work around by determined players, and somewhat unsatisfying especially for turn based players who often like to take their time. Plus how do you handle bathroom breaks, eating, and things coming up in real life elegantly? Blech.
In the current version there are several systems designed around verisimilitude to create logical and interesting constraints. I already have a post on Attention Points. You have a 4-5 digit number, they roll over, and there is a lot of flexibility for decisions. For me personally games like Star Dynasties and Old World are too restrictive. Of course this is partially a requirement for them with much simpler and more abstract simulations. Plenty of people prefer the board gamey style of Old World. I own and enjoyed both of those games.
Mechanical Solutions
An additional system, which I learned a week or so ago Paradox implemented but made a mess of in Stellaris, is the information/intelligence network. You need a node in a province to see basic information. There are roughly 8 tiers and new 8-tier aspects open up at certain levels.
For instance around level 3 you can assign missions targeting Characters and not just Provinces. Surveying and Surveillance are the two missions that get you basic info on those entities respectively. This system ties into Secrets, a system I developed in ~2014 which, with additional aspects like Desires, is discussed on the Axioms Fandom Wiki.
This system also handles deploying resources to *use* the information the basic missions gain you for intrigue and espionage purposes. It also deals with the Secret Location system in provinces and prospecting for resources. All of these are described in previous blog posts.
Interaction
Interactions with land and characters require you to spend attention and resources on gaining relevant information. Once you have the appropriate level however you get all the important information. Axioms, in one of my concessions to abstraction, lets you know if a deal will happen and why or why not. Compare the systems in Crusader Kings vs Star Dynasties. The latter just gives very broadly interpretable color codes while the former details everything.
Similarly for economics and population interactions, like whether the people support you or want to revolt, as long as you have the right infrastructure you can see all the details. Of course Secrets and Desires as well as the plans of other characters, hostile or otherwise, remain hidden unless you seek them out. You don’t know if there is a Secret that would make good blackmail or a desire that you could promise to fulfill to draw someone to your side. That requires more intense and specialized Intrigue investments. You can and will be surprised by things in a way more Civ style games don’t enable.
Managing your characters and populations as well as your economy, with a limited, but generous, Attention Point allocation provides organic and emergent difficulties. You have a lot of options to ameliorate or subvert limitations on your ability to expand or rule. A player should never feel like they are in a straitjacket unless they really, really screwed up. Setbacks should be setbacks not game enders.
Immortality
One thing about strategy game players that they hold over real life potentates is effective immortality. You are a single driving force across all of history with singular and powerful focus. Since games cut out all the boring and distracting points of ruling. Of course you can put limitations on even a player character but that is just another thing players will work to transcend. Meanwhile the AI tries to play based on the individual Consciousness(yes like all capitalized words that is a data class in Axioms) of the character.
Axioms has a unique strategy to apply a pressure to the player that is similar if not identical to the vagaries of real life. The rule of fucking cool(this may or may not actually work on particular players). Taking inspiration from fantasy media Axioms tries to distract you and shift you off course with surprising opportunities and unique gameplay.
Players who find an ancient, abandoned, underground magitech railroad system or network of decaying terraforming devices will ideally be seduced away from careful and optimal play. Perhaps a sealed away elder god or sentient forest will convince you to take up their cause. Maybe like a television serial killer you will be sucked into the game shadowy dealings and crimes putting one over on stodgy and sanctimonious AI rulers.
Or perhaps you will discover that the leader of a massive steppe confederation gained his position by imprisoning a dragon, the holy symbol of his people, under a great mountain, using the unique Axioms magic of characteristic transference to take on iconic aspects of draconic appearance, thus misleading his people into believing he was blssed by the divine.
Maybe you are a bit more mundane and become obsessed after eons of a lonely existence with connecting all of the fantastically large and numerous river systems created by the world generator into a single massive network with canals. This experience is similar in some ways to the madness of real life rulers. Consider the pyramids for example.
Wait! Can I… Can I do that? Maybe, maybe not, but I’ve got to find out!
Or maybe after all your hard work a quirk of fate marries the beautiful princes you spent the last 20 turns trying to court to some schmuck. And you just lose it. I don’t judge. Burn the world.
In any case this post is about the perennial problem of strategy design, mostly regarding 4X or map painter games, which is good old snowballing and the limitations of AI in strategy games. The first issue is the difficulty of the AI. If you get tired of a specific topic I do recommend skipping to the next header. Especially if that section is just review of stuff you’ve already thought about.
Players Will Always Beat The AI
A lot of ink has been spilled on the debate over “cheating AIs”. That is AIs that get the strategy version of a bullet sponge buff. Tons of resources, units summoned out of nowhere, perfect knowledge of the entire map. In Total War the tactical AI gets large morale boosts, or it did back in the day. I don’t recommend cheating AI as a strategy but it seems a little hypocritical for time warping omniscient deities in a vast interconnected network of fellow deities to complain about cheating.
The human player has a monstrous pile of advantages over the game AI, even a cheating one. I considered putting the infamous meme of a massive pile of stuff outside an undersea pineapple here to illustrate the difference between the player and the AI. Players can rewind the game an arbitrary amount of time. They can start an entire campaign over with tons of general and specific knowledge of what will happen retained. They can go on the internet and look up guides by the best of in some cases hundreds of thousands of people who play a given game.
Whether turn based or in real time with pause the player has an infinite amount of time, if they so choose, to devise a plan and in the case of most games the same amount of time to execute their decisions while the game AI has to compete with all the other game factions for the limited time between turns that is highly limited because designers don’t want players to get bored waiting. Which is fine and basically unavoidable but you do have to consider it when you complain about or merely discuss how to improve the AI experience in strategy games.
What Does Good AI Even Mean?
Since you and I are timebending higher dimensional gods, I can state here that before you consider the details of making an AI good you probably want to consider what you even want to make the AI good at without, it being contradictory. Even within the same game, much less subgenre or among strategy games over all, different people don’t evaluate the AI the same way. Some players want it to try to crush them the same was a player in a competitive multiplayer game would. Others want it to challenge them but ultimately lose, and a third group plays strategy games more to explore than to dominate.
I like to min-max and speedrun strategy AI. I’ve beaten Star Dynasties in 3 turns with only a little savescumming. But I personally am mostly in the 3rd group. I don’t like boardgamey strategy and I don’t care too much about the often dreaded “complexity”. I want a game that creates an interesting world with surprising emergent events and I want to feel like an archmage or a warlord of a sneaky stabby boy or a grey eminence. I want verisimilitude in the game.
AI designed to create cool stories and AI designed to replicate a human board game player are pretty hard to reconcile. I won’t say no one could figure it out but I sure can’t. Paradox has something of a conflict in their product because of this issue. It has an additional hurdle of historical hardliners as well. Firaxis somewhat avoids this conflict by not marketing their games as what they are not which other companies fail at. Their AI is still a bit shaky, though.
Antifreeze Mechanics
We finally arrive at our ostensible topic. How do we stop snowballing and blobbing? This is almost as fraught as the previous issues. Do we want to? I enjoy a good blob, especially in games designed for it. Well some games probably do want to stop it. Axioms itself restrains snowballing and blobbing through verisimilitude related mechanics. On the intended map size it would be pretty hard to blob in any case because I want to have 40000 provinces. None of my design is focused on snowballing or blobbing directly.
If we did want to reduce it what should we do? Well lots of things have been tried. Both hard and soft caps on expansion and on stability of large territory. Most boardgamey Civ style games use corruption or happiness or bureaucracy or administration or some other word that papers over a pretty simple equation. More simulationist games are similar but not as detailed as Axioms and try to make at least somewhat organic limiters. Paradox games tend to use softer but not simulation based methods like steadily increasing research costs or caps on military units. Some of them have global penalties for expanding like infamy in Victoria 2 or aggressive expansion in EU4.
In the end the AI can’t truly stand up to a player steeped in the knowledge of the speedrun and boundary pushing communities. Can you win Medieval 2: Total War in 14 turns with just Egyptian peasants? With a few key exploits yes. Can you conquer the whole world in Europa Universalis 4 with Ryukyu or Ulm? With some exploits and incredibly intense micro, sure. A common idea these days is that the game should just give you a win screen once you are obviously dominant, especially in Civ style games, no need to keep going after you have 51% of the world. The problem is you generally actually become unstoppable well before the game code can really effectively detect it.
Player Limitations
In the original plan for Axioms there was a turn timer, which rolled over extra time, such that you were limited in a way inspired by the limits of real world leaders. You didn’t have time to micro every little thing. I eventually decided this was mostly unfeasible. Hard to code properly, pretty easy to work around by determined players, and somewhat unsatisfying especially for turn based players who often like to take their time. Plus how do you handle bathroom breaks, eating, and things coming up in real life elegantly? Blech.
In the current version there are several systems designed around verisimilitude to create logical and interesting constraints. I already have a post on Attention Points. You have a 4-5 digit number, they roll over, and there is a lot of flexibility for decisions. For me personally games like Star Dynasties and Old World are too restrictive. Of course this is partially a requirement for them with much simpler and more abstract simulations. Plenty of people prefer the board gamey style of Old World. I own and enjoyed both of those games.
Mechanical Solutions
An additional system, which I learned a week or so ago Paradox implemented but made a mess of in Stellaris, is the information/intelligence network. You need a node in a province to see basic information. There are roughly 8 tiers and new 8-tier aspects open up at certain levels.
For instance around level 3 you can assign missions targeting Characters and not just Provinces. Surveying and Surveillance are the two missions that get you basic info on those entities respectively. This system ties into Secrets, a system I developed in ~2014 which, with additional aspects like Desires, is discussed on the Axioms Fandom Wiki.
This system also handles deploying resources to *use* the information the basic missions gain you for intrigue and espionage purposes. It also deals with the Secret Location system in provinces and prospecting for resources. All of these are described in previous blog posts.
Interaction
Interactions with land and characters require you to spend attention and resources on gaining relevant information. Once you have the appropriate level however you get all the important information. Axioms, in one of my concessions to abstraction, lets you know if a deal will happen and why or why not. Compare the systems in Crusader Kings vs Star Dynasties. The latter just gives very broadly interpretable color codes while the former details everything.
Similarly for economics and population interactions, like whether the people support you or want to revolt, as long as you have the right infrastructure you can see all the details. Of course Secrets and Desires as well as the plans of other characters, hostile or otherwise, remain hidden unless you seek them out. You don’t know if there is a Secret that would make good blackmail or a desire that you could promise to fulfill to draw someone to your side. That requires more intense and specialized Intrigue investments. You can and will be surprised by things in a way more Civ style games don’t enable.
Managing your characters and populations as well as your economy, with a limited, but generous, Attention Point allocation provides organic and emergent difficulties. You have a lot of options to ameliorate or subvert limitations on your ability to expand or rule. A player should never feel like they are in a straitjacket unless they really, really screwed up. Setbacks should be setbacks not game enders.
Immortality
One thing about strategy game players that they hold over real life potentates is effective immortality. You are a single driving force across all of history with singular and powerful focus. Since games cut out all the boring and distracting points of ruling. Of course you can put limitations on even a player character but that is just another thing players will work to transcend. Meanwhile the AI tries to play based on the individual Consciousness(yes like all capitalized words that is a data class in Axioms) of the character.
Axioms has a unique strategy to apply a pressure to the player that is similar if not identical to the vagaries of real life. The rule of fucking cool(this may or may not actually work on particular players). Taking inspiration from fantasy media Axioms tries to distract you and shift you off course with surprising opportunities and unique gameplay.
Players who find an ancient, abandoned, underground magitech railroad system or network of decaying terraforming devices will ideally be seduced away from careful and optimal play. Perhaps a sealed away elder god or sentient forest will convince you to take up their cause. Maybe like a television serial killer you will be sucked into the game shadowy dealings and crimes putting one over on stodgy and sanctimonious AI rulers.
Or perhaps you will discover that the leader of a massive steppe confederation gained his position by imprisoning a dragon, the holy symbol of his people, under a great mountain, using the unique Axioms magic of characteristic transference to take on iconic aspects of draconic appearance, thus misleading his people into believing he was blssed by the divine.
Maybe you are a bit more mundane and become obsessed after eons of a lonely existence with connecting all of the fantastically large and numerous river systems created by the world generator into a single massive network with canals. This experience is similar in some ways to the madness of real life rulers. Consider the pyramids for example.
Wait! Can I… Can I do that? Maybe, maybe not, but I’ve got to find out!
Or maybe after all your hard work a quirk of fate marries the beautiful princes you spent the last 20 turns trying to court to some schmuck. And you just lose it. I don’t judge. Burn the world.
