Hey, everyone! This was an article I originally published on Patreon, intending to edit it later depending on the feedback I received. That feedback never came, and since I shut down my page at Patreon, the article became inaccessible. So I'm going to post it here so that it doesn't simply gather virtual dust in an external drive.
These were my thoughts after rationalizing plenty of my design process during Unholy Arts, which eventually finally clicked when I learned about systems theory, and hopefully serves as a framework for people who wish to create emergent gameplay without falling into the trap of just coding whatever makes sense along the way, and instead, make a general plan of how their simulation intends to work. I sincerely think this is an underappreciated line of thought and research for game development, which could be really useful for plenty of projects in the future. Please be patient with the start of the article. The introduction is probably dense, but the most important concepts are not that difficult to understand once you get to them, I think.
Do also note that you will find notes by the form of (*Number) or (*1). Look at the corresponding (*N) note at the bottom of the article for further insight or references.
Anyhow, here's the article:

-------------------------------------

Emergent gameplay has been used in the games industry for decades, to the point of being a core concept to explain why entire genres are fun. The Civilization franchise, along with the whole 4X genre, Will Wright's series SimCity and The Sims, as well as some of the more sophisticated city builders and life simulators exploit the same core ideas; and it has only been relatively recently that AAA action-adventure games have been making conscious efforts to introduce emergent gameplay to breathe life into their worlds.

However, it's likely that many of the developers behind these games didn't use the terms or principled approaches that have been established today(*1), more often than not guided by instincts and a marked vision of what their games ought to be - and although we have great historical examples of success, the fact that we're talking about them is likely a case of survivorship bias, as we can also find plenty of attempts of implementing emergent gameplay where the returns are questionable.

I posit that the industry is still at a very immature stage of understanding emergent gameplay, and there aren't great reasons for it, since we have decades and decades of work in different fields that we should have been learning from. And I know because I'm one of those developers who has been, albeit somewhat successfully, designing guided by instinctive heuristics rather than systemized knowledge. We should start talking about building broader, sturdier bridges towards some academic disciplines that we've sipped some concepts from, and how we could perhaps even build some theoretical frameworks of our own.

What is systems theory?

Systems theory refers to the different analytical frameworks used in different disciplines to study systems of elements with their own characteristics that interact with each other. Because, despite my best efforts, this is a mouthful, but I want everyone to follow along, I'm going to use a practical example. If you're very familiar with all of this, jump to the Wikipedia quote.

Let's take ecology. Everyone has heard, in some form or another, the idea that nature is self-regulating. Vegetation uses soil, water and sunlight to expand as much as possible, herbivores feed on it and try to reproduce as much as possible, and carnivores feed on them and try to reproduce as much as possible.

Core elements and interactions in our model

Now that we have established the elements, their core properties and interactions, we can build mathematical models to predict the evolution of the system as a whole over time.

When one of these groups becomes too large, it influences the growth of at least one of the others, which in turn will eventually constrain their own growth. This is an emergent property of the system, which we should be able to predict by developing the core mathematical model.

The model now represents emergent dynamics as well

You might have two flies buzzing next to your ear. One of them might be saying "What's wrong with these charts!? Where's the fungi, the earth's minerals, the sunlight, the-!?", make it shut up, we're simplifying this for the sake of education. The other fly, however, says "the idea that nature regulates itself is as correct as the idea that self-regulating markets always produce desirable results. It works well on a weekly basis, but eventually the natural development of the system might result in oligopolies or collapse of the ecosystem". Reward that fly, feed it, give it a position in your company, it's got good insights.

Now, the tuning of different numbers in the core properties, base interactions or variables in the system might lead its dynamics to behave in one way or another. There are starting conditions that could result in the system exploding, or imploding, or reaching an equilibrium, or evolving into something different, and these frameworks help us to follow the chain of causation. To bring it down to a concrete example: the population of herbivores may overgrow, without the carnivores catching up in time, up until a moment when they raze the immense majority of the vegetation, and the one factor that stops this dynamic is mass starvation. Because we have a model, following the data to identify when this is happening and why is possible.

From discipline to discipline, many of these tools acquire different connotations. Even though positive and negative feedback loops are mathematically the same in ecology's systems theory and business' systems theory, the latter sees positive feedback loops as something great that creates room for growth, while the former sees them as a risk factor that might provoke severe issues in the ecosystem if they aren't properly balanced with negative feedback loops. Similarly, these tools will acquire severely different connotations when we're trying to craft a different game experience. When we're designing the system, rather than studying it, the model will give us an easier time locating under which conditions this event could take place, and identify the numbers we need to tweak and how.

Now that we're here, I can finally copypaste Wikipedia's definition of systems theory and it will make sense for most people:

> Systems theory is the transdisciplinary study of systems, i.e. cohesive groups of interrelated, interdependent components that can be natural or human-made. Every system has causal boundaries, is influenced by its context, defined by its structure, function and role, and expressed through its relations with other systems. A system is "more than the sum of its parts" by expressing synergy or emergent behavior.

It's likely that a sizable portion of the readers have at least some familiarity with some of these concepts, even if they didn't know about the most abstract terms, because the study of systems has grown increasingly important to understand the contemporary world and to devise how to best interact with it, so one form or another of systemic analysis has taken root in vastly different disciplines: while programmers might have noticed how the core concepts of systems theory closely resembles some parts of object-oriented programming(*2), a significant chunk of the work of sociologists is studying how individual elements (or persons) interact with systems (institutions, society at large), or even how systems interact with other systems.

Now, let's delve into why all of this is relevant for emergent gameplay design. My main study cases will be an old title called Patrician III, and my own game, Unholy Arts, though I will also be making minor approaches to Opus Magnum, Victoria 3 and the two latest 3D Zelda games, as well as calling attention into patterns across different genres.

Patrician III and the scalability of gameplay complexity

> Description of Patrician III

Patrician III is an old, niche game that sits at the odd position of having become a small historical reference of the subgenre of trading games, thus tempting the current owners of the IP to refloat the franchise every few years, either under its original title or as spin-offs such as Port Royale and Rise of Venice, almost always with heavy difficulties to please old fans, and ultimately, verging on the edge of establishing the basis of an easily recognizable, long-lasting franchise, but never quite getting there. Its core mechanic is trading: you buy and sell goods from different markets that both consume and produce, with prices continuously being updated according to offer and demand. Your ventures may go wrong if you buy goods expecting to sell them somewhere else where there should be scarcity, but an opponent sells those types of goods there first, making you choose between holding onto them or looking into a different city where they might be better desired, which might make you lose significant time.

Production and consumption come from each cities' population, which may work on non-represented small scale economic activities, or on player-owned (human or AI) manufactories. If you gain money, you can turn that liquidity into goods, used to obtain physical capital in the form of ships, manufactories or offices to buy and sell more goods and produce them yourself, all of which you use to gain more capital, faster. Lower good prices in a given city makes its population rise in wealth, which will make them demand more expensive goods and thus create more opportunity for profit, although there's always going to be a sizable poor population. I'm going to avoid speaking in detail about special mechanics such as pirates, sieges, plagues, popularity, city halls, feasts, expeditions and the Hanseatic League, all of which interact at one level or several with the core system, but aren't really important for our analysis.

Core elements and interactions for Patrician III's goods systems

All in all, the foundation of anything you want to achieve in the game lies at trading goods. You need to buy and sell to make money, without which you cannot take larger decisions such as getting ships or buildings. Making enough money for a ship or a manufactory might require hundreds to thousands of goods traded. While the game is a sandbox, the starting screen invites you to reach goals such as multiplying the population of all the cities in the game by 4, for which you'll likely have to build roughly over a thousand buildings, thus requiring to make the money of well over a million trades.

The players' curve of diving into systemic games

If you are wondering if this game could be considered a glorified excel spreadsheet, the answer is yes. If you are wondering if anyone ever found this game fun, the answer is surprisingly yes, too. But how is that possible, against all common sense? There are some reasons why anyone would sink a few tens or even hundreds of hours into this game.

- The core mechanics result in emergent dynamics, which results in a prolonged learning curve. As the player learns of or notices different emergent properties of the systems, such as: the advantages of protecting or disrupting the chain of production; the need for long distance travel to distribute region-specific goods; the imbalance in offer and demand provoked by building manufactories and increasing population; and so on, they will begin to approach problems in a different manner, seek solutions to problems they hadn't noticed, and ultimately play more solidly. This generates novelty in how the player interacts with the game.
- Tools to interact with the core mechanics at different scales. Not all of them are immediately available at first, but the game has a lot of automatization options, such as establishing trade routes for convoys, or making your offices at individual cities automatically buy and sell when certain prices are reached, up until certain quantities.
- Ultimately, an advanced player does not look so much at individual interactions as much as they look at the big picture. Whereas the problem to solve at the start of the game was: "Where do I sell these iron tools?", a player with a very large trade network will wonder: "Can I afford the risk of developing cities focused on producing unprofitable building goods, increasing the speed of economic growth and creating a new layer of demand, or could I run into several crises that make me run out of liquidity in the process of making it worth it?", thus jumping from understanding how to take advantage of the most basic interactions to thinking about the emergent properties of the interrelated systems.

We shouldn't advance towards the next step of analysis without remarking the obvious: while this game contains a potential for fun that has earned it a small share of enthusiastic fans, who keep returning to it and resurrecting discussions and old guides, it might as well be considered a small footnote in the history of the industry. There have been relatively few games that attempt to replicate the things it did right, and after a few attempts of failing to one-up it on its hyper specific niche, "trading empires" games have simply moved in different directions. Anyone who'd want to develop some sort of general framework that aimed to predict any game's capacity to captivate players, would need to be capable of explaining such disparities in player experiences. That complete general theory for any and all games might be a bit too ambitious, but I have a few insights that could serve to explain the subfield that is systemic gameplay experience.

Let me start by identifying __general categories of the different stages a player may find themselves in when learning to interact with a systemic game at a conceptual level(*3):

1: The player's first contact with a game's systems are the core interactions between elements, such as [buying and selling goods] in Patrician, [jumping] in a platformer, [ordering an unit to attack] in a 4X.
2: The player sees consequences of those interactions in different contexts, and learns of contextual cues that teach them <X interaction is good when Y>, such as [buying is good when the price is below X], [attacking is good when I reach X attack].
3: The player deduces an abstract rule that will allow them to make useful judgements in more specific contexts they have not yet been encountered, such as [buying is good when the price here is good in relation to other markets], or [attacking is good when there's this positive relation between my stats and the opponent's].
4: The player learns of the interrelationships between different mechanics, which allows them to devise and plan complex plays, such as [buying too much pig iron for my iron tool manufactories reduces the availability of pig iron] -> [competing manufactories will have to pay extra to produce iron tools] -> [there will be somewhat reduced availability of iron tools] -> [the iron tools I'll produce will reach higher prices].
5: *The player learns of the interrelationships between different systems*, which allows them to devise and plan general strategies, spot vulnerabilities and thus minimize losses if something goes wrong. This way, the thought process of a high level Patrician III player might be: [In order to make as much money as fast as possible, I have to supply as much demand as possible as early as possible] + [My ships cannot be everywhere at the same time] -> [I'll use my ships to buy and deposit a general amount of goods in each city anticipating the demand generated during the following weeks] + [I'll have those goods automatically sold when the price is good enough] / [Competitors may try and supply those demands instead of me, making my automated trades never trigger] -> [I can export excess goods to different regions] || [I can lower the prices I want to sell at, at the cost of reducing profit margins] || [I'll have to buy less goods for a while].

To sum it up in short words: the natural curve for players is to learn from specific experiences to later deduce increasingly generalized principles, which might then be interrelated to reach more strategic depths of thought(*4)(Why is all of this important?). Note, however, that this curve isn't linear for a specific game from start to finish, but rather, the players' expertise of different mechanics and abstract principles will advance separately, so a player could be at the latest stage of some areas while their understanding of specific mechanics is still blossoming. Most stages of this curve have been present in a wide variety of genres since the early decades of the industry, including those not popularly identified with emergent gameplay. You'll find that the better designed platformers today tend to guide their players through different stages of learning when introducing new mechanics by presenting challenges with increasingly complex content, and may include extra stages for their more experienced players that demand of them to reach or have reached either levels 4 or 5, depending on how you define mechanics and systems(*5).

And this takes us to an interesting paradox. While the genres more traditionally identified with emergent gameplay (4X, simulation, management) tend to have a far deeper potential for the later stages to be significative, they tend to do a far worse job at guiding the player through these stages.

Consider this structure: [the player unlocks an action or ability -> the only way to leave the current room is by making them use the action in a specified manner against a specified target -> in order to advance to the next room, the player must use the action to solve a relatively simple puzzle -> enemies that can only be defeated through the new action or have a significant weakness that may be exploited by the new action start to appear or become more frequent -> at some point near the end of the level or area, a somewhat more complex puzzle must be solved through the use of the new action] ; Optionally, there might be a considerably more complex related puzzle at some later location as extra content. I bet you can think of action, adventure, platformer, puzzle and even FPS games that have this pattern at the very least at one point in the game, if not several. This structure handholds the player (in the best of ways!) through at least three of the first stages I defined earlier, and the extra challenge invites the player to reach the next stage by themselves.

In contrast, for 4X, simulation or management games it has been sometimes been historically acceptable to either A) Teach the player the controls, and where are the buttons for the fundamental interactions, or with some luck B) Make a not-really-a-tutorial campaign that pushes the player towards different milestones, letting them figure out the complexity generated by emergence on their own, which is what Patrician 3 and some similar titles do. The first solution is straight up deficient, while the latter might easily fail to teach players about the relationships between systems and how to take advantage of them, or even worse: teach them to think about the complexity of the game as a problem that hinders their experience, rather than as a source of fun. It isn't a problem born out of nowhere: we can identify some difficulties in the facts that creating tutorials or guidance for games meant to be played in a sandbox environment leaves too many variables up in the air, or that dropping a comprehensive explanation on complex mechanics can make even an interested reader doze off. For the Patrician franchise, this means that the vast majority of its games' potential to generate fun is gatekept for any player without a particular interest in the intricacies of market dynamics and logistics, and they better had it before they first touched the game, because the game itself isn't sweating in efforts to make them develop that interest.

Let's add two factors, the first being the self-perpetuating problem that players and developers of systemic sandbox strategy games tend to be people naturally inclined to abstract problem solving, who have been educated in the non-written rule that you either learn to play these games by stubbornly trying once and again, or by looking up guides, which used to be found in forums 20 years ago, or more recently, in Youtube. The second factor is the fact that companies, trying to maximize growth, take whatever IP based on complex emergent gameplay they have, want their sequel to grow in players, but are well aware that, given the current development formulas, the majority of new players of the franchise who might have only a mild interest in the game will have a frustrating experience. Therefore, decision makers in the company or development team are incentivized in not letting designers explore positive avenues to further develop complex systems, which have a long learning curve, and thus building on top of what made those previous titles great; and rather make them design laterally, in ways that emergent gameplay doesn't benefit from(*6). Which leads us to the following problems: old fans of the franchise might fail to find the experiences that make them remember the old titles fondly; and, since the opportunity cost of adding lateral elements of gameplay benefits games that do not focus as much on emergent gameplay, the fans of radically different franchises are going to feel far more satisfied, which will ultimately make those get better reception and grow faster in playerbase, while companies learn the doubtful lesson that 4X and systemic simulation and management games have more constrained avenues of growth. To sum it up: for the designers of many emergent games, the most significant problem to solve today is how to painlessly guide new players towards a comprehensive understanding of systemic thought (or systems theory) within your game, the same way that games that rely less on emergent gameplay have learned to guide their players towards learning new mechanics without it turning into a chore.

I think everyone agrees with the premise: Emergent gameplay generates fun through the interactions with interrelated systems. After the previous analysis, I think I have a solid base to extend this premise to: the capacity to do so is constrained by two factors: the systems' potential to generate rich interactions, and the constraints in the player's experience to actually reach those emergent interactions, and derive meaning from them.

What can be done to fix the learning curve in emergent gameplay?

If we agree on the premise that, for game designers of complex emergent gameplay, one of or the most significant problems to solve is to make sure that even unexperienced players of the genre are capable of approaching the gameplay through systemic thought, how do we do that? Let's approach some solutions we've seen implemented in the last decade.

- The games of the Victoria franchise may be jokingly subtitled as "social and economic systems: the game". These games don't shy away from deploying endless not-too-figurative excel spreadsheets of data representing the hundreds of thousands of people inhabiting the hundreds of provinces in the map, representing their income, the goods they produce and consume, their material quality of living and political inclinations, which are the central piece for the player to influence as they grow and reform their country, often through decisions with consequences so indirect that might be pretty difficult to predict for a novice player.

The tutorial of Victoria 2, dense and sleep-inducing as it is, understands pretty well that in order to teach the player about the budget, the effects of heavy taxation and customs on the economy must be mentioned, or that you cannot field an army if your country cannot produce or buy military goods. This is, it refers to the relationships between systems from the very start, because a player unaware of them simply cannot take rational decisions far beyond following a step by step guide they've read online.

While Victoria 3 still proves a complex beast to approach for many of its players, some of its features make it clear that its developers clearly identified that the commercial viability of the game was constrained by its learning curve. Its most significant one is, in my opinion, its system of nested tooltips. If you hover over a farm building, you will see that it uses "arable land", contributes to "urbanization" and uses "infrastructure". What does any of these terms mean? Well, you can learn about it by hovering over them, which invokes new tooltips, referring to new concepts that once again invoke new tooltips. This system is great because it allows players to learn at their own pace, helping them to slowly but surely build an increasingly more complete picture of all the mechanics in the game, and how they relate to each other(*7). Rather than merely using them to teach about game concepts, you can jump between tooltips to see who's working at your rye farm, how large their wages are, what are they spending it on, how literate they are, and so on. Compare this to older titles such as Patrician, where you have to constantly open and close a few different menus in cities, offices and ships to find a few numbers you need to update your trade routes.

- On a very different line, Opus Magnum is a smaller puzzle title that challenges the player to design production lines through the placement of mechanisms capable of transporting, rotating or transforming objects. The core elements of the game are relatively simple; but the plurality of principles that determine how to make the idea you have in your head work and how efficient it'll be might easily spiral out of control - which is great, because it means there's a relatively small requirement to start having fun and a very large potential to continue honing your growth. The level design is coherent with this philosophy: you can make your production lines large, slow, inefficient and ugly and still beat the level, *but* after beating a level, you will be shown the player stats for different marks in time, used space and quantity of used pieces, challenging you to find a better solution or even an optimized one.

On top of that, its difficulty progression does a great job at pacing the increase in complexity, slowly but surely presenting harder problems, progressively pushing the player towards breaking them down into individual systems with their own input, transformations and output, that take care of individual tasks and work together, this is, it handholds (again, I'm using this word as a compliment) the player into using systemic thought to design their own solutions. While its solutions might be too specific to establish a direct translation into very different games, we should take into consideration the merit of choosing boundaries for your game that work well with its learning curve.

- There has been endless discussion on the design of the two latest 3D Zelda titles Breath of the Wild and Tears of the Kingdom, after the first of them radically broke conventions, introduced a large share of innovations in open world design and made a clear bet on emergent gameplay mechanics, to great success both in terms of sales and critics. If you haven't watched it already, I suggest taking a look at the conference by a few of BotW's developers at the GDC from 2017(*8).

These games generate emergent gameplay through its "Physics" and "Chemistry" engines(*8), which define rules by which different "materials" and "elements" interact with each other. Naturally, Nintendo being Nintendo, they made sure that these games had surefire methods to guide the player through the learning curve, for which they used shrines: locations scattered through the map that provided a controlled environment where certain principles of these engines could be taught both in its fundamentals and through practical problems. The players have to be lured to these locations, so as general rules, they should: A) Be easy to find, B) Provide some sort of challenge, and C) Provide a reward. However, not all shrines have the same exact function: some are meant to teach fundamental concepts, others show interrelations between different mechanics and principles, and others are meant to be challenges, rather than stepping stones for the player's development.

This is a solid solution for the challenge of designing a fixed open world meant to be enjoyed through emergent mechanics, but I think there should be some focus put on its limitations in order to improve the formula. It is a somewhat common, if perhaps not too dominant, criticism of these games, that they might turn stale, boring or repetitive for their size. I personally find these opinions conflicting, given that they're usually embraced after having enjoyed the games after tens of hours and trying to explore the vast majority of the map, when simply not aiming for a completionist playthrough would have resulted in a more enjoyable experience; but after giving them enough thought, we might find solutions aimed at managing players' expectations and helping them to better pace their playthroughs. Notice this contradiction in the design of shrines: they serve somewhat different functions in terms of game structure, as mentioned earlier; yet they look the same, and they almost always have the same reward: tokens that may be used to increase health or stamina. This makes it so that players often feel as though they should always be aiming to reach and complete shrines, regardless of their actual state in the learning curve. A player might have a very solid grasp on how sand works and how it affects movement, objects and perhaps its significance in combat; then they find a shrine designed to teach fundamental mechanics of sand. Before entering, the player wants to complete the shrine because health and stamina increases are a core part of progression in these games, but the challenge might prove too simple and even intrinsically unrewarding, and ultimately a drag of time. As a game designer, you don't want this player to feel compelled to complete this shrine if it's far too divorced from their state at the learning curve.

I wouldn't get rid of shrines as a core concept, but as hinted earlier, some aspects should be rethought. If shrines that fulfill different purposes are presented in different manners (such as having different colors or roofs), players will soon learn to identify these and decide whether they're worth their time or not: a player that finds an introductory shrine upon entering a region that features lava might understand that it's a safe introduction to a new element from the chemistry engine, while a player who has already traversed half the area and is fairly familiar with lava will understand that it's safe for their gameplay experience to skip it. Players who have completed the storyline and half the shrines in the map, might decide to extend their gameplay for a few hours if they have an easier way to find and identify the most challenging shrines(*9).

Let's summarize everything so far: 

-As a rule, game designers must keep a model of the player's learning curve, where players are first given a taste of the fundamental mechanics and are encouraged to experiment until they conceptualize general principles, which are later interrelated to understand emergent mechanics and dynamics. 
-Rather than painting a narrow road to make sure they don't miss on anything, they should account for different progression speeds and divergent paths, offering challenges and rewards proportional to the player's development and goals. For this objective, soft handholding, openness in game goals, opportunities to skip, contextual cues that help players identify and categorize different kinds of content that they might prefer are all helpful. This becomes even more important in open world and sandbox games.
- Easy access to game concepts and data, and opportunities for experimentation allow players to try and test their theories on emergent properties and dynamics while minimizing the cost in terms of effort and frustration. 
- When designing campaigns or quests, players should be guided towards positions where they can take advantage of emergent properties, before having to face emergent challenges, in order to teach them to see emergence as a source of fun and achievement, rather than as a source of difficulties and frustration.
- When a large chunk of a game's enjoyment might be found in complex emergent interactions, special effort must be put into making sure that the player reaches later stages of the learning curve and has access to tools that allow them to manage plenty of interacting elements at minimum cost, in order to help them get the most out of the game.

Unholy Arts and designing emergent systems and narratives

Let's delve into a very rich aspect of emergent gameplay that we haven't touched so far: emergent narrative. It is one thing to create a complex set of elements, interactions and rules that allow for deep chain reactions, and thus interesting decision-making, but that is different from emergent game worlds naturally blossoming into interesting stories, like Dwarf Fortress and Rimworld achieve. That was exactly what I wanted to create when I started developing my current game over three years ago (over five years as of time of re-publishing on Itch), and I'd like to share some initial insights that have matured into more solid, proven concepts, as well as potential difficulties that might arise in a project such as this.

Let me provide some background. Unholy Arts is a NSFW game with BDSM themes -of which I'll actively avoid to make explicit references in this article-, which I decide to start developing when I find a few financially successful projects with a production value I can reach with the skills of a lone, fledgling game developer, and which could provide substantially different experiences through NPC AI concepts I had worked with in a discarded prototype. Specifically, my hypothesis back then was as it follows: a lot of the appeal of these games comes from the development of the relationships between the player character and the NPCs, but due to those characters, their dialogues and their actions being completely scripted, they don't have the potential to grow on their own and take original decisions, which would make them feel more alive.

So, how do you make a NPC "grow on their own"? By making them change the way in which they make decisions depending on their experiences. Therefore, the game would require mechanisms for the player and NPCs to interact with the world and each other, resulting in events that influenced the NPCs in some significant way, and therefore, AIs whose behavior evolved in accordance with those events(*10).

Core idea of the model and its interactions

So how should the parameters defining these AI characters be chosen? I parted from the following idea: characteristics that define our behavior, such as what we call 'personality', are the result of not only biological factors but also and mostly conscious and unconscious learning from life experiences, kept through evolution due to the utility of adaptative behaviors that help us better face different circumstances(*11). A very logical choice would be creating parameters that define personality, but I went with personal life values (called 'drives' in game) because I thought they could be more interesting for certain narrative aspects. Naturally, because characters shouldn't behave the same way with everyone else nor in any context, I also chose to represent relationship values and mood.

We define some of the tools we're going to use in Unholy Arts

The personal values I went with are love, pleasure, cooperation, domination, self-improvement and ambition. The proportion in which these values are present in a given character determine whether they're capable or not of selecting to pursue a certain goal, and how likely it is they try to do so. To put some examples: a NPC with high love and pleasure will put more effort into pursuing strong and passionate relationships; a NPC with high pleasure and domination will try and bend other characters to their will; a NPC with high cooperation and domination will attempt to establish a hierarchy where they and their closest ones are on top; a NPC with high cooperation and love will punish those who pursue conflict. Regarding the results of these goals: characters who fail in the pursuit of their ambitions will see a raise in their self-improvement values; characters who successfully manage to stomp on others will learn that imposing their will is a proven method to pursue their goals; characters who find themselves the victims of aggression but are protected by third parties will see cooperation with better eyes. This way, we have a dynamic system that invites characters to engage in events that may or may not go their way, therefore, either their values will be reinforced or they'll go through adaptative changes in their worldview.

And now, let's go back to the most abstract forms of systems theory. You may remember that systems may have positive and negative feedback loops, which were influenced and did influence other parameters in the system, sometimes moving it towards different points of equilibrium. In the example with the ecosystem, one of these feedback loops going out of control could break the entire thing, thus evolving into something different. We can find similar processes in Unholy Arts' system of drives. You may have a scenario where one character's continued, unilateral aggressions are successful, eventually imposing a culture where it's expected of everyone to try and fend off by themselves or just accept what's coming to them; you may have a scenario where some characters' continued aggressions are sometimes thwarted by a third party, tempering the aggressors' initiative and teaching victims to look out for help; you may have a scenario where one some characters' are successful at establishing such fulfilling relationships that most characters' grow to prefer spending time socializing; and most of the time you will have scenarios where most of these dynamics reach a general equilibrium that doesn't completely fall off a certain extreme, but still has the seeds to grow different circumstances over time, eventually breaking that equilibrium. Naturally, since the player is controlling one of these characters, they too have agency to steer these dynamics in a particular direction, and a particular commitment to a certain philosophy will certainly have clear effects, if they're successful during gameplay.

Model finally representing some emergent dynamics

The theory worked out fairly well in practice. While the game ultimately ended its development prematurely due to an over-ambitious story, this system works as intended in the game, ultimately receiving a lot of feedback from plenty of players, sometimes remarking their satisfaction from getting to play something new, entertainment at noticing how NPCs could behave fairly differently from playthrough to playthrough, curiosity to discover how much NPCs could change and wanting to learn how to make them get to a certain point, and sometimes confusion and frustration too. Given what I've learned so far, it is my hope to be able to lay out some general principles and potential pitfalls, so that these types of designs might be replicated in a more streamlined process that removes some of the uncertainty of experimenting.

Ludonarrative consistency

Let's go back to one of the ideas I remarked earlier about the potential limits that an emergent gameplay design may run into: "the capacity [for emergent gameplay to generate fun] is constrained by two factors: the systems' potential to generate rich interactions, and the constraints in the player's experience to actually reach those emergent interactions, and derive meaning from them". I wish someone had drilled that into my skull a long time ago.

Some of the criticisms I received some time ago were related to players finding illogical some of the behaviors of the NPCs, which might be for the most part be broken down into: "NPCs are attacking me too much", and "[Character] is being mean in conversations for no reason", which fell as buckets of cold water because every time I went to check the code, everything was working as intended. They stopped the moment I added flavor text that hinted at a character's challenges, assaults and socialization intentions.

NPCs only choose to challenge or assault another character because they've selected a "mission" that has specific requisites and weights, such as "Humiliate" (The NPC wants to improve their position in the merit rankings), "Weaken enemy" (The NPC wants to stop the progress or aggression of a character they've identified as potentially dangerous) or "Liberate friend" (The NPC's target has forced a third character into an unfavorable position, and the NPC wants to change that), but in the grand scheme of things, the reasons why the AI is taking certain decisions will fly over the heads of most people unless it is explicitly communicated, which is easily solved by adding messages associated with those missions (For "Humiliate": "You've taken too much spotlight lately. Don't you know how fast flies burn?", for "Weaken enemy": "I heard of your misdeeds. I'll punish you before [authority character] has to.", and for "Liberate friend": "Who gave you the right to order others around? I'm going to put you in your place.").

When a NPC initiates socialization (think of something similar to The Sims' conversation system), they similarly have specific missions they want to achieve, but making them immediately disclose their intentions at the very start has nothing to do with how many real life conversations work, so I made two groups of messages: "Friendly" and "Flirty". However. One socialization mission is "Taunt", intended to start soft conflict with a given character, either to tempt them into initiating a fight or to push them into a mood unfavorable for socialization (I guess you could say I coded bullying?), but because no one in their right mind would accept a conversation with a hostile prompt message, NPCs with a "Taunt" mission use "Friendly" flavor text at the start. I expected players would generally learn that they shouldn't always take NPCs' word at face value. I was wrong! So I ended up coding an extra feature that made the player character receive extra insight as a generic conversation advances, with the progression speed depending on their social stats and relationship with other characters, initially displaying generic messages, but ultimately using specific messages for different missions. This way, the player will get these messages when talking to a NPC with a "getAlliance" mission: [ "This character may have friendly intentions."->"This character may want to raise their friendships."->"This character wants to take a relationship to the next level." ], but these when talking to a NPC with a "taunt" mission: [ "This character may have friendly intentions."->"This character has hostile intentions."->"This character is trying to taunt someone." ]. This solution seems to have been successful so far(*12).

When it comes to emergent narratives, helping players to determine causality is just as important as it is in emergent gameplay complexity, to the point that even if you have supposedly great reasons to want to make your game deceive the player, as you may want to do in a systemic game that involves socialization, diplomacy or politics, you must also make the game confess their capacity to lie and offer in-game tools to the player to identify it. This, in fact, should be elevated to an even greater principle: the mechanics and emergent dynamics of your game's systems should be designed to generate ludonarrative consistency, this is, they should work together and be properly communicated to generate the idea that the game's world is a cohesive living environment, in a way that results in a plurality of experiences, both responsive to the player's agency and avoiding excessively cryptic processes. One of the ideas that called me, and likely many others, to the idea of emergent narratives, is the capacity of a game to grow to create tales that surpass the intentions of their creators, but when this is the result of a pitch-black chain of unidentified emergent interactions, no one is going to know that such a tale actually exists.

Rather than blindly building a set of systems that sounds interesting in a vacuum, a more sensible approach might be to define a set of stories you wish your game to be capable of telling, breaking them down to core elements, interactions and systems capable of generating *and communicating* them, and letting it grow from there, thus ensuring that it can actually result in a core minimum narrative experience. While there might be a reasonable temptation to simply code systems as they grow out of control, you must reconcile with the facts that the resources that may be put into a project are never going to be limitless, and that increasing complexity comes with increasing cost to maintain and grow, therefore, efforts must be focused into that which might actually be relevant. In this way, the somewhat esoteric design of Unholy Arts makes sense when the initial mental drawing board already contained some chains of causality where the actions of the player character, or even those of a non-playable character, could ultimately result in slowly transforming the social dynamics of the simulated characters, resulting in a noticeably different experience for the player as the game progressed, rather than shooting blanks in the dark and expecting them to grow into a tree. And for this, it is necessary for a game designer to understand the logical processes in which the different elements and interactions forming a system can be put together to actually grow into an emergent experience.

Closing words and thanks

I think we're at a point where, even though there's a great deal of enterprising and ambitious game developers, not enough is being written or discussed about theory of game design. This results in game design being a far more unpredictable field than it has to be, with many pioneering projects failing to meet expectations and decision-makers at large organizations often choosing to oppose innovation almost on principle, due to the dangers of such black boxes. An inherent risk in trying to become an authoritative voice in a nascent field of thought is getting proven wrong from one day to the next, but I'd like to take that risk if it means to start a discussion that blossoms into pushing game development to new avenues. I'm not going to deny some degree of selfishness, as I ultimately also want to play more games like mine, made by other people.

While it's hard to properly quote all the influences that take one to where they're currently at, especially if you don't keep track of dates, I'm very certain that I should grant a lot of merit to the science communication Youtube channel Primer(*13), and the game design communicator Mark Brown(*14), both of which I recommend to look into to all readers who have gotten here, as well as to the creators and developers of different simulation franchises that peeked my interest in history and different social sciences. As for those who have, in different degrees, accompanied me in this journey, I cannot thank enough the many players of Unholy Arts who supported me either through feedback, paying my bills or both.

References

(*1): Will Wright deserves a special mention here. At a conference twenty years ago, he explained how his team used statistical models while designing SimCity. While he didn't delve in detail into it, one of the images used at that conference might be used as an introduction to different approaches for studying systems. I'd argue that him being so far ahead of the curve in this aspect was a fundamental pillar for him to build such a prolific career.

(*2): While it wasn't specifically devised for it, the idea that object-oriented programming signified a jump towards closely representing general systems theory was noted almost 30 years ago:

https://scholarworks.lib.csusb.edu/cgi/viewcontent.cgi?article=1126&context=jiim

(*3): As differentiated from a hand-eye coordination level. It is one thing to be capable of executing complex, precise plays (hand-eye coordination), and a different one to conceptually understand how too take advantage of different mechanics in the most efficient way. It may be a common mistake to think that a game or genre does not have one of these types of learning curves after taking it for granted.

(*4): In case anyone asks: why does a game need this level of complexity? Complexity is yet another tool to generate challenge and induce feelings of accomplishment, which generate fun and reward, and thus engagement. Not every game needs to be designed with the explicit goal of pushing the player towards becoming a mastermind in global logistical operations, but using this framework might help you identify the level of complexity you want your game to reach, where it's currently headed towards, and better gauge the limit in extra mechanics before you leave the drawing board, rather than removing or underutilizing them when they're already implemented.

(*5): I have limited knowledge of platformers, but some Super Mario titles have been doing this since basically forever. Another popular example where this might be easily identified would be Celeste.

(*6): By designing laterally, I mean expanding the game in directions far apart from their main focus. Examples would be a fishing minigame in a pre-Breath of the Wild 3D Zelda title or treasure hunting missions in Patrician. For the former, exploring very different concepts of gameplay in ways that do not build over each other is the norm, and doesn't really subtract from the overall experience; for the latter, it is a missed opportunity to add spice to gameplay elements that benefit from building on top of them.

(*7): Some of Victoria 3's design on user interface, as explained by the own developers: https://forum.paradoxplaza.com/forum/developer-diary/victoria-3-dev-diary-29-use...

(*8): Breaking Conventions with The Legend of Zelda: Breath of the Wild ,

(*8b): Very short article on the "Chemistry engine", based on the actual conference, in case you want a quick reference: https://venturebeat.com/games/the-legend-of-zelda-breath-of-the-wild-makes-chemi...

(*9): Note that the development teams of these games made a fairly good job at placing the shrines at specific locations at the map to make sure there was some correlation between the conceptual difficulty of their challenges and how long the player had been playing, but ultimately, these techniques can only go so far in an open world map. This is even more true in Tears of the Kingdom, which doesn't heavily incentivize the player to follow a specific route through their first hours of playtime.

(*10): While the idea immediately evokes the concept of what's currently known as self-learning AIs, I didn't think those models were most appropriate to model human behavior in the ways that I intended.

(*11): This way, stress and anxiety aren't intrinsically bad, but tools that evolution has made our species keep due to their past utility to help us face danger and troubles, even if radical changes in our species' environments have turned them fairly maladaptative in most contexts. In this way, it could be argued that emotions and biologically rooted mechanisms that build personality were useful because they're akin to very rudimentary attempts at making rational decisions, when reason as we contemporarily understand it wasn't an accessible tool.

(*12): I didn't link it until very recently, but I remember similarly frustrated complaints about AI leaders in Civilization 5, which had their root in their "Likeliness to be Deceptive" mechanic: https://civilization.fandom.com/wiki/AI_trait_(Civ5)#Meanings_of_the_Traits

(*13): Game Theory (the math field, not game development) communication channel, Primer: The link to their channel breaks for some reason, but you can find him on Youtube as "Primer" or "PrimerBlobs"

(*14): Mark Brown's channel, Game Maker Toolkit: https://www.youtube.com/@GMTK