Frank Lantz recently interviewed me on Donkeyspace, his excellent Substack, which generally focuses on the current AI boom but, in reality, is about his ongoing work on the human condition. As my responses would be of interest to readers here, I’ve reposted the interview below.

Is there a competitive scene for Civ, with tournaments, ranking, etc? If so, do bots have any role in this scene, either as part of the game or for training/analysis?

There is no true competitive scene for Civ although there are a number of small ladders that do the best they can. It’s not really a game that lends itself well to the satisfying resolution needed for competitive play – the closest I ever saw was a grassroots mode where the winner was determined by the first to capture ANY city on the map, which Civ 4 eventually supported as an official mode. However, the lack of a competitive scene means that there is a smorgasbord of different, generally friendly, sub-communities which focus on things like succession games, democracy games, team games, team democracy games, games-of-the-month, challenge games, and so on. Generally speaking, these communities are trying to make a solitary game more social, even if the games played are technically still single-player. A democracy game, for example, is run by a specific player with a group of citizens who vote on important decisions (and sometimes vote out the current player or divide power amongst a cabinet or switch to a new government style or…). The bots are not of much interest here beyond being a consistent measuring stick to use to measure success.

The one exception I can think of is Sullla’s Civ 4 Survivor series (https://sullla.com/civ4survivorindex.html). He’s a long-running 4X blogger, streamer, and critic (provided critical feedback for both Civ 4 and Old World), and he organizes and streams “tournaments” which pit Civ 4 AI leaders against each other to see which ones perform best under different environments and rulesets. He has now added a fantasy version of the tournament where viewers can bid on different leaders before the games begin and then track their success, as one might do in “real” fantasy sports.

I’m curious about all-human, no-AI Civ. Do you know if it’s usually played as a free-for-all or symmetrically (1v1, 2v2, 3v3, etc)? Is it very different from the single-player game vs bots?

Team games and free-for-alls are both popular. Indeed, I’ve spent a good chunk of my career trying to encourage players to forgo free-for-alls for team games as the latter tends to be a much smoother experience (fewer losers, positives emotions from teamwork, less waiting if the game supports simultaneous turns), but there is some instinctive pull that draws players to free-for-alls like moths to a flame. (It’s the same instinct that causes players to always choose the largest map possible and the maximum number of opponents, often to their own detriment.)

Multiplayer is very different compared to the single-player experience, where there is an unspoken, and often unthought, expectation that the AIs will play “fairly” and not suddenly backstab the human (which players will describe as “crazy” AI) or all gang up on the leader as they approach victory. In contrast, humans don’t have any problem – at least conceptually – with other players backstabbing them or ganging up on the leader. It might annoy them, of course, but because they can put themselves in each others’ shoes, they realize they might have done the same thing. Nobody, however, puts themselves into the shoes of an AI. It doesn’t matter if we understand that the AI is just acting like a human might act; AIs are second-class citizens.

When designing games which use AI, it’s important to remember that there are two types of competitive games – games with two sides and games with more than two sides. Two-sided games are inherently zero-sum and thus require no diplomacy at all – all the AI needs to do to evaluate a move is add the move’s value for itself and the negative of the move’s value for its opponent (does this move help me more or hurt my opponent more or some combination of the two). In contrast, games with multiple sides also involve diplomacy, requiring the AI to evaluate who to target, which can involve social and emotional reasoning for which the AI is not extended the benefit of the doubt when it does something the human doesn’t like.

(Of course, many games are actually on a continuum between these two extremes – most free-for-all Eurogames severely limit how players can impact each other so that diplomacy is of little use. Race for the Galaxy, for example, is often accused of being multiplayer solitaire – although the other humans add noise to the system, and mastery comes from predicting that noise. AI works perfectly well for these types of games as the mechanics themselves hinder diplomacy.)

Human-only free-for-all games of Civilization look a lot different from traditional single-player as there is often a lack of trust between humans, which leads to much more defensive play. In single-player, high-performing humans understand how important it is to push out settlers as fast as possible to found new cities; the AI will rarely punish you for doing so as rushing the human is both hard for AI programmers to execute and would also be a bad experience for the players so has been avoided intentionally. In the rare case where the AI does punish the player, the human has an easy emotional out by just reloading or quickly starting a new game, options not available for second-class players (meaning the AI). In multiplayer, players still try to expand quickly but do so in a high-stress environment where they know that an undefended new city could be a game-ending gift to their opponent.

(Old World, by the way, includes a Competitive AI game mode, which is explicitly for players who understand the subtle issues of an AI trying to win against the human at all costs. Under this setting, the AIs will start to dislike you just for winning, will rush a player for expanding too quickly, and will absolutely gang up against the leader near the end. Making this mode an option players have to turn on protects us from most of the standard prejudices that humans bring to a game with theoretically equal AI opponents.)

It seems likely to me that the 1P vs bots version of Civ is the “actual”, canonical version of the game, and the all-human version is a kind of variant. Does that make sense?

It could be considered the canonical version – Civ 1 was single-player after all, and multiplayer was never supported in the initial release until Civ 4 – although that’s mostly a result of the logistical issues with playing a multiplayer game of Civ. A two-team game of Civ is, in my biased opinion, one of the best strategy multiplayer experiences that most people haven’t tried.

The issue of “infinite city spam” seems to be a constant topic in Civ discussions. This seems like exactly the kind of thing you would need to manage with AI opponents. Was this an issue on the Civs you worked on?

Infinite City Spam has always been an issue for 4X games which allow free settling, and all versions of Civ have tried different limitations to slow it down, from city corruption to exponential maintenance to global happiness to minimum distances between cities. (With Old World, we adopted what has worked for space 4X games since Masters of Orion – fixed city sites.) Allowing the player too much leeway to cram in as many cities as possible onto the map leads to many, many problems, but it’s especially a problem for games which adopt one-unit-per-tile as it reduces the space for maneuvering between cities, turning the map into a permanent traffic jam. The incentive to maximize the number of cities per tiles is another good example of how we intentionally code the AI to play suboptimally by not pushing ICS to an extreme, so taking that option away from the human as well can avoid imbalances between the human and the AI that we don’t want. Further, having well-spaced cities leads to a better general play experience, so there is little reason to sacrifice that just so that one side can get 10% more science or production.

I loved the story about how players learned to exploit the AI’s “land your fleet at the city with the least defenders” rule. I imagine that beating the highest difficulty levels involves finding exploitable weaknesses like this in the AI’s strategy and abusing them, is this true?

These cracks in the AI are probably somewhat akin to finding various speed-running shortcuts in that, after they are discovered, it becomes hard to resist abusing them. (Many of the community-run challenge games will explicitly bar certain types of play that are deemed to be too exploitative.) The AI programmer for Old World, Alex Mantzaris, first got my attention as the player who discovered a code exploit in Civ 3 that minimized corruption as long as you founded your cities in equidistant rings around your capital, which became the dominant way to play until we patched it out (which led to the weird experience that some players missed the fun they had optimizing the equidistant ring puzzle that we had unintentionally created). However, because these strategies often either break the theme or are very unpleasant to execute, we put a high priority on stamping them out in patches so that players don’t optimize the fun out of their games.

How different are the designs of the AI opponents in Offworld Trading Company and Old World from those you made for Civ?

Offworld was quite different from Civ (and Old World) in that the problems that the AI faced (usually determining which investment had the highest probable rate of return) was something that algorithms usually do better than humans, especially since the game ran in real-time. Further, because black market attacks were both limited and anonymous, the AI didn’t need to grapple with the emotional side of diplomacy as a little Mutiny of a Geotherm was a much smaller decision than a declaration of war. Indeed, Offworld largely feels like a real-time Eurogame where the game has intense competition via mostly indirect conflict. If you don’t have a source of water, and I stop selling my water to drive up the price (or use espionage to trigger an artificial shortage), the effect can be devastating, but it doesn’t feel as mean as conquering the cities you founded and named after your kids. Because of the indirect conflict, Offworld actually works best as a free-for-all; indeed, we were never quite able to make a very compelling team mode for the game.

Old World has many of the same design challenges as Civ – the cursed problem of diplomacy, the human having infinite amount of time to min-max everything, the necessity to give the AI an artificial advantage at higher difficulties – but instead of trying to solve these problems by just writing a better AI, we addressed them at a design level, by making the game explicitly asymmetrical. In reality, all single-player 4X games are asymmetrical (the AI is either not able or not allowed to play the game the same way the human does), but players like to pretend that they are symmetrical. That ostensible symmetry leads to a lot of problems; besides the issues with diplomacy that I’ve covered, there tends to be problems with how games start and end. An AI that begins the game with a single settler is extremely vulnerable to a human rushing it early (which is not a strategy we let the AI pursue). At the end of the game, non-transparent victory conditions (like cultural or religious victory) are extremely unsatisfying ways to lose the game (in which a random popup informs you that you just lost to some other nation you might barely even know).

Thus, in Old World, our AIs start the game AHEAD of the players, as established nations with multiple cities, but are also only able to win the game via victory points, a very transparent measurement of their cities and wonders. Ambition victory, which is managed primarily through the dynamic event system and gives the player ten different ambitions to achieve, is only available to the human, so we never had to make compromises about which ambitions were fair or unfair for the AI to pursue. In fact, the event system doesn’t apply to the AI at all (we simulate the per-turn value of events for the AI as they tend to be positive on average) because we didn’t want to limit what events could do. An event might lead to an unexpected peace deal if, for example, your enemy’s heir shares your personal religion, and she has now taken the throne. These types of events highlight how the AI occupies the role of a second-class citizen; a peace deal like in the previous example is perfectly reasonable for a human to get, but they are not appropriate for the AI. How would the human react if told that they are no longer at war with a weaker nation because its AI got a peace event because their leader is besties with someone in your court. A significant number of players would just shelf the game at that point – their nation is the Middle Kingdom, after all, the center of the universe. There is no room for an AI protagonist in a single-player game.

Players often talk about moves in strategy games in terms of “greed” and “punishment”. Do you think this kind of talk is just metaphorical, or do you think there actually is a kind of moral dimension to these moves?

I do think that strategy games can teach us about ourselves, about our strengths and our weaknesses with different types of reasoning. A perfect example is that games can teach us to separate a good decision from a good outcome; I’m sure you appreciate how success at poker requires being able to make that distinction, and it’s hard to imagine an activity that teaches that lesson better than games. I can imagine a parallel universe where Reiner Knizia was born in Republican Rome, and Cato spends his latter years decrying how the youth have stopped playing board games and are now losing their virtue and discipline. There are a bunch of lessons a good game, even an abstract game, maybe especially an abstract game, can teach: the sunk-cost fallacy, the endowment principle, understanding probability, long-term vs. short-term decisions, avoiding tilt, and so on.

We recently played a bunch of the board game Pax Pamir together, a game neither of us had played before, and you were much better than me. Do you have something like an algorithm that you could write down that captures how you think when you encounter a new game and are deciding which moves to make, or are you just intuitively winging it?

Relative to the average gamer, I tend to do pretty well the first few times through a game (and then fall back to the pack), and it usually comes down to figuring out the most likely mechanic that will deliver victory. With Pax Pamir, I felt it was unlikely that any of the three coalitions would gain dominance in our first few games, so victory would come down to whoever got the most of their own pieces on the board, so I placed as many spies and gifts as I could as that seemed the cheapest way to be in the lead. (Tribes, on the other hand, make you a tempting target.) I also realized that the game was NOT actually an engine-builder even though it gave the outward appearance of being one. The strict tableau limit, the fact that placing cards competes with using cards for actions, and the opportunity for your rivals to kill your cards means that one needs to think of cards as temporary, with their placement bonus being more important than their ongoing capabilities. I think many new players assume the game is an engine-builder because it looks like one, but engine-builders require permanence – the whole point of playing a long-term card early is knowing that it will pay off later. When Tom Lehmann designed Race for the Galaxy, he gave himself an early constraint that no card could damage another player’s tableau, as it would lead to a completely different experience at odds with being an ideal engine-builder. Pax Pamir is perhaps that alternate version of Race – Pamir is not a bad engine-building game, it’s a good some-other-sort-of game.

Do you think that it would be possible to make a game-playing AI that played “for fun” the way we do? That was interested and curious, that learned the game over time, that could get bored, angry, distracted, addicted, proud, etc? If so, would that be a third category, beyond the “fun” AIs that are really just opponent-themed game rules and “good” AIs that are attempting to play optimally? Can you think of any games that have done anything like that?

This question raises another question that I wonder about – is there any point interviewing me about machine learning “AI” just because I work on game “AI” as the two fields are so fundamentally different? The big difference is that, to some extent, most ML AI involves some sort of black box, and we’ve discovered that if you try a lot of black boxes and cram an enormous amount of data into them, you’ll eventually get great results. However, one is never really sure WHY the AI is making the choices it does, which means that it can be a useful tool for a game where the rules have zero chance of changing (in other words, go and chess) and where performance can be reasonably evaluated objectively (we only care if the go or chess AI wins, not if the human has a good experience). Both of these vectors are at odds with actual game design work, where iteration is a given and, generally speaking, we want the AI to grasp defeat from the jaws of victory.

Also, before answering the question of an AI playing “for fun”, I am contractually obligated to reference the other line that Sid is well-known for – to paraphrase, we should always ask ourselves who is having the fun, the player or the computer? Further, it doesn’t matter how much internal emotional depth the AI has if that is not made transparent to the player, who will probably just interpret the AI’s mood swings as random chance, or worse. (If we postulate a future world where humans extend the same theory-of-mind to AIs that we extend to one another, perhaps the answer will be different, but I also suspect that if players really wanted this kind of depth in their opponents, then single-player game modes would be a lot less popular.) Thus, I am largely skeptical that a “genuine” emotional AI would make an ideal opponent. In contrast, “fake” emotional AIs (no magical machine learning, just old-fashioned integer math) are quite useful. Since Civ 3, I’ve had AI opponents describe their attitude towards you using a simple enum, from “friendly” to “cautious” to “furious” – levels which have concrete effects on how the AIs play and also transparent inputs that make intuitive sense.

A lot of people are worried about AI destroying civilization (the actual one, not the game.) Are you worried about that? Does your experience designing AIs for games influence how you think about this issue?

I have a hard-to-suppress instinct that if James Cameron hadn’t made a movie about AI-controlled robots attempting to destroy humanity, we wouldn’t be having this discussion. No matter how generous our reading of ChatGPT or other models are, even if we are willing to extend the label of intelligence to them, they don’t have any agency, let alone any needs, memories, or goals. If we don’t prompt them to write our term papers for us, they don’t do anything on their own. So, it’s really a question of what we let AIs control because, similar to the problem with using machine learning for games, the main issue is that these AIs are inherently unpredictable. So, let’s not give AIs autonomous control of heavy weaponry, alright?

I gave a talk on games and meaning at GDC 2023, which is now available on YouTube:

However, I fully scripted the talk ahead of time, so I decided it would be worth taking the time to post the slides online, in three parts to have mercy on your browser.

Conflict between designer intent and player behavior actually predates video games as it is a common issue with historical wargames. One of the basic challenges of wargame design is how to encourage historical play when no one knows what the likely or even plausible alternate results could be – for example, could the South have ever won the Civil War? – and how to recreate the pressures that led to bad decisions even when we all know the outcome. Maybe Napoleon should have known better than to invade Russia, but even my kids know not to start a land war in Asia.

A good example of this problem is Volko Ruhnke’s Labyrinth, a game about the War on Terror that starts in 2001 and pits the US against Al-Qaeda. It must have been a difficult game to design for many reasons, one of which is that many of the US’s historical actions, such as the invasion of Iraq, backfired horribly, which players would presumably want to avoid. However, when you read the rules for the game, it comes across like a Bush administration neo-con fantasy world. The mechanics enable a domino effect of Western support throughout the region. Establishing a democracy in one Middle Eastern country will cause it to spread to its neighbors, just like we were told would happen in Iraq back in 2003. When I first encountered the game, I was a little shocked, because it is easy to assume that the designer had fallen for this neo-con propaganda.

Years later, the designer actually addressed that question head on:

What if you as the designer have a different view of how the world works than the historical contestants? To get your players to behave historically, for your model to work and produce historically plausible outcomes, you have to incentivize your players the way people at the time were incentivized. What was the US administration’s strategy and what did they think would work? I have to allow that to work in the game; otherwise, the US player doesn’t play like the US in the War on Terror… regardless of whether Volko thinks those ideas are correct.

This raises the question of what is the point of a historical game? Is it an attempt to model history? To actually experiment with alternate outcomes? Or is it to help us understand why people made decisions that we find baffling or shocking today? With Labyrinth, Ruhnke is not trying to recreate history; instead, he is trying to recreate the historical mindset of the time. As we’ve covered, the ability of a game to simulate the actual world is extremely limited and, thus, misleading. Labyrinth shows a clear example of an alternative – simulating the pressures, the desires, and the fears of historical actors, which can be much more illuminating. The study of the past is not just about what people did but also about why they did it, and a game can probably do a better job of teaching the question of WHY people made choices they did than any other medium.

Let’s move that question to a different era, the medieval world of Crusader Kings. Now, they could have made the game with a very different focus – maybe how best to conquer the world or what is the fastest way to advance out of the Middle Ages – made it more like Civilization, in other words. Instead, they leaned into empathizing with medieval rulers, putting you under the same pressures that made it difficult to maintain a dynasty and keep you realm from fracturing. Which leads to…

…lots of posts like these. How to kill off your heir? Is it worth it to kill off your brothers? Indeed, what is the best WAY to kill off your brothers? Now that’s an interesting question for players to ask because…

… fratricide was actually official policy during a certain period of Ottoman history. For example, when Selim I assumed the throne in 1512, he quickly executed his two brothers. Mehmed III had 19 of his brothers and half-brothers murdered. Eventually, they began imprisoning their family members instead, which was quite an improvement. So, Crusader Kings certainly meets Ruhnke’s goal for a historical game, to encourage players to behave historically regardless of what the designers at Paradox think, who presumably have neither killed nor imprisoned their own brothers.

It requires creativity for a game to recreate a historical mindset as opposed to being just a historical re-enactment. For example, Europa Universalis includes a Randomized New World mode which makes the game less realistic but instead more true to the human experience of the age, of being an explorer and not knowing what is over the horizon.

There is a similar issue with Victoria, which designer Chris King outlined a number of years ago. Historically, colonies just didn’t pay. They tended to be a net-loss for the controlling countries as it cost more to subjugate, manage, and defend than it brought back home to the colonizer. However, a game about 19th Century Europe in which those nations did not race for colonies, even if to the detriment of literally everyone in the world, would clearly put players in an ahistorical mindset. In Victoria, they solved this problem by making colonies a necessary part of their economic model, providing the raw goods and, eventually, markets that the player’s factories and businesses needed. It’s not a realistic model of history, but it is a realistic model of what 19th century Europeans thought was important to them at the time.

An important consideration here is whether players will understand the line between the game incentivizing historical behavior and the game espousing a world view.

In Cole Wehrle’s An Infamous Traffic, you play English aristocrats who are profiting from selling opium in China during the 19th century. You are playing bad people doing bad things, and the designer underlines this by how victory works. The money you earn in China gets converted into frivolous prizes back home in London, including my favorite, a Fancy Hat. The game is telling you that it knows you are doing bad things, and you shouldn’t feel good about the fancy hat you ended up with as your prize. A game’s framing can matter a great deal to help players separate a historical mindset from actual reality.

What if, at the end of Civilization, instead of telling you how much land you conquered, the wrap-up screen told you how many cultures you destroyed and how many languages have disappeared?

I started my career very passionate about making games about history, games that could sit next to a book or a documentary as a legitimate secondary source on world history. My history thesis paper in college was a game that simulated the life of a shopkeeper living in early modern Oxford, based on my research in the Oxfordshire Archives. It didn’t just model the change in prices over time but also your standing in the local Mercers and Grocers Guild as well as your family life as the line between personal and business affairs was not as fixed back then, so you have to worry about your children and their future.

Two years later, I started working on Civilization III, and I was super pumped. I had just read Guns, Germs, and Steel and was inspired to make the videogame version. However, I quickly ran into the goofy reality of video games, of how players will twist your games into what they want to play. I learned that you can’t just put Horses and Wheat and Pigs on one continent but not on another and expect players to be ok with starting in the wrong place.

And then I met the twin evils of Civilization, Infinite City Sprawl, which is how the game encourages cramming cities into every possible spot on the map, and the Eternal China Syndrome, which is how once the initial expansion phase is over, the game becomes static and dull. These effects have nothing to do with history and everything to do with players seeing past the historical setting to the game’s inner math.

So, after making Civ 3 and abandoning the idea that I was making a history simulator, I had to ask myself, what could I actually communicate with Civ 4. Why make that game at all? In doing so, I started down a path that led me to why I make strategy games.

Here is the Civ 4 Civics screen. In Civs 1 through 3, you adopted an ideology – Monarchy, Despotism, Communism, Democracy – which came with a bunch of bonuses and penalties, basically judgments on the designer’s part on what those ideologies meant. I dropped that entirely for Civ 4 and implemented a build-your-own-government system where you choose where power lies, the type of economy, how the legal system works, and so on. You could have a Police State with a Free Market or you could have Slavery mixed with Free Speech. Or you could mix Universal Suffrage with State Property. The whole point was to defy ideological labels to get the player to see past them.

If the twentieth century has a single theme, it is that ideology itself is a failure. Dogmatic leaders used ideologies to demonize the “opposition,” which usually meant helping the strong to terrorize the weak. From Nazi death camps to the Soviet gulags to China’s Cultural Revolution to America’s McCarthyism, the twentieth century was full of ideas that gave power to autocratic leaders not afraid to destroy the lives of those who resisted. Much as we hate to admit it, these leaders were often supported by masses of people who believed in the stories these leaders told, the ideologies they espoused.

Demagogues love the labels that ideologies provide because they obscure and dehumanize the opposition; both sides of the Cold War made liberal use of the terms “Communist” and “Capitalist” to define and differentiate each other, even though the United States government has slowly adopted communist programs piecemeal over the last century. Why exactly was the U.S. – a country with social security, medicare, welfare, a minimum wage, labor laws, and trade unions – killing people to keep Communism out of Vietnam? In fact, if you took a typical Red-fearing, union-busting industrialist from 1923 and sent him 100 years into the future and explained to him how America works now, he would probably assume that the Communists won after all!

Labels exist to separate and control people, and I wanted the civics system to encourage people to look past the labels and at the actual choices a society needs to make when governing itself. It was no accident that I attached Mt. Rushmore to Fascism; carving mammoth statues of your country’s leaders into a MOUNTAIN is fascist, even if we do not live under capital-F Fascism. Our own self-labeling as a Capitalist Democracy does not protect us from charges that our country is damaging the world when our policies hurt real people.

I often get asked whether Offworld is a free trade game or an anti-capitalist game or some other statement on the world economy. I don’t think games should make broad statements like “capitalism is good” or “capitalism is bad” – it’s just too simplistic, and as we’ve discussed, injecting a heavy-handed message can so easily go off the rails.

In Offworld, if Iron costs more than Steel, something has gone wrong, and you win by taking advantage of those discrepancies. There is a reason why everything costs what it does. Sometimes, that goes beyond just supply and demand to government policy and cultural factors, but there is always a reason, and if we don’t like what something costs, we should find out why instead of just complaining about it.

More broadly, I made Offworld for the same reason I made Civ 4, to push against dogmatic ideological thinking, the idea that there is one solution to everything. In a good strategy game, the answer to every question – which resource is most important? what should I research first? – must always begin with “Well, it depends…” The point of having to make tough choices and to adapt to the environment is that there is never, ever just one right approach to every situation. Ideologies inevitably lead to a belief that there is one set of solutions to the world’s problems, and I believe that a good strategy game always challenges this type of thinking.

OK, now I am going to tell you something that is definitely in the 10% useful part of this pie chart. We game designers have no idea what we are doing. To be a good designer, you have to abandon the idea that you are anything other than an explorer.

You know how we all feel like we are impostors? Well, what if we are actually right? It’s actually bad to lose this feeling. If you think you have figured out game design, your career is over.

We are not creating ordered systems. We are creating chaos for the player to find order in.

I would extend this quote to say that designers who try to make ordered systems put their intentions, the meaning of their games, at risk because players will break those systems. Players don’t care about the imaginary game you have in your head.

Indeed, player understand our games better than we do. When evaluating design talent, the most important trait I look for is humility. Designers need to be able to hold two contradictory ideas in their heads at all times – to hold true to their design vision even when success is uncertain BUT ALSO to always assume that their vision is meaningless until they see it the hands of real players. They need to have both a big enough ego to follow their own path but a small enough ego to assume that they are probably wrong.

So, why make games at all if the players ultimately take them away from us? Why am I still here, trying to climb up that hill?

Well, what if that infamous Barbados-Grenada match was actually the most interesting game of soccer ever played? Maybe it’s amazing that a soccer team ended up defending both goals?

Maybe it’s amazing that the socialist Landlord’s Game turned into the capitalist Monopoly?

Maybe it’s amazing that Sweatshop created more empathy for sweatshop managers than for the suffering workers?

Maybe it’s even amazing that Spent somehow made people less sympathetic towards the poor.

Strictly speaking, these four examples are all failures, but the way they fail doesn’t at all suggest that games are useless.

Instead, their failures show us just how powerful games can be.

Right now, we are like children playing with fire, we don’t know what we are doing, and we might end up setting the wrong thing ablaze.

In the spirit of humility, I don’t have the answers for how to make sure our games don’t end up conveying a message that is the opposite of what we intended, but I do know of some games that have succeeded better than others, and I bet you do too.

Just remember that simply stating that your game is about X or Y doesn’t make it so. The only people who truly understand your game are your players, so if you want to know what your game means, make sure you ask them.

Thank you for your time.

Part 1, Part 2

I gave a talk on games and meaning at GDC 2023, which is now available on YouTube:

However, I fully scripted the talk ahead of time, so I decided it would be worth taking the time to post the slides online, in three parts to have mercy on your browser.

Besides the question of whether we know what we are doing as designers, what about the question of whether games can teach us anything about our world.

Or, maybe, let’s set the bar lower and see if games can at least teach us anything about sports.

To do that, we need to talk about baseball analyst Voros McCracken.

Who, despite his preposterous name, has no relation to either Zak McCracken or the Alien Mindbenders

Instead, Voros McCracken revolutionized our understanding of baseball with an idea he first published on Usenet in 1999. He called it DIPS, which stands for Defense Independent Pitching Stats.

The basic idea is that while pitchers do have control over balls and strikes, once the batter hits the ball, the results are no longer in their control. In other words, barring a strikeout or a walk, pitchers don’t control how many hits they allow.

This may seem like a fairly simple observation, but baseball is a very old game, and for over a century, everyone had assumed that the opposite was true – that some pitchers were better at getting batters out than others.

The initial response to McCracken’s idea, which threatened to turn our understanding of pitching upside-down, was shock, disbelief, even hostility.

Although Bill James, the patron saint of progressive baseball analysis, was initially skeptical, after doing the research, he determined that McCracken was correct and that he felt “stupid for not having realized this 30 years ago.”

So, why am I talking about DIPS? What does this mean for video games? Well, one part of the appeal of games is that they can theoretically simulate the real world and teach us about it, that we can make choices and see those choices be modelled accurately. But, to use just this one specific example, how could a game written before McCracken’s insight on pitching have any claim to accurately model baseball? The programmers writing these games would absolutely make some pitchers better than others at preventing hits because that was how everyone thought baseball worked before McCracken. And of course, if garbage goes in, garbage comes out. These games could only simulate a faulty understanding of how baseball works.

To underline this point even more, consider this article Bill James wrote in 2015, arguing that baseball managers were using their starting pitchers incorrectly. For decades, teams have used a five-man rotation, meaning that there is a new starting pitcher every fifth day so that each one can pitch at full strength after four days of rest. James argues that teams should instead use a three-man rotation but with much lower pitch counts, relying more on relief pitchers.

Let’s say someone wanted to test this theory with a baseball simulation. Well, even with a sport like baseball that is ideally suited for simulation as it is essentially a turn-based game, there is no way to get good results on a three-man rotation because baseball simulations are written by trying to get their internal numbers to match real-world results, not from some deeper understanding of how baseball actually works which would then produce accurate results. Because no one has tried a three-man rotation in real life, no one knows what would actually happen, how a pitcher would hold up to pitching every three days instead of every five. Game designers would just be guessing.

So, what can games simulate? Strangely, the best example I can think of is a game trying to recreate a situation MUCH more difficult to simulate than baseball, life as a border agent in a totalitarian country. Papers Please succeeds because instead of trying to simulate reality, it is trying to simulate the personal tensions someone in this position might feel.

The game puts you in difficult situations as a border agent processing immigrants who have compelling stories for why they are trying to cross the border. Would you stop a young girl fleeing from abuse just because she doesn’t have all her papers in order? Who will you let in and who will you keep out? What laws will you enforce and what will you turn a blind eye towards?

However, letting people in illegally can lead to citations which carry fines that might lead to your son dying because you don’t have enough medicine.

Is this an accurate simulation? I mean, who knows? But it creates a genuine emotional conflict which we can all relate to – Is there a right thing to do when helping someone in need will hurt your family? Losing your family is a loss condition, so you can’t just perform as a paragon.

Through this tension, Papers Please gives players an understanding of why resistance against an oppressive system is so hard for people with real lives and, thus, why the powerful are able to stay in power. 

So, to put it simply, games can simulate empathy much better than they can simulate reality.

Speaking of which, here’s a classic line on one of game’s most famous simulations: SimCity doesn’t actually simulate a real city. It simulates the inside of Will Wright’s brain.

Except that’s not exactly true. Very crudely, here are the two poles of 20th-century urban planning. Le Corbusier, who was a proponent of top-down, rational city planning, which separated residential, commercial, and industrial areas. In contrast, Jane Jacobs challenged this idea with proposals for mixed-use development which reflected how cities traditionally grew without central planning.

When Will Wright talks about urban planning, he is much more likely to praise Jacobs than Corbusier. Her more contemporary ideas are the ones he would commonly refer to in his sprawling game design talks.

For example, in this interview, when asked about the inspirations for SimCity, the one urban planner he mentions is Jane Jacobs, not Corbusier.

However, Wright was not making a game in the abstract. He was trying to create a whole city on a very real Commodore 64, and the ideas of these two designers required very different types of coding. Jacobs’s mixed-use urbanism, which focused on pedestrian flow, would require agent simulation, which would be much too complex for an 8-bit system. On the other hand, Corbusier’s residential, commercial, and industrial superblocks could be handled by much simpler cellular automata, which is what Wright choose to use. In other words, the limits of the technology determined what type of city SimCity would simulate, regardless of what Will Wright might have actually believed.

So, SimCity ended up with the famous residential/commercial/industrial split that a rationalist planner like Corbusier might admire, and which – it needs to be said – is today considered bad urban design that leads to crime, slums, and general economic and social decline. As an admirer of Jacobs, Wright probably understood this too – so that leaves us with the question, what meaning should we take from the first SimCity if it represents an urban model that the designer himself doesn’t even believe in?

Is this intentional design? Accidental design? Something else?

There is actually a successful city builder based on the type of agent simulation needed to support Jacob’s ideas. Pharaoh doesn’t use districts; instead, its systems are built around little walkers that move around your city and do their jobs, so that the layout of your streets and the adjacency of your buildings actually matters. The game is considered a high-water mark for city builders, and a testament to how choosing the right model can matter.

I’d like to talk about another game, Kent Hudson’s narrative simulation, The Novelist, which explores the story of the title character who has troubling balancing his three biggest priorities – his wife, his work, and his son.

The game presents you with choices over the course of nine chapters, moving you up or down in those three different categories. The inner math is zero-sum so if you gain two points in your marriage, you lose two points between your work and your son.

However, after playtesting, Hudson realized that his game’s meaning was the exact opposite from what he wanted:

My game was telling players: You can’t have it all. Life is zero sum. You can’t win.
I don’t believe that statement to be true, but people were taking a message from the game that I fundamentally disagreed with.

Games can escape the intentions of their designers just so easily.

I think one of the issues games like The Novelist face is that it’s hard to find human meaning in a game with just simple math at its core. Yet, games absolutely can teach us about ourselves. Telltale’s Walking Dead games provide a great example of this by showing you how your choices compare to everyone else’s. If you are one of the 25% of players who killed Stephanie, you might reflect on why you made that choice when so many others didn’t. Maybe the best way for games to be about people is simply to inject more real people into the game.

Let’s talk about another example of designer intent going awry. This is a SPENT, a well-intentioned game that wants to build empathy for the poor by showing players just how difficult their life can be, how they sometimes need to choose between paying the gas bill, repairing their car, and attending their grandfather’s funeral. That’s a bold goal, but is it effective?

One researcher aimed to find out. Here is an article from Psychology Today about an experiment she ran to see how effective SPENT was at increasing empathy for the poor.

She writes:

After I analyzed the results from this study, I was dismayed to find that playing the game had no effect on positive feelings toward the poor. In fact, the game had a negative effect on attitudes among certain participants – including some people who were sympathetic to the poor to begin with.

The problem is agency – when holding the mouse and making the decisions, it’s very natural to assume that the poor have the same agency that you do as the player. Consider this choice right here – should you spend the money to attend your grandfather’s funeral? The problem is that it’s very easy for the player to not spend the money by just hitting the Skip the Memorial button and then end up thinking: Why do these poor people have such a hard time saving their money?!?

One very interesting finding was that the game did produce empathy… when people watched the game instead of playing it. From my perspective, this is a devastating finding because the whole thing we as game designers have been going on and on about for decades is how games are empathy machines because they put you in the shoes of someone else’s life, but here we see the exact opposite effect, and to make it worse, a passive, non-interactive medium is the one that produces empathy instead.

However, maybe things are not so dire. Why, for example. does Papers Please succeed where Spent fails? The answer is actually just game design. Papers Please took the time and energy to give bite to your decisions – either from what happens when you turn away those in need or from how your acts of defiance hurt your family. In Spent, there is no actual cost to pressing the Skip the Memorial button and saving the money, which keeps the player from actually empathizing with the protagonist.

However, even if designers take the time to build out all of the mechanics needed to create real emotional tension, things can still go awry. Consider Sweatshop, a game designed to raise awareness about the hostile labor conditions in modern sweatshops. Indeed, this game earned the honor of being banned from the Apple App Store for its depiction of child labor and unsafe working conditions, which perhaps hit a little too close to home for them.

The game puts you in the role of the sweatshop manager who, in order to meet increasingly unreasonable quota demands from the corporation, has to cut corners by lowering safety standards, hiring children, and pushing workers past their limits. 

This is what Simon Parkin, one of the designers, had this to say about their intentions and the game’s meaning:

Whereas a film documentary might piece together the sweatshop story through footage and anecdote, the game allows players to experience the system from the inside with all its cat’s cradle of pressures and temptations. [A] game can present the system in a more objective manner thereby building a different sort of empathy and understanding.

However, trying to get a message across with interactivity is playing with fire. This is what journalist and game designer Tom Francis said about his experience playing Sweatshop:

At the end of it, I thought, shit, it’s hard to run a sweatshop. Previously, I was like, oh, it’s terrible these conditions in the sweatshop. Now, I’m like, man, you don’t know what pressures they’re under. It’s hard to meet these quotas!

The problem is that the game puts you in the role of the manager, so your empathy is for the pressures he is under instead of the workers. You end up understanding why managers make the compromises they do and why children end up being mutilated.

Now, there are a couple of different ways to look at that. If players are able to step back and think about what they just did, it’s sort of amazing that a game could get you to kill kids to hit your t-shirt quota.

But I think it’s just as likely that, in less obviously baleful situations like a sweatshop, players will always subconsciously identify with whoever they control in a video game. What does that mean for games where you play the king, the queen, the ruler, or – more generally – the status quo, the existing power structure?

Perhaps the most famous example of a designer’s intent being thwarted is Elizabeth Magie’s The Landlord’s Game from 1906. It was designed to shows the negative effects of rampant capitalism, with an alternate set of rules to show how all the players would be better off if they adopted a tax system where rents were paid into the public treasury instead of into the landlords’ pockets.

The original ruleset contain a very interesting passage that lays out the designer’s intentions. Magie points out that players will quickly realize that, under the default, monopolistic ruleset, “one player will own everything on the board.” The Landlord’s Game was Das Kapital made of cardboard and dice. She invented player elimination to prove out the evils of monopolies. Unfortunately for Magie, collecting rents from your properties and pushing your rivals into bankruptcy proved to be a lot more fun than having all the money going to the public treasury, and…

Today the game is known as Monopoly, minus the socialist tax ruleset. The lessons here are subtle – Monopoly absolutely does demonstrate how a capitalist system will concentrate wealth in the hands of the few and impoverish the many, which is what Magie intended after all, but I somehow doubt…

…this is exactly what she had in mind or if players perceive of the game as a critique of capitalism.

Fun is an insidious requirement for a game to be played and, perhaps more importantly, re-played. Games that aren’t much fun tend to just disappear, and we have to grapple with that as designers.

The very nature of a game makes it extremely difficult to express a strong position on an issue. In order to be a game with different potential strategies, Prison Architect has to suggest that rehabilitation and punishment are both equally viable options. The game-shaped box it is in prevents it from picking a side, regardless of what the designers think.

What it can do is show the problems with each path – you can punish prisoners by searching for contraband every day, which means your addicts will go into withdrawal when they can’t get their drugs and act out violently. On the other hand, you can create job training programs, but that lets the prisoners get their hands on screwdrivers and other items that can be turned into weapons. You can have visitation programs but then you’ll discover a pipeline of drugs being smuggled into the prison.

The game is not – and never could be – an accurate simulation of prison because that’s impossible, but it can help players understand the tradeoffs, compromises, and tensions that they may not have considered before playing.

Now let’s talk about Defcon, a game about nuclear holocaust. (We are really hitting the high points, aren’t we?)

An interesting study was conducted on how playing the game affected player’s opinions of nuclear war.

The experiment separated the subjects into two groups, a control group that read articles on the dangers of nuclear war and a treatment group which played Defcon instead. There were significant differences in how these two groups changed their opinion after the experiment. Although the control group became more worried about a nuclear war in the near future, the Defcon players strangely became less concerned. On the other hand, the game players were more pessimistic that they would survive a war. The researchers’ conclusion, based also on qualitative data, was that playing Defcon was more effective at showing players how destructive nuclear war would be so that they then assumed that our governments would be more incentivized to never resort to nuclear war.

However, there is one important wrinkle in the overall results, which are divided up here by high, medium, and low frequency gamers. Note that every single group became more concerned about the threat of nuclear warfare except for one – the high-frequency gamers in the treatment group, meaning the ones who play games the most frequently. The hypothesis is that core gamers quickly saw past the setting and no longer saw a game about nuclear war and instead saw an RTS game with an unusual art style. This highlights a huge challenge for trying to communicate using game design – if you are working within familiar genre constraints, over time, both the game’s setting and meaning will eventually disappear.

A similar finding showed up in a study run by Dr. Stephen Blessing and Elena Sakosky based on a Geoff Engelstein thought experiment about whether players of Incan Gold would change their behavior based on simply changing the setting of the game. Incan Gold is a push-your-luck game where you delve into an ancient temple for gems and artifacts but risk losing it all the farther you go. To see if the setting affected players, they reskinned the game twice – first, as a firefighter game where you rescued victims instead and, second, as an abstract version where you are just playing for points.

The results they found were that players did change their behavior based on the setting, at least at first. This graph shows how often players returned to the base, which means that they have stopped pressing their luck. In the firefighter setting, this means rescuing less victims, while in the abstract version, it simply means scoring less points. In the experiment, the firefighters would push their luck more, taking more risks to save more people. However, and this is the important part, by the fourth game, the results had largely converged and players of all three versions were playing the same way. Players were now seeing past the setting and just optimizing to score the most points, whether they were called gems or victims or just points. Setting can matter, but we need to be aware that players will eventually gravitate to the game’s inner logic and start to ignore the setting. The more the setting and the rules are disconnected, the bigger a problem this becomes.

These four games have very different settings and meanings, with a very different set of messages and emotions for the player. And yet, there are significant parts of these four game that play out the exact same way, could even be built on the same shared codebase. Putting players into an established genre dulls the designer’s intent because, over time, players will stop engaging with the message and meaning of the game and instead just fall back on instinct. They are now playing shooter #34, not a philosophical game about a submerged dystopia or a jaunty adventure with a lovable rogue or a contemporary high-tech military thriller. Instead, players are warped back into their dorm room in 1994 and booting up Doom. Meaning is not a layer built on top of someone else’s game. A game’s meaning starts with its basic building blocks, the core actions that the player is going to be repeating over and over again.

Part 1, Part 3

I gave a talk on games and meaning at GDC 2023, which is now available on YouTube:

However, I fully scripted the talk ahead of time, so I decided it would be worth taking the time to post the slides online, in three parts to have mercy on your browser.

Hi everyone, I’m Soren Johnson, and welcome to You Have No Idea How Hard It Is To Run A Sweatshop

Here are the published games that I’ve worked on over my career, some of which I will touch on over the next hour.

I also do a podcast where I interview game designers about why they make games, and the episodes tend to run long, so check it out if you have any 4-hour road trips coming up.

So, the talk I am about to give is something of a sequel to one I gave in 2010 entitled Theme is Not Meaning. I argued then that we need to stop assuming that a game’s theme or SETTING determines its meaning and, instead, that meaning comes from the mechanics themselves. Since then, I’ve spent a lot of time thinking about games which do a good job of constructing meaning from their mechanics and also about ones which do it poorly. I’d like to talk about some of these examples today and examine how best to build meaning in our games. Games have great power to affect players but, and I’m going to be bringing this up over and over again, that power can so easily slip out of the designer’s hands and sometimes end up delivering the exact opposite message from what the designer intended.

First, let’s talk about soccer and the 1994 Caribbean Cup, where they introduced a unique rule concerning Golden Goals, which are overtime goals that instantly end the game. In this tournament, Golden Goals would now be worth 2 goals instead of 1, which could matter if two teams are tied in the standings and need to use the difference in goals scored as the tiebreaker.

So, here are the standings before the final game between Barbados and Grenada.

Let me explain what we are looking at here. These columns are the number of games played and the wins and losses so far. Note that if Barbados beats Grenada, there will be a three-way tie.

In a three-way tie, each team will have 3 points in the standing because each win is worth 3 points.

So, in the event of a three-way tie, the team which advances is the one with the best Goal Differential, which is Goals For subtracted by Goals Allowed.

Thus, even though Barbados is currently in last place here, if they can beat Granada by two goals, they will advance from their group by having the highest goal differential. Remember when I said earlier that a Golden Goal is worth double? That’s called foreshadowing.

Near the end of the game, Barbados is up by exactly two goals, with a score of 2 to 0, so if the game were to end at that moment, Barbados would have the goal differential they need to win the tiebreaker.

But then, in the 82nd minute, Grenada scores and, even though Barbados is still winning the game, they will not advance because they have lost a point in goal differential. Barbados needs another goal, or they will be eliminated.

So, if Barbados can score a third goal in the final eight minutes of the game, the standings will look like the chart on the right, Barbados will get back to a +1 goal differential, win first place, and advance to the next round.

Five minutes pass, and Barbados is running out of time to score the third goal they need. If nothing changes, they will win the game but be eliminated.

However, someone on the Barbados team remembers the special Golden Goal rule and realizes that if they score on themselves and force the game to overtime, they would have a full thirty minutes to score a Golden Goal which would then be worth two goals and put Barbados into first place as you can see on the right. And so this happens…

Barbados brings the ball back to their own side and intentionally scores on themselves.

Obviously, Grenada figures something is up, and they quickly realize that they now have two very different paths to first-place, represented by both charts here. Either they score a goal to break the tie – you know, the old-fashioned way – OR INSTEAD they intentionally score on themselves to lose the game but win the group by goal differential.

Which leads to what must be the most unusual soccer game of all time because Barbados now needs to defend BOTH of the goals, and Grenada just needs to score in either one, leading to this totally, totally normal arrangement of players.

Ultimately, Grenada fails to score in either goal, and the game goes to overtime. Naturally, Barbados does score a Golden Goal, now worth double, and goes onto the next round. Their plan to score on themselves and win in overtime actually worked.

Here is the response of the Grenada manager to this outcome:

I feel cheated. The person who came up with these rules must be a candidate for a madhouse…. Our players did not even know which direction to attack: our goal or their goal… In football, you are supposed to score against the opponents to win, not for them.

Of course, Barbados did nothing wrong. They are just playing by the rules of the game, and the rules put them in a situation where scoring against themselves made sense. I’ve certainly had the same experience where a rule or mechanic I’ve added to a game ends up creating a perverse incentive that leads to the exact opposite player behavior that I wanted to encourage.

One way to put this is that nobody knows anything.

Here’s the rest of the quote from screenwriter William Goldman – Not one person knows for a certainty what’s going to work. Every time out it’s a guess and, if you’re lucky, an educated one.

Certainly, that can be said for video games. I would never have bet on a ultra-detailed dwarf simulator using ASCII art.

Or on a game with 1 winner and 99 losers.

Or on a turn-based game that plays by itself.

Or on a twin-stick shooter minus one of the sticks.

Much of game design is determined by constraints which are out of the hands of the developers. DOTA 2 and League of Legends both inherited their design from the original Warcraft 3 mod Defense of the Ancients, which had two distinctive mechanics that were just baked into how Warcraft 3 worked – Creep Denial (which means being able to kill your own minions) and Last Hitting (meaning that the experience points go to the last hero to hit the target). Dota 2 kept Creep Denial, and both games kept Last Hitting, which suggests that their existence is a happy accident as no one actually quote-unquote designed them. They were just inherited constraints from a game in a different genre.

So, is game design just sorting through accidents like this to see what works?

Consider the case of Fortnite, which started as a game about building forts while fighting off waves of zombies. Even after Fortnite blew up into the world’s largest battle royale game, these building mechanics stuck around as this weird, inherited vestige of the original game. It took Epic years to finally do the obvious thing and just get rid of them entirely so that the end of a Fortnite death match was no longer this strange, frantic building contest.

Indeed, it’s hard to describe Fortnite as intentional design at all when you draw out the strange path things took. Here is a rough history of the various games and mods that led to Fortnite Battle Royale as we know it today. Arma is a realistic tactical shooter while DayZ is a free-form survival game while PUBG is a 1-vs-100 cage match. Nobody planned this all out – it just happened. Fortnite was more discovered than designed.

So, when we design a game, are we really just throwing darts? Do we have any idea what we are doing?

Another problem with game design is that we often overcorrect from whatever lesson we just learned. With Offworld Trading Company, because the game had random maps, each game started with a short exploration phase where you scan the map and found your colony with only part of the terrain revealed. We learned quickly during Early Access that the multiplayer community didn’t want exploration at all because it added luck, and they wanted NO LUCK at the game’s start, so we added a mode with a fully revealed map for these players.

When we built Old World’s multiplayer, we assumed that we would have a similar multiplayer community which would want us to minimize randomness. So, at great effort we maintained a…

…No Characters mode where the game has literally no characters, no events, no families, nothing that could add randomness to the game. However, almost no one uses this mode. To our surprise, the multiplayer community loves the randomness; they love seeing and adapting to fresh events each game.

The lesson here is that every game finds a unique subset of players from the audience that you think you are targeting and from the ones that you aren’t.

You might have made the same mistake if you had listened to my GDC Offworld Postmortem. None of us really know what we are doing. Remember that at GDC, we are all talking about why we THINK we were successful not why we actually were successful, which is essentially unknowable. Half of what you hear at GDC is BS. Half of the rest doesn’t apply to your game. Most of what remains you can’t implement. The rest might be useful, maybe? Keep in mind, this chart itself might also be BS.

For an example of the type of thing you might have heard at GDC that turns out to be wrong, we’ve been told for years by product managers that people buy cosmetics in free-to-play games so that they can peacock, to show off their skins in front of everyone else. Well, in Valorant, all animations, graphical effects, and sounds are only visible to the purchaser. Your enemies don’t see them. So, it turns out that people will pay for cosmetics even if they are the only ones who can see them.

Part 2, Part 3

At this year’s GDC, I mentored Justin Gary on his Tabletop Summit talk on SolForge Fusion and, afterwards, he invited me onto his excellent game designer podcast Think Like A Game Designer. It’s a bit like my podcast except its 90% tabletop designers (although I will be having more tabletop designers on in the future). As I’m not afraid of going too long, he had to split my recording into two parts, but they serve as a pretty good overview of my career. Take a listen if you’ve got some free time!

Part 1: https://www.thinklikeagamedesigner.com/podcast/2019/11/14/think-like-a-game-designer-49-soren-johnson

Part 2: https://www.thinklikeagamedesigner.com/podcast/2019/11/14/think-like-a-game-designer-50-soren-johnson2

If you follow this blog, you are probably aware that my latest game, Old World, just came out on Steam and GOG, and so the game has been in the media spotlight for the last few weeks. Accordingly, I put together all of my recent appearances if you can’t get enough of my voice!

• Three Moves Ahead
• Waypoint Radio
• Strategy Can Be Fun
• Space Game Junkie

Finally, here is our 3-hour release day stream, which was a lot of fun: