Games: Agency as Art (Review 2)
By I.O.
Part of what makes the theory of games so captivating is our eagerness to embrace an optional struggle. On any given day, I might procrastinate eating for a few hours because the effort of making myself a sandwich seems like just a bit too much. Yet, when it comes to games, I willingly engage in unnecessary struggles and am eager for greater difficulty. I relish the arbitrary constraint of dribbling a basketball, even though the game would be easier without that constraint. I prefer opponents who will challenge me, and I resist the urge to cheat, even if it would mean victory. Even if I lose, as long as the challenge is fun and intriguing, I may want to play again to experience the struggle once more. This is the mystery that Games: Agency as Art by philosopher Thi Nguyen seeks to explore.
Games are a complete inversion of typical motivation. The pay-off isn’t the point—the activity is. The constraints and challenges are designed to shape an experience that is difficult but might also be interesting, fun, or even beautiful. When I play a game, I play for that experience.
Which isn’t to say I don’t care about status, social cohesion, or any number of other functions that a game might serve. Nor that I am indifferent to winning; I care quite a bit. In some situations, I am highly competitive and want to dominate for the sake of domination. When my nephew-in-law starts bragging about winning at Super Smash Bros., I want nothing more than to take a baseball bat and slam him right off the stage. Instead, what I am saying—and what Nguyen argues in his book—is that the challenge has a certain valuable quality, separate from the consequences of playing. The arbitrary constraints, goals, and abilities come together to form a desirable experience that is valuable for its own sake.
When one plays for the sake of the experience, Nguyen calls it Striving Play. In such play, one must adopt the goal of winning in order to have the full experience, but there is a distance from the goal. The goal is adopted as a means to having the experience but isn’t itself the true goal. Losing might hurt, and winning might be fun, but the true purpose of Striving Play is the experience, and the struggle. Nguyen focuses most of his book on Striving Play because it is so much more interesting than Achievement Play (playing with the sole purpose of winning), and reveals more about human nature.
For example, if I know that I am playing Twister later this week, I will put zero effort into training because it is a game where failure is the point of the game. When I play board games with my wife, I don’t want to be much better than her because it would suck the fun out of the game for both of us. And who wants to be so good at a drinking game if they never have to drink? Nevertheless, suppose I intentionally fail or do not take the goal of winning seriously. In that case, I will not be able to have the whole experience of being immersed in the game. I have to genuinely adopt the goal of winning even though winning is not the point. Again, in a total inversion of typical motivation, we adopt winning as an end as a means to having a struggle. “In most of life, we justify our goals in terms of their intrinsic value or the valuable things that will follow from them. In games, we justify our goals by showing what kind of activity they will inspire.”
To summarize, when playing a game, we willingly take on challenges and goals someone else has given us. These are taken seriously within the context of the game yet hold little significance outside of it. By adopting the seemingly arbitrary constraints and goals set out for us, we enable ourselves to temporarily step away from our everyday existence and immerse ourselves in an agency designed by someone else.
This dynamic is at the center of Games: Agency as Art and sets up three of the most critical takeaways from the book.
Agency as Art
Agency is a strange word, but you better get used to it because I will use it often during this review. Agency means active and intentional action (e.g., thinking, deciding, doing, planning, or even adopting a goal). It is everything philosophers discuss when discussing free will (regardless of whether you believe in the concept itself). For example, when writing a research paper, you must exercise your agency in various ways, in what Nguyen calls agentic modes: research mode, creative mode, rigorous mode, communicative mode, and a nit-picky proofreading mode. All of these modes require different agentic capabilities to accomplish the task given to you.
Nguyen argues that just as a painting might capture a beautiful sunset, games can capture human agency—not (just) for the observer, but principally and primarily for the player. This isn’t to say that games can’t have other aesthetic qualities, such as beautiful graphics or a great soundtrack, but rather, games have a medium unique to themselves; human agency. As Nguyen says in the book:
“John Dewey suggested that many of the arts are crystallizations of ordinary human experience. Fiction is the crystallization of telling people about what happened; visual arts are the crystallization of looking around and seeing; music is the crystallization of listening. Games, I claim, are the crystallization of practicality. Aesthetic experiences of action are natural and occur outside of games all the time. Fixing a broken car engine, figuring out a math proof, managing a corporation, even getting into a bar fight-each can have its own particular interest and beauty. These include the satisfaction of finding an elegant solution to an administrative problem, of dodging perfectly around an unexpected obstacle. These experiences are wonderful-but in the wild, they are far too rare. Games can concentrate those experiences. When we design games, we can sculpt the shape of the activity to make beautiful action more likely. And games can intensify and refine those aesthetic qualities, just as a painting can intensify and refine the aesthetic qualities we find in the natural sights and sounds of the world.”
Consider Chess. "Doing math, philosophy, and the like, can give rise to aesthetic experiences of calculation, puzzle solving, and glorious leaps of the mind. Chess takes that sort of activity and crystallizes it. Chess offers us a shaped activity particularly fecund in aesthetically rich experiences of the intellect."
Interestingly, rather than an object having an aesthetic quality, it is your own agency that evokes emotions of beauty, harmony, tragedy, hilarity, etc. Nguyen calls it Process Art as opposed to Object Art. Other examples of Process Art might include cooking, dancing, doing math, or even making art. The outputs of these processes may also have their beauty, but we are interested in the process itself, the aesthetic experience of using a specific agency.
My own experience of this sort of crystalization is with Machine Learning. There is a particular pleasure in building a great ML model. You have all these constraints with your data and challenges you must overcome, from mundane coding bugs to dirty data and the omnipresent lack-of-data-you-actually-care-about. And it was fun. Not just in moments, but the whole process of building a model was a great puzzle that begged to be solved. I even dreamed of data tables in my perverse version of the Tetris Effect.
And then, I left the world of Machine Learning to pursue something I was more passionate about, and I lost the aesthetic experience that I had grown accustomed to. While ultimately, it was the right choice, I began to miss the challenge.
But then, one day, I stumbled across a video of some British guy having a sublime experience doing a Sudoku—going so far as describing the particular puzzle as the “universe singing.” He was so passionate and in love with the game that I knew I had to give Sudoku another try after years of forgetting the game even existed. And lo and behold, I enjoyed it—especially Cage Sudoku, which added a constraint beyond normal Sudoku. Then, after a couple of weeks of playing the game, I realized it was the same aesthetic experience as building an ML model. Not entirely the same by any means, but there was something about thinking through the constraints of the puzzle that felt like thinking through the constraints of the data. In both situations, you are trying to control for all the possibilities to ensure only one possible solution, or interpretation, remains. Now if people ask me what data science is like, I tell them it is a bit like Sudoku—you play around with constraints until everything lines up and you are left with only one possible solution.
Nguyen would say that Sudoku is the crystallization of a specific agency that I will call reasoning by constraints. Other games crystallize other types of agency, such as Chess capturing the agency of calculation and leaps of logic, the game Sign crystalizing the agency of stabilizing meaning, and Monopoly crystallizing profit maximization.
Together, "A collection of games can, then, constitute a library of agencies." Recorded bits of human experience ready to be taken up. Not an escape from reality, but each game distills a specific human activity. “One game might be focused on high-speed reflexes, another on calculative look ahead, or on diplomacy and bargaining, or on manipulating alliances and shared incentives.” Games "are a method for inscribing forms of agency into artifactual vessels: for recording them, preserving them, and passing them around." A way to communicate agencies from one human to another. Two people who have never met, separated by geography, time, and language barriers, can play the same Sudoku puzzle and experience the same agentic experience.
I remember the first time I played Pokemon when I was seven and chose my first “Starter Pokemon.” There are many special moments in games, but that moment will always stand out: the crystallization of the experience of choosing your first pet. Other art forms may be able to crystalize the experience of receiving a pet. Still, no other art form can crystalize the experience of choosing your first pet because no other artistic medium uses “agency.” The ability to crystallize choosing, problem-solving, reasoning, and other forms of agency is a feature unique to games.
Leveling Up Your Agency
The idea that games are a medium through which we can share agencies brings up an intriguing possibility: can games expand our agentic capabilities? Nguyen thinks so. "Games can experientially immerse a player in an alternative agency, making that mode of agency more available to the player elsewhere in life. Games can help to build a broader menu of possible ways of being an agent," which can, in turn, expand our freedom and autonomy, sort of like a more realistic version of downloading Kung Fu directly into your brain, as in The Matrix.
Nguyen doesn’t commit to any particular theory of freedom or autonomy but argues that games expand regardless of the theory you adopt. I’ll avoid the philosophy and will instead go straight to what I think are the mechanisms underlying such a psychological change. I find these mechanisms are more plausible if you consider them from a childhood development perspective, but it is important to note that these mechanisms are not unique to children.
For one, a game may introduce you to a new agency you have never experienced before. Perhaps you are clumsy and lack the experience of graceful movement. In such a case, you might take up a sport or game that requires graceful movement, such as gymnastics, to learn to embody that way of moving and learn grace. Or perhaps you are naive and too willing to trust others, so you might pick up Poker to learn to develop your lie-detection abilities.
Second, a game (or perhaps shifting between games) may enable you to be more fluid in your ability to shift between agencies. For example, you may be stuck in philosopher mode, and your friends consider you a jerk who is constantly correcting and arguing with them. You need to learn to shift your agency, to adopt a new mindset and goals. Since games ask you to adopt arbitrary goals for a time that you can throw aside, perhaps games teach one to quickly lay aside a certain agency to adopt one more appropriate when circumstances change. Or consider my friend who watches math videos before playing a game of poker to get himself in the suitable agentic mode. The shifting between agencies is itself a type of agency worth pursuing.
Lastly, a game may teach you to enjoy an agency you hadn’t previously. This aspect is perhaps the most interesting to me, and so it is where I will spend the rest of the section.
In making this argument, Nguyen is not talking about gamification (which he is concerned about, more on that in the next section). Instead, he argues that in embodying a specific agency, we can learn to appreciate the aesthetic quality of that agency. Consider my earlier example of Machine Learning and Sudoku. I didn’t particularly enjoy Sudoku as a kid but learned to love it as an adult because it offered a similar aesthetic experience to Data Science. Now just reverse the order of those events. Could someone learn to love Data Science by playing Sudoku?
Nguyen’s personal example is that he didn’t start out as a very good philosopher. His teacher said he was great with questions but didn’t have the temperament to sit down and think logically through the problem. But then Nguyen started playing Chess. Chess taught him the joy of thinking through a problem carefully and logically. There was a quality to such thinking that he began to get a knack for, and now that he knew how to access that agency and the accompanying aesthetic quality, he was able to apply that same agentic mindset to philosophy, and thus learned to enjoy the part of philosophy he had previously struggled with.
How does such transformation happen? In most of life, our challenges are beyond us. Too complicated, too complex, too vague, and too different from what we are used to, and so not enjoyable. But games are designed to fit our skills and abilities. "In games, the problems can be right-sized for our capacities; our in-game selves can be right-sized for the problems; and the arrangement of self and world can make solving the problems pleasurable, satisfying, interesting, and beautiful." A cognitive fidelity between the game and real life allows enjoyment from one arena to transfer to another. "Games are a teleologically crisper context for action and evaluation,” allowing us to recognize and appreciate the beauty of a specific agency a little easier.
Are games the solution to the intention-action gap (i.e., Akrasia)? I don’t think Nguyen would go that far, but perhaps it is an essential missing part of the equation. If Nguyen is right, games play a crucial role in helping us to overcome the gap between what we want to enjoy and enjoying it. Just as he learned to love philosophy through Chess, we can learn to love the aesthetic experience of a certain form of agency - the agency of the person we want to become.
Here is another way to think about it: how can you learn to love the struggle of physical activity if you never learn to love the struggle of a sport? How can you learn to love the struggle of a good intellectual challenge if you never meet a challenge that is the perfect intellectual size to struggle with and solve? Perhaps games are the key to expanding our ability into realms of challenge we had previously found daunting, tedious, or a nuisance. Perhaps games are one key (among many) to conditioning ourselves into becoming the person we want to become.
Sometimes we dismiss such aspirational goals as “signaling.” People say, “You don’t really want to do X; you just want to be seen as the type of person who does X.” They assume you only value doing X (e.g., writing) because you think it will make you look smarter, more competent, richer, or whatever else. These comments always bothered me - if you want to be seen as the type of person who does something, it is likely because you value those who do that thing. We shouldn’t lose our aspirational values for fear of our vanity. But perhaps the reason many are so dismissive of such values is that we are not very good at helping people to learn to love the struggle. We have not perfected the art of helping someone to embody and love the agency of the person they would like to become, and so the aspiration seems surface level despite its deep roots in our values. But perhaps games can help us bridge the gap.
The idea that aesthetic experiences can shift our values is not a claim unique to games. Art has long been praised for its ability to influence our values. It is why Eliezer Yudkowsky wrote a Harry Potter fanfiction to promote rational thinking, why a dragon-tyrant eating villagers may have inspired millions of dollars in donations, and why the EA forum chose to organize a fiction writing contest. We might also consider the aesthetics of religion and propaganda. Art seems to have the power to shift our values - for good and ill (and more on the latter in a bit).
What does this process look like concretely? If your current approach to reasoning is flighty, would Chess help? If you are prone to overfitting theories, would Sudoku help? I don’t have strong opinions on this. No one argues that reading Dostoyevsky once will make you a good person, and Nguyen isn’t arguing that playing Settlers of Catan will make you a good steward of resources. At the end of the day, this is a book of philosophy, not science or self-improvement. Nguyen merely wants to claim that a game's seemingly arbitrary constraints, goals, and abilities are not so arbitrary and may serve the important function of expanding our agency. I agree and hope that his work inspires others to put his claim that games expand our agency to the test.
Value Capture
Anything that can shift your values has excellent potential for good and ill - whether religion, propaganda, or even games.
Nguyen’s concern isn't that Call of Duty will turn us into mass shooters. Nor is it about addiction. The concern is that our values will be shifted and over-simplified. This shift towards over-simplification is related to Goodhart’s Law but goes one step further. Goodhart’s Law states that once a metric becomes a target, it ceases to be a good metric. As an example, imagine a teacher who feels forced to teach-to-the-test because test scores are the metric by which they will be judged. Initially, test scores might be a good metric for quantifying student success, and therefore teacher success. But as the teacher feels mounting pressure to improve test scores, the teacher shifts the focus of their lessons away from ensuring the children are learning and towards ensuring the children will perform well on the test. The teacher probably knows the metric is wrong-headed, and is angry that their performance is rated on a simple metric that doesn’t capture how well they are doing their job. But they must teach-to-the-test anyways, leading to an overall lower quality of education even while the metrics show improvement.
The key difference between this example and Value Capture is that once the metric is taken away, the teacher will sigh a breath of relief and return to how they had originally taught. The shift in focus was only temporary.
Imagine a different scenario where a school competes to see which classroom can get the highest test scores. In this scenario, the teacher buys into the game and is excited by the metric. They do their best to optimize their students for test-taking, and teach-to-the-test becomes a mantra. In this scenario, even once the competition and metric are taken away, they may develop a taste for it and are excited to continue to optimize for test scores. Maybe they enjoy it so much that they want to apply that same mindset and optimization approach to other areas of their life. Nguyen calls this “Value Capture.” It is not Goodhart’s Law, but rather a consequence of the law wherein not only does the metric lose its value as a metric, but gains value as a terminal goal for the person who adopted the metric.
If Nguyen is right, aiming for a metric is always dangerous because metrics are essentially a type of point system. There's always the danger of not just missing the target by aiming at the wrong thing (Goodhart's Law) but changing what we consider the target as our attention shifts to quantifiable metrics (Value Capture). Scott talks about a related phenomenon in Unpredictable Reward, Predictable Happiness, where he says he needs to think of rewards as “that thing which changes behavioral programs” rather than as some sort of fuzzy concept of “the target” or “good things that I like.” In other words, the metrics we create are not the true target, but a method by which we change our behavioral programs.
Nguyen’s favorite example is Twitter. Twitter, he argues, gamifies communication. Maybe you join the platform to engage with people, change minds, or stay current on the latest news. But once you are on there, Twitter has embedded you in a game of its own design. They gave you a point system of likes, re-tweets, and followers. These metrics are correlated with your goals, so you play the Twitter game. But after a while, you are less interested in changing minds than in dunking on your most hated politicians because that gets you the most likes and retweets. As you pursue the metric, it de-correlates from your initial values.
My favorite example is that of academia. So many start with high aspirational goals of pushing the frontier of knowledge, only to get stuck in a game of maximizing the number of publications, the prestige of the journals they publish in, their h-index, or whatever other metric that the academic system has imposed on them.
He who controls the point system controls you, argues Nguyen. The nuanced values you developed throughout your life, such as the desire to connect with others, find the truth, and contribute to society, are lost to simplified metrics as social media companies gamify communication, academia gamifies knowledge production, and your workplace gamifies production.
Quantification and metrics wouldn't be problematic if we could match our metrics to our goals exactly. But it is a deep truth that what matters most is the least measurable. We try to operationalize love, happiness, truth, utility, or whatever else. But our metric will always be subtly off from what we care about. Perfect operationalization always escapes us. This is always a problem, but Nguyen argues gamification makes it even worse. When we gamify, we will try to optimize for the metric, which all but guarantees that the metric will de-correlate from the original value we cared about. Worse, pursuing the metric will become more fun and rewarding than seeking that original value.
This community, in particular, should be familiar with the difficulty of operationalizing human values as it underlies so much of the fear around AI alignment. How can you capture, code, and quantify human value? Our values are complex and multi-faceted, and capturing that complexity in a log-loss function seems impossible. And since it seems impossible to code our values into a computer, any AI system trying to optimize the value function we give it will inevitably diverge from what we originally wanted it to do. This scenario doesn’t require true intelligence or even consciousness, just a bad log-loss function and the ability to code.
The solution, Nguyen might argue, is that AI needs to learn striving play. That is to say, it needs to learn that any individual game (or task) is embedded in a larger context of many games that are all worth playing, and it need not optimize so thoroughly that the possibility of the next game becomes limited. Rather than intelligence maximization, agential fluidity becomes the goal: an increased emphasis on general and decreased emphasis on intelligence.
Don’t ask me how to do this. I am not offering this up as an actual solution to alignment—that's far above my pay grade. Nguyen doesn’t even mention AI in his book. Instead, I want to use AI alignment as a sympathetic metaphor to illuminate something more controversial and closer to Nguyen’s arguments.
It is ironic that a community so full of people worried about the potential of an AI to pursue a simplified metric with disregard to other values is so full of utilitarians who debate which bullets they are willing to bite for the sake of moral consistency. It’s a good thing normies made the majority of content on the internet and not us—our community is a terrible example for a fledgling AI in need of understanding the dangers of an over-simplified value function!
Most of the concerns people have about utilitarianism, and EA for that matter, essentially reflect the same problems this community has about AI; megalomaniacal pursuit of an overly narrow value function. The root fear that maximizers will pursue a metric and lose sight of other values remains the same.
Perhaps we must take seriously the idea that, as Nguyen fears, we have been Value Captured. We played a few too many trolley problems, trained a few too many machine learning algorithms, and played a few too many video games. Now we’ve got oversimplified metrics and an optimize-everything mindset. Our values have shifted from the rest of the world not because we are more rational but because our rational methods demanded that we have something quantifiable and, therefore, overly simplified. Perhaps we have fulfilled Nguyen's fear that
“we will be drawn to systems, institutions, social practices, and activities that closely resemble games, and we may be tempted to adjust our own goals to make our lives more closely resemble game-play. We will be attracted to whatever systems can give us game-like levels of value clarity in our non-game lives.”
I don’t think I will solve the problem here, and tracing the exact genealogy of this community's morals is a task far beyond a book review. But Nguyen’s concerns about quantification and metrics are worth engaging with partly because of the parallels to fears about AI. There is a genuine concern that we have oversimplified the set of human values and therefore any optimization is as poorly aimed as any unaligned AI.
Should we then adopt striving play as a model for how to approach altruism? That certainly seems to be how many approach the topic. They adopt some altruistic goals while thinking about where to donate, but they don't seem to be trying too hard to maximize on any dimension. Meanwhile, EA seems full of achievement players who are going full out and doing everything in their power to win. Like professional athletes, they dedicate their lives to the struggle to win.
There are certainly trade-offs to both types of play. And despite my sympathy for Nguyen's concerns about metrics and quantification, I'm not yet entirely on board. I will keep his thoughts as an omnipresent concern while trying to maximize the value of my donations.
Conclusion
No book review can truly do justice to all of the fascinating lines of thought in a book like this or capture all the reader's implications when reading. Suffice it to say there is more I wish I could have written about.
While the book occasionally drags on through arguments I did not care about (i.e., the reality of striving play seemed obvious to me), it is nevertheless a good and fun read. Nguyen’s prose is easy to follow and not too academic. Part of what makes the book so fun is something I skipped entirely for this review to keep the length down: Nguyen himself is an encyclopedia of games. One wonders where he finds the time to play so many games while fulfilling his other responsibilities. It is good that he found a wife who seems as obsessed with games as he is—as far as I can tell their relationship seems to consist entirely of competition.
The criticisms of the book that I do have are all a bit unfair. He wrote a book on the philosophy of games, and I desperately want him to expand the project's scope.
I wish he would expand on the nature of aesthetic experience. It is a key term in the book and one I have an intuition for, but I feel the need to push back on. I understand that beauty is an aesthetic experience, but what makes it distinct from fun, schadenfreude, oneness-with-the universe, or any other experience humans are capable of?
Another area I would love for him to build on is the empirical side of his claims. I’m woefully ignorant of whether there is empirical work showing that art truly changes values or merely shifts awareness to a different set of values. And even if we take it for granted that art does change values, how skeptical should we be that games, or even non-gamified metrics, can shift our values? If games expand our agency, shouldn’t we expect gamers to have incredible agential fluidity compared to non-gamers—allowing them to enjoy engaging with a wider set of real-world challenges compared to non-gamers? My own experience runs counter to this hypothesis, but perhaps there is a selection bias going on.
Relatedly, I would have loved some clarity on what the process of value transformation looks like in practice. If it is true that games can expand my agency, then what games should I be playing? How often? How can I tell if it works (i.e., which metrics)? What would a concrete plan of agentic expansion through games look like?
But in the end, this is a book about the philosophy of games, not science, self-help, or even ethics and aesthetics. The questions and criticisms Nguyen focused on may have been suitable for others or the field if not for me.
If you do not have time for the book, I recommend listening to some of the podcasts he has been on. The three I have heard (Ezra Klein, Mindscape, and Complexity) have all been excellent dives into the nuances and dangers of metrics and quantification and the (false) moral clarity they provide. Nguyen doesn’t offer much support or nuance on the benefits of quantification, so how to handle those trade-offs is left to the reader (or listener). Still, it is a criticism to add to the list when thinking about metrics.
Overall, Games: Agency as Art is an excellent read that has changed my perspective and given me a new way of thinking about and approaching many topics, including things not mentioned in this review. It may only be for some, and some will be so opposed to some of his arguments that they may struggle with the book. Nevertheless, I would love to see more engagement with these ideas, and I recommend the book to anyone interested in any of the concepts discussed in this review.