Kenta Saijo works as a sales representative for Hiyori Printing, at twenty-eight years old he is married with a young son. He lives in Saitama, a city whose urban border brushes against the larger Tokyo metropolitan area. After meeting with a client one day, Kenta sees his wife laughing with a man he has never seen. In a society where slight gestures can carry weight, her casual touch of this stranger’s arm causes him concern. He has two choices he can make: Should he confront his wife? Or cast away his suspicions for fear of disrupting the careful balance of their lives?
So begins a possible life in Game of Possible Lives (2025) by Tomo Kihara, an arcade styled life-game that prompts its players to navigate critical moments in a fictional protagonist’s life, choosing with a joystick tilted left or right whether to end a marriage or live never knowing if your wife is having an affair. As the latest in a long lineage of games simulating life paths – which dates as far back as the 9th century – what differentiates Kihara’s game from its board and digital ancestors is a thoroughly modern application of large-language-model technology for narrative generation, supported by vast quantities of statistical national data. Powered in this fashion, each playthrough of the Game of Possible Lives is a unique live generated fictional life of someone who could plausibly exist. Imbuing the binary A/B decision making with a greater level of relevancy than the flippancy of associated with the arcade medium might otherwise signal.
Yet, complexifying this instinctual need to associate, Kihara’s game opens with a warning. It alerts players to the fact that the game’s AI generated “plausible lives” are reflective of the “biases and injustices of the real-world society,” in this case Japan. Beyond making us aware that characterisations and events portrayed may be distressing, this attentiveness towards the way data and data dependent systems construct reality is a deeply rooted concern for the artist game designer. As our conversation teased out, Kihara’s interest in the life-game as a historic genre goes beyond its ludic possibilities to an active desire to address what and who it determines as hegemonically centralised by a given society.
In this manner, to play Game of Possible Lives is not only to play-god in a seemingly realistic portrayal of a life but also to entangle yourself in an ongoing debate about the ways governments understand populations through raw statistics and to ask how an AI industrial revolution will change that picturing of a populace. A tall order for an often whimsical game that friends might sit around, chat over and drink beer to.
Side Quest: I thought a useful place to begin our discussion today could be gamic ancestry. In a preview video for the Game of Possible Lives, released on your website, you briefly parse a history of board games that have in various ways gamified and distilled life and its living.
These ancestors include the Shengguan Tu, (c. 836) Tang-dynasty promotion games in which players competed to attain the highest bureaucratic rank; the virtue and vice based The Mansion of Happiness (1800) that taught Christian ethics to British children; and the score based Checkered Game of Life (1860) from the US. Also referenced is one video game, Alter Ego (1986) that was developed by the psychologist Dr. Peter J. Favaro and which based its life events on interviews with over one-thousand participants. The game was released by Activision on the Apple II, Commodore 64, Macintosh and MS-DOS.
As you’ve noted, each of these games can be considered a “playable archive of its era’s normative assumptions,” I wondered whether you could expand a little on what studying these games taught you about game making and about the pitfalls of modelling the breadth of life into a limited system?
Tomo Kihara: What fascinates me about these games is that by compressing something as complex as a life into a simple set of rules they end up preserving that era’s certain values in a playable form.
Conversely one of the pitfalls when designing one of these life games is that it requires you to decide what aspects not to include. Sometimes that happens consciously and sometimes subconsciously. Focusing on the absence of certain lives and values can be quite interesting. I was looking at the history of games that simulate life in Japan and around the time that Checkered Game of Life (1860) came out in the U.S. there was a social climbing board game called Shusse Sugoroku that was printed in separate versions for men and women. In the boy’s version – which came out around 1890 – the goal is to enter parliament and to get there you pass through squares like studying in the West, as this was when Japan was massively modernising. Whilst most of the girl’s versions are written about how you can become a good wife to a certain man, (Formanek and Linhart, 2002) which shows how narrowly society viewed gender roles at the time.
Elsewhere, I read that the ratio of snakes to ladders in ancient Indian versions of the game – such as Gyan Chaupar, or Moksha Patam (c. 13th century) – may reflect how each tradition viewed karma at the time. (Bierend, 2015) Jain boards, dense with snakes and short on ladders, suggest a pessimistic world view full of temptation and spiritual pitfalls; the Hindu and Muslim devotional boards, more evenly balanced and feels forgiving.
Within this lineage the biggest reference for Game of Possible Lives would be Alter Ego that came out in 1986 which was later ported to a web version. I played the initial version, and the premise of living like an American guy in around 1986 is really fun but soon you start realising that you have to follow a very WASPy (White-Anglo-Saxon-Protestant) notion of success. Such as going to university and getting married. And yes, although you have the choice not to do such things, life gets really hard if you don’t follow these values. In all my playthroughs, the topic of race never came up. The game quietly assumes you’re an American who faces no day-to-day discrimination, so there’s no way to live as a person of colour or as a sexual minority. Those lives simply aren’t written. It originally shipped in separate male and female versions, but beyond that binary, a lot of lives never get depicted at all.
What is personally interesting for me is that the developers of the recent remake modernised the UI and everything, but they intentionally left the narratives and the story elements within the game untouched. In a post they state that they wanted to preserve how aspects of life were depicted in the 1980s and I think that preserving the game as a historical body of what life was like for some specific type of people in America is quite valuable. (Fabulich, 2023)
I guess that the Game of Possible Lives extends this lineage with additional questions, such as what happens when the role of narrating a life passes from a human to a language model? And once that model is frozen, local, offline and never updated — can it become what those older games became: a playable archive of how the AI of a particular moment told a life from beginning to end?
SQ: Thanks, this brings us nicely to a two-parter question, which concerns the way that the Game of Possible Lives and the possible lives within it have been composed. Foremost in relation to the types and amount of data that you sourced for the project; and the usage of local open-weight language models to generate both plausible individuals and possible life paths, from this sea of information.
Let’s begin with the data itself. A principal source for this was the Japanese National Census 2020, which collected information on one-hundred-and-twenty-six million people living in Japan, including foreign nationals who had been in the country for over three months. (Statistics Bureau of Japan, 2024) And then there’s a compilation of secondary sources that contribute information on wages, life expectancy, divorce, university entry rates, sexual orientation, and so on. Can you walk us through the act of gathering, understanding and figuring out how to use this data?
TK: I use these statistical data as an anchor to make the stories feel more possible, though I should make clear that plausibility here is a matter of fictional texture and real world gravity, they’re not able to reproduce a real person and the game is never trying to be a factual simulator of any kind. I think it’s important to emphasise that first. That said, the data enters the game in two ways, one as seed information for creating a character; and then the other as a reference material for how the character’s life unfolds in real-time.
For the first, the character seed, as you mentioned I use Japan’s national census in which I am also a respondent… so I am in there somewhere. Japan is actually one of the countries that takes its census most seriously. It’s conducted every five years compared to a decade in the UK and the U.S., plus foreign nationals are obliged to respond too if they’ve been in the country for more than three months, regardless of what type of VISA you are on. Socially there’s an expectation to answer the census and the census data is very accessible through the government’s website. They have their own custom API where programmers can use the anonymised data to understand more about what is happening to the country. So it’s something very accessible.
I always wanted to use that data in some kind of game format and so this has been a really good opportunity. At the start of each playthrough a fictional person that could exist in Japan today is assembled from categories in the census, such as place of residence, nationality, sex, age and one of the twenty industries of employment. That assembled person becomes the seed which is then handed to the language model and from there the character’s profile takes shape.
SQ: Do you recognise any gaps in the data available that you would have hoped to include?
TK: Although it is quite robust, I see the census data as being biased from the very start. As philosophers like Michel Foucault and others have argued, the notion of census emerged as a technology for states to govern their populations more efficiently. The moment something becomes data – in most cases – it’s already an index built for authority to exercise control. (Foucault, 2003) So I’m always cautious when using this data because you have to be conscious of the people who are not in the system. One clear example for the Japanese census is that it only asks about sex assigned at birth, which has exactly two options, male or female. There’s no question that captures your gender with any nuance. Whereas countries like Canada do. Game of Possible Lives tries to draw on data about sexual minorities compiled from non-profits and companies, although these surveys are very limited and most of them very focused around the Tokyo area.
There have been lots of gaps in the model and what’s equally problematic is that due to how language models work, lives that are statistically underrepresented tend to be generated less often. Or when they are generated, they tend to produce cliche narratives. Thinking about these absences, the lives the system fails to generate — is also part of the project as well.
SQ: It almost suggests a meta-game paradox. If you were to counter-calibrate the system to better represent someone marginalised in the data then you’d falsify the national statistical perspective, erasing the issue. But not to you enter into the problem of perpetuating the bias.
TK: Yes, that is the tricky part. That’s why the statistical data is a grounding system and why I never claim this is actually a truthful system that presents society as it is.
SQ: Would it be easy to recreate a similar source of data for a different national context? And is there somewhere you’d be most excited to apply the project to next?
TK: I live and exist within the Japanese census statistics and know the context quite well. Conversely I want to be cautious about working with data I have no lived connection to. That being said, if there were experts and on-the-ground collaborators I’d love to think of alternative versions for other countries as well. Although there are a lot of countries where equivalent census data doesn’t exist.
SQ: Turning to how the LLM works in the game, if I have understood correctly the census data informs the construction of the character profile, whilst those secondary sources we mentioned are used to generate the story elements. Could you expand a little on how these data are fed into the model and whether – based on your lived experience and maybe even the experiences of players – the profiles created and story arcs generated within the game feel ‘truthful’?
TK: Because I want to ensure that the model is swappable down the line, I’m intentionally deciding not to fine-tune a model to a certain set of data. Instead the system uses what technically would be called a keyword-based retrieval-generation system (RAG). In the game, let’s say a marriage event is narratively generated then the model will retrieve statistical data of anything that’s related to marriage, divorce rates for instance, and then put it in the context for the generation of what happens next. Whenever a certain context happens, be it a career change or educational milestone, the model will have relevant statistical data to ground things into reality.
As for the question around if the game feels truthful, since this is a game that unfolds differently every single time, I never really have a full picture of what is happening and can only say from my test-plays whether it feels real or not. Sometimes I encounter a life story that feels too surreal, for instance a playthrough where a retired J-pop idol turns into the prime minister of Japan. For me, what feels truthful is when I see a lot of repeated patterns that I see and hear actually happening in real-life around my friends in their thirty-somethings. There are a lot of cases in Japan where a married couple ends up divorcing or not having a good family relationship because one of them gets ordered by their company to work in a different city. (Jameson, 1993) It’s quite common for the entire family to move to that new city and what happens is that, say it is the husband being asked to move, then the wife has to quit her job and jeopardise her career. Noticing that story pattern come up again and again suggests this model is somewhat producing what I can consider a truthful representation of life in Japan in 2025. But again, it’s very hard for me to tell.
SQ: And on the other side of the coin, how has the reaction been to these AI generated lives? Do players of the game report an affinity or empathy with the characters being portrayed and the choices they face? Do the choices they’re forced to make ever put them in difficult positions or have the potential to trigger negative emotions?
TK: The reception at the Mori Art Museum, where Game of Possible Lives was first shown, was much better than I expected. At times there was an hour-long wait for people to try it out and what surprised me most was how popular it became as a group activity. I’d imagined it to be a solo-player experience though it seemed more fun when people huddled around the cabinet debating the life of someone who doesn’t exist. Whether the question was ‘should you marry this person?’ or ‘a friend is inviting you to join their startup?’ it prompted people to discuss their own lives and life values. It allows people to discuss their own lives with fictional life as the medium. One person messaged me later to say they started dating their boyfriend because when they were playing the game she liked how her partner was very actively making choices that support raising children. She was like “this is good husband material.”
SQ: I wish them a long future together.
TK: Many happy anniversaries. As for the question around the potential to trigger negative decisions and emotions, I could have designed the model not to generate these events but a life game that only permitted happy events would be lying about life, and I didn’t want to build a lie. With the museum I had a lot of conversations about safeguarding and what we ended up doing was placing content advisories explaining the possibility to see triggering topics inside the game at the start of play and also on the gallery wall. I also added a dedicated quitting button so people could leave anytime if the game made them feel uncomfortable.
One thing I didn’t want this to become is what you would call an empathy machine, where you pick someone from a minority background and play their life to feel like you’ve understood them. (Sora-Domenjó, 2022) That’s partially why everything is random and you don’t get to choose who you become. It emphasises the fact that the lives we’re born into are never ours to choose.
SQ: Pulling back to the multiplayer framing and how the tabletop nature allows for people to gather around. Many of the life games we talked about are multiplayer and competitive. One person succeeds as the best bureaucrat, etc. Whereas Game of Possible Lives inverts that dynamic towards collaborative multiplayer. Yet at the same time you have the ability once concluding a life story to change one choice. Does that introduce competitive play through comparison?
TK: That’s something I never initially intended, though what you mention was happening in two slightly different ways. At Mori there were three arcades in the space and people were walking around them comparing what was happening between each game. One person might walk around and tell others that the players at the arcade in the centre tried to do a start-up but they failed miserably so you shouldn’t do start-ups. A lot of information was going back and forth. And the aspect of having one chance to relive from a certain point, that’s always a moment where players start fighting over where to relive from and what choice to retake. I feel where the players choose to relive from highlights the values of each person in a weird way.
SQ: As you mentioned, Game of Possible Lives was first shown at the Mori Art Museum where it was part of their Roppongi Crossing triennale programme, which for its 2025 to 2026 edition took the title of What Passes Is Time. We Are Eternal. I’d like to first ask about the context of showing the work within an esteemed institution and whether there were any particular artists you were particularly excited to be featured alongside? And equally I would love to know more about your subsequent displays, which include a hotel lobby. What different interactions or receptions do the spaces engender?
TK: I was generally very nervous and it was an honour. Roppongi Crossing is a triennial focused on spotlighting emerging artists such as Miyu Hosoi – who I think is roughly the same age as me – alongside worldwide famous artists like the duo A.A. Murakami. It motivated me to build the arcade cabinet in a sculptural way, so that it can sit next to other artists who come from a fine art background. I went way over budget but it was worth it in the end to see so many people lining up and playing it.
One thing about choosing the arcade format was that it allows the work to survive outside the exhibition context. It’s currently being shown in the lobby of the Hotel Anteroom Kyoto where you have people coming in who wouldn’t go to museums, play the game and give feedback to you in very different ways. There’s something interesting about the arcade format as a new way for artists to be making these experimental games so that it can live outside an exhibition context.
SQ: Would you ever consider it being released online? To reach more people and to intersect into predominant ways of understanding Japanese culture and Japanese lives?
TK: That’s possible too. I’m experimenting with how you can bundle local language models in a Steam game for machines with good GPUs. Local model is always preferred for many reasons but I guess it’s ok for the online version to use the cloud AI services too if that is explicitly stated.
As for this being used to depict a Japanese sense of life for a wider audience, that might be difficult. I noticed the game’s narrative shifts with the language you play in. Playing in Japanese tends to retain a Japanese sense of life, while playing in English drifts into a more American sense of what life should be. Recent research measured the same drift in Claude’s values across languages. (Matt Kearney et al., 2026) This drift is part of what the game explores, so I’m looking forward to swapping models to examine the differences.
SQ: Finally question, beyond the life games we talked about earlier, are there any games that have been particular sources of inspiration for Game of Possible Lives or for your practice more broadly?
TK: Not really a game game, but for Game of Possible Lives I got inspired by the Veil of Ignorance thought experiment by American philosopher John Rawls. (Rawls, 1971) The idea is that if you had to decide the rules of society without knowing who you would be born as, your gender, class, health, then you’d be forced to choose something fair enough to accept from any position. I encountered this notion in university and although Game of Possible Lives is not a direct implementation it borrows some elements. You could say it’s the opposite because you receive the starting point by lottery but there’s no way to change society whatsoever. Maybe in the future I could create a game where you can suggest any change in society but you still don’t know who you begin as.
I often find influence in board games, because my first job was an internship at a board game studio. The crown for my favourite board game goes to Modern Art (1992) by the legendary Reiner Knizia. He’s the Hayao Miyazaki of board games. Basically, in the game you become an art dealer selling paintings at auction, representing galleries in cities such as London, New York and Tokyo. You try to rip off the other galleries by pretending your artist’s work is the next big thing. Knizia did two important things in this game. One is that he converted several real-world auction mechanisms into a game format, which was groundbreaking at the time. And second, ironically, he did the most contemporary-art thing possible: he built a critique of speculation and hype in the artworld, and of what institutional validation even is, arriving almost accidentally in the same place as theorists like Arthur Danto and George Dickie. (Danto, 1964; Dickie, 1983)
As for digital games, I’m drawn to games where choices carry heavy consequences that are generated new each time. I really enjoy roguelikes that tend to follow the Berlin Interpretation, (Lait et al, 2008) which was this conference of roguelike developers coming together to define what a roguelike actually is. For close to fifteen years I’ve been playing Brogue (2009 - 2018) and only recently managed to carry back the Amulet of Yendor – which is an item in the depths of the dungeon. That’s something special and rewarding after fifteen years of play.
SQ: Would you call Game of Possible Lives a roguelike? Does it fit the conference’s definitions in some ways?
TK: I would never dare to call it a roguelike but it is somewhat inspired by it. I love random environment generation and the notion of an apparatus that makes interesting choices. Previously I used to design puzzle games for a company and it gets boring to play-test the level in Unity for one-hundred times to check that there are no bugs, because you already know the answers. I guess the cool thing about making roguelikes or any LLM game is that every time you play there is something new and you never get bored even if you are the one that is making it.
Upcoming opportunities to play Kihara’s Game of Possible Lives include the competition exhibition Adaptive Breathing taking place at the Silesian Museum in Katowice, Poland, as part of New Media Week 2026 (September 30 – October 4, 2026); and soon after it will be presented at the Games As Art Festival in Tampere, Finland (October 9 – 11, 2026).
Game of Possible Lives was produced by Tomo Kihara with additional development from Ryota Okawa (Polygoose Studio), sound design by Higashi Takumi, English font design by Zin Nagao, editorial support from Kou Shirayone, image model provision thanks to Cody Claus and Brandon Neri, circuit design support by Kaho Abe, cabinet production by Yusuke Kokubu (Yukuido Ltd), plastering by Yawata Sakan and material support from Aica Kogyo Co. Ltd.
Author’s note: this interview was conducted in relation to video documentation of Game of Possible Lives.
Bibliography:
Arthur Danto (1964) ‘The Artworld’ in The Journal of Philosophy, Volume 61, Number 19 (pp. 571-584).
George Dickie (1983) ‘The New Institutional Theory of Art’ in Aesthetics: Proceedings of the 8th Wittgenstein Symposium (pp. 57-64).
Doug Bierend (2015) The Timelessness of Snakes and Ladders. re:form.
Dan Fabulich (2023) About. playalterego.com.
Michel Foucault (2003) Society Must Be Defended: Lectures at the Collège de France, 1975-76 (David Macey, Trans.). Picador.
Susanne Formanek and Sepp Linhart (2002) ‘Playing with filial piety – some remarks on a 19th-century variety of Japanese pictorial sugoroku games’ in Board Game Studies, Volume 5 (pp. 39-64).
Sam Jameson (1993) Japan’s ‘Bachelor Husbands’. Los Angeles Times.
Matt Kearney et al. (2026) Claude’s Values Across Models and Languages. Anthropic.
Jeff Lait et al. (2008) The Berlin Interpretation. RogueBasin.
John Rawls (1971) A Theory of Justice. Harvard University Press.
Carles Sora-Domenjó (2022) ‘Disrupting the “empathy machine”: The power and perils of virtual reality in addressing social issues’ in Frontiers in Psychology, Volume 13.
Statistics Bureau of Japan (2024) Population and Households of Japan 2020: Overview of the Results of the 2020 Population Census.
Video Credits:
‘Kenta Saijo’ playthrough video of Tomo Kihara (2025) Game of Possible Lives. Courtesy of the artist.
Exhibition impression video of Tomo Kihara (2025) Game of Possible Lives in Roppongi Crossing 2025: What Passes Is Time. We Are Eternal (2025 - 2026) at Mori Art Museum. Courtesy of the artist.