Latest / The Killscreen Podcast / He Fed a Classic Anthropology Text To Make An AI Game. Here's What Happened.

Transcript

He Fed a Classic Anthropology Text To Make An AI Game. Here's What Happened. · The Killscreen Podcast

• Here is the verbatim transcript of the conversation between Jamin Warren and Michael Hoffman. Jamin Warren: All right, lovely. You can see me okay? Michael Hoffman: Yeah, I can see you. I can hear you. Jamin Warren: Okay. All right, great. Let’s get started. I appreciate you making time to... appreciate you making time to chat. It’s, um, it’s... it was really fun like going through all of the... the papers and the work and everything. So, um... Michael Hoffman: You read them all? [Laughs] Jamin Warren: I did! I did actually. Yeah, yeah, yeah. Um, I, you know, I love reading like, um... I love reading... you know, academic papers, but they’re about games, so it’s like something I’m interested in. And so like I’m bringing some context and some knowledge, so it’s not like completely... it’s not like completely, you know... it’s not like completely like out of my... out of my like range of interest or anything like that. So... Michael Hoffman: Yeah. Well, great. No, it’s great to... to be here at your podcast and uh, yeah, I look forward to our talk now. Jamin Warren: Yeah. Well, I think maybe first to get started, like how did you make your way to like Artificial Intelligence as something that you’re like interested as it relates to... like as an anthropologist? Was your background as an anthropologist first and then you made your way to like AI or were you already interested in the technology? And yeah, tell me a bit about like your process to getting to... to where you are. Michael Hoffman: Yeah. Well actually, um, AI is something that like I started very early. Like, you know, when I was like 17... 16, 17, like started to program and like worked on little games, like uh, board games actually. And already at that time like, you know, AI interested... AI in board games. Um, and then later actually started to... to study at university. I started out with as a computer scientist in, um, in Munich. I am uh... uh, I’m right now actually I’m in... in Munich again. Um, and like so I started in Munich and um, and from there later then after the... the undergraduate like I went into um, anthropology um, to study something different. Jamin Warren: Okay. Yeah, I mean it’s interesting also I think with AI, it’s one of these technologies that is um... like familiar for people who play games. You know, it’s um, it’s a new... it’s a new technology I think for... you know maybe people who aren't um... it’s a new technology maybe for people who like don't read a ton of science fiction or their interactions with it are, you know, Clippy or... you know, it’s kind of like... or like maybe chatbots every now and again. But if you play games, I mean AI has been... it has been a... a key part of like game development for... for like quite some... for quite some time. At least that’s... that’s been my experience. Michael Hoffman: Yes, yes, yeah. Since... since... since the early days of AI like... I mean gaming was always a... a central part and a lot of innovations came out of like uh... um, from... from the gaming side like um... Jamin Warren: Yeah, yeah. Um, well I want to talk about your work... maybe this is uh, you know, a little bit... a little bit silly, but like when you’re explaining like what an anthropologist like does to someone who’s not familiar with the field... Michael Hoffman: Sure, yeah. Jamin Warren: What is the... you know, what is the like the dinner party definition that you give for like the... the work of anthropology? I think that’s helpful for kind of grounding... grounding the conversation we’ll have. Michael Hoffman: Right. So... so an anthropologist actually like goes... like usually like you... you go uh, to another setting. It can be also at home, like to another environment. Usually it’s like further away like um, and you dive into a culture. And um, for a period of at least like 12 months I would say, like to get like a whole kind of yearly cycle. And then like uh, you look at like a specific topic. Like for example in... in my case, like just to... to make this a little bit more uh... uh, give you a bit more flesh, like I went to Nepal, Western Nepal. I lived in like a uh, town called Tikapur. Um, it’s like far Western Nepal in the lowlands. And I looked like um... I studied the lives and mainly like politics and economics of um... uh, former debt-bonded laborers. Jamin Warren: Okay. Michael Hoffman: Um, yeah. Which... which were all of them uh, are part of a ethnic group, the Tharu group in that case. So it’s quite interesting like their cosmology, but like also like, you know, everyday politics, everyday labor, everyday work. Like how... how it’s... how is the life there? And um, and then describe like um, usually you describe like these systems and it’s a sort of like translation process as well. Um, into uh, yeah, a readership then like monographs are usually nowadays in English. Um, and yeah, and so you make like this translation process as well. And um, yeah. Jamin Warren: Yeah. And then the work that you do, it’s to... it’s to better understand like a particular culture? Is it to... do you come in with a kind of like a research goal in mind? Like I want to understand... I want to understand this about this group of people? Or is it more open-ended than that? That like you pick a... Michael Hoffman: Well yeah, I mean like uh, you know, usually you have like an agenda what you want to do like, but for example in my case like you always have to play a lit... little bit by the ear. Like I wanted to do something in India on a Special Economic Zones and like I ended up in Nepal and like that was just the uh, Civil War just ended in um, 2006. There... so there was the... the Maoist insurgency from 1996 to 2006. And first time I came to Nepal was like 2008. Um, and um, yeah, so it was just over and it was very, very exciting. Uh, a very interesting place to be at that time. Um, it’s a post-conflict uh, setting. And like so I ended up there and um, yeah, and then you just play it by the ear and like you realize like what’s going on. And um, yeah in my case like then like I... I discovered like uh... uh, yeah all sorts of things which we can, you know, if you want like we can talk more about. Jamin Warren: Yeah, yeah. I’d... yeah, I’d love to hear... so you... you studied bonded labor. What would you... what was the big uh... like the big observation? Like the big... what would you say that... like the big uh... big result? Michael Hoffman: Um, yeah so... so the big observation was kind of like so how did this kind of like context of the... the Maoist insurgency, how does this affect now like the everyday life of like people who were previously debt-bonded laborers on farms? Then moved out like lived kind of like in... in kind of their new lives and like uh, worked in brick kilns or like small-scale industries. And like at the same time this is uh... is a setting where like politics change quite a lot. And so like the question of course is of course like so is... is... are they free? Or is like... do they end up in a sort of like a uh, neo-debt-bonded setting? And like to what extent does ideology play in there? Like this kind of like uh... uh, questions I looked at. Later I looked at like... I did a second study on um, industrial labor, like further... um, further in the East in a town called Nepalgunj. Then went up to um, Pokhara uh, and like looked at like small-scale industries because Nepal doesn't have like... doesn't have like big industries, no? Like big steel plants or something like that. But more kind of like small-scale industries, food industries or like... I spent a lot of time actually like to get into factories is quite complex like to get the access and the trust of people to do a study there. So I also went like to outside like construction sites and then like went into a... a sand mine for example that um... actually um, was also very interesting. Jamin Warren: Hmm. Yeah. Yeah, yeah. Um, yeah. Well I think... Michael Hoffman: Yeah. No, I was just saying so basically as an anthropologist you describe kind of like the... the life worlds of like other groups. And um, yeah, that’s basically the job, I would say. Jamin Warren: Yeah. Yeah, yeah. Well I certainly can understand uh, like the appeal of games uh, as a medium. I mean that... that’s a lot of the... you know, not all games are like this but many of them do um, put you in a... in a... in a different place, in a different time and expect you to sort of like make your way there. Obviously you’re not doing... you know, you’re not doing uh, you know a research project necessarily. Although I guess there have been, as you noted in one of your papers, there is a... a history of like in-game anthropologies, right? Like applying those anthropological tools as some of these like virtual worlds become... become to resemble more like human societies. They’re populated by real people with real... Michael Hoffman: Yeah, yeah, yeah. Yeah, you know, like anthropologists like started to look at games more seriously in like the early 2000s and they first looked at like virtual worlds and... and tried to study these like as an anthropologist. Just like immerse yourself in World of Warcraft or Second Life, such virtual worlds. And then later they looked more like in the 2010 onwards, they looked more like at like the production of video games. Well how... you know... with the exploitation or like how does it work? Like how is kind of like the whole political economy of like... how... how of game production working? And yeah. Jamin Warren: Yeah. Yeah, yeah. Does um... I mean I guess the, you know, for someone uh, who is like maybe scep... for... for a skeptic, they would say, you know, what is the value at doing in-game anthropology? What... what would you say is like what’s the counter to that? Like why is it important that you would look at in-game worlds as a... using an anthropological lens as opposed to something, you know, in the... in the real world? Michael Hoffman: Mm-hmm. Yeah, so my... my take actually uh, is that like anthropologists should make um... uh, video games. Actually like there’s... there’s hardly any anthropologists who... who have done it. Like also because game making is really, really hard. And like making a game that kind of like meets sort of like on the one hand the goal to make game fun, but like also like make... make it... you know, align it to like the cultural standards um, is... is really... is really difficult. Um, but I think like, you know, like it gives you the experience. Not everyone has like the... the fortunate uh, to go to... to another place or like to really have the time like to dive into uh, another culture. Not on a superficial level, but like really like to try to understand it like um... And you know, I think like that games maybe could be... could be a way like to uh, give people a chance like to experience such... such settings, such cultures. And like, yeah, it’s... I think like that... that could really uh, be a new way like of um, experiencing um, yeah, worlds. Jamin Warren: Yeah. Yeah, I mean I suppose one of the... one of the like concerns would be that like when people play games they are not like quote-unquote "themselves," right? So that... that... that they are different people in some ways. But I guess, you know, when you operate in the real world, people put on all kinds of masks, you know, when they’re at work or with their, you know, partners or with their families. Like this is just maybe just another uh, another type of mask. Um, but underneath it there’s still like, you know, meaningful things to understand about... Michael Hoffman: Yeah, yeah, yeah, yeah. No, that’s a... that’s a very good point. Um, and I think also like that’s, you know, like the... the ethics when you do something like that is... is... is... is really crucial. Like who... who do you... who do you portray? And like to what extent do you allow people that you portray... or should you uh... well I think you should like get them involved into like the game making process. Um, yeah. And so like that’s... that’s... so if someone plays uh, these games, it’s like uh, yeah, that’s... that... that’s a... it’s... it’s a... it’s a ethical issue as well to consider. Jamin Warren: Yeah. Yeah, I mean it is also interesting, you know, for game makers, they build these like social systems, but they don't employ any of the people who study social systems. You know, it’s like... it’s like a bunch of engineers and then they create this universe and it has like millions of people in it and there are like no people from the humanities... they don't have an anthropologist on staff. They don't have like a ethicist on staff. They don't have a... you know, a political economist on staff. But they’re doing all these things that are... that you would... you would want experts to like help you figure out. Michael Hoffman: No, and that’s totally fine as well, you know. But like... you know, like there should be also like some... some other space where you like bring in people with such expertise like. Jamin Warren: Yeah, yeah, of course. Well um, let’s talk about the... the... the uh, the "Anthrogaming" project that you worked on. Um, let’s like start with that term, like Anthrogames. How are you defining that and um, you know, why... why is it important? You mentioned earlier that you think it’s important for anthropologists to be like making... uh, to be making games. Michael Hoffman: Yeah, so I’d like... the... I would say um, the definition is of it is that like you convey uh, anthropolog... anthropological knowledge through uh, through games and make... make these life worlds of other people uh, experiencable. Jamin Warren: Oh, hello? Do you hear me? Michael Hoffman: Oh, yes. Sorry. I think we just had a... we had a little glitch there. Jamin Warren: Sorry, the internet cut out for... for one second. Um, yeah. Michael Hoffman: Yeah. Jamin Warren: Um, and to make the... the life worlds for people... sorry could you repeat that last sentence again? Michael Hoffman: Okay, yeah. Okay, so I said again... um, uh... so an "Anthrogame" in... in... in my understanding is like a game that like conveys um, cultural knowledge and anthropological insights into uh, like into a... int... in... uh... to... to an audience who wants to play it. And um, I think the... the key of uh, an characteristic of an Anthrogame is that like it’s um, so-called "ethnographically thick." Like an ethnographic is a term that anthropologists use like when they write about other cultures or societies. And like so it’s a... it’s a thick description. You really want to give it like uh, uh... describe it not on a superficial way but like in a more in-depth way. And like so if you can um, convey like this kind of like thick writing, like which is hard to do, but like there are anthropologists who are able to do that. And like if you can convey that into uh, an uh, anthropological game, like uh... into a game, then like it... it can turn into an Anthrogame. Jamin Warren: Got it. Have there been attempts in the past to... to make Anthrogames? Either by anthropologists or by game designers maybe working within a anthropological lens? Michael Hoffman: Yeah. Yeah, I mean um, just a few. Just a few. And like uh, for example one by an... done by an anthropologist is um, uh, "The Long Road of Young Pen." Okay. Um, it’s done by like a London-based anthropologist uh, Andrea Pia. And it’s about um, it’s about Chinese labor migration from... from rural countryside to the city. And kind of like um, what labor migrants experience when they go under this um, on this voyage. Um, and another one uh, would be I think uh, you could read "Never Alone," the game uh, as... as an anthropological game because it was done together with um, members of the... I believe it’s the Iñupiaq community. You can help me if I’m wrong. And like I think they’re in Alaska. Um, and like so it’s a 2D platformer game where you play like um, uh, a little girl from the community together with a fox. And like um, in this game like you learn about the uh, the cosmology of um, this specific group. So I think like this could be read as like an sort of like one of the first Anthrogames actually. Jamin Warren: Yeah. I... I mean I... I’m interested in this question in part because um, you know when we think about game genres, right? They’re usually focused on the mechanic or the activity that is in a particular game, right? So is it a First Person Shooter or is it a Platformer? And you end up with maybe less experimentation on the like the setting or the... the subject over time. And so, you know, games have done... they do have a history of like history... there have been, you know, attempts at doing history in games. I think in part that comes from like the legacy of like tabletop roleplaying games and its connection to wargaming. And like there’s been a desire to kind of recreate like wars and battles and that really functions really well. It’s very interesting to me I think from a genre perspective that like um, you know, stepping into a culture and experiencing a culture that is not a fictional culture is... it’s just not been a place that games have really spent much time. So it’s interesting people... they just... for one reason... So I was going to ask you, why do you... why do you think that is? Why do you think that game makers have been sort of reticent to make games like Never Alone which was I think was really a... like a landmark for like that kind of culture work in games. Why... why do you think game makers have stayed away from... stayed away from those types of things? Michael Hoffman: Because I think it’s... it’s really hard to do in... in the sense of like um... like... you know, like if you want to do this seriously you would like have to go first like to dive into a culture. Probably it’s better like if you go somewhere which is a little bit more far away. Like because then you have like the... the benefit of like an outsider coming inside. And like actually that’s always good. Like so you kind of like get a better understanding. When you... while you’re inside your own kind of society, like you often don't see it anymore. Many things that like outsiders see when they come in first. Um, so you know, and... and... and that takes a long time. So like you could work with such experts or like um, yeah it’s in... in general like it... it’s probably very time consuming. This is probably one um, constraint. And the other one is like that um, fictional games like give you as a game designer also a lot more freedom. Because like you can design like all kind of um, worlds um, but you’re not constrained now by the reality. Now if you’re doing some anthropological game like, um, well ideally you should um, ground it in like uh, in uh, the reality of like the society. And in order to be respectful as well uh, about the... for the... towards the community that like you’re representing in the game. Jamin Warren: Yeah. Yeah. No, that’s a... that’s a very good question. Uh, I mean just... just a very good observation. And I hope that more game makers like spend that time... Michael Hoffman: Yeah, that... that would be great. That would be great, yeah. Jamin Warren: Yeah, yeah. I mean you end up with like kind of like the setting, like the Assassin's Creed series. They get a... like that sort of... Michael Hoffman: I think, you know, like I also think um, it’s a great question that you’re asking. Like why hasn't this been done? Like maybe also like in... in my take on it is also that like it’s a sort of double translation, you know? Like so I’m taking like anthropological books and like everyone could do that like um, and like which are already a translation from um, of like... it’s a specific subjective view at the end like of an uh, one anthropologist on uh, that society. And then like you translate that again into uh, a game. So in a way it’s like a... a double translation that you have to do um, if you... if you do it like in... in that way um, and don't want to go on like this whole fieldwork trip. Jamin Warren: Yeah, yeah, yeah. Well... well I do think some of it also is like, you know, game makers they... they come to games with a desire, typically they come with a desire to like experiment with the like the... the mechanical nature of games. And then the setting, that kind of piece of it, comes much later. And so, um, uh you know, I’m... yeah. So I’m of the opinion that like you can do so much more like the setting can carry, the aesthetics can carry so much more of the effective... the effective emotional weight of games. Um, you don't need to spend as much time. Like Never Alone like, you know, from a... you know, from a... from a mechanical standpoint is not a, you know, is not a... is not a crunchy game in that way. But has the same, you know, that sort of meaning making that... that type of game does is... is really powerful because you set it in a place that, you know, you set it in a place that’s going to be really different and it’s a different experience for people to step in. Um, yeah. Michael Hoffman: Mm-hmm. Mm-hmm. Jamin Warren: Well uh, you know, for your... so for your project... Now you came to this... you came to this idea of uh, like Anthrogames or the... the... the... the core of your work is around um, taking this core anthropological text by... by Malinowski and like turning it into something that um, ostensibly could be used for educators. I... it’s a big open question I think for... my... my wife is an educator and so I think she... she... she... she’s a high school teacher. And so she is extremely skeptical... extremely skeptical of like uses of AI like in the classroom. And so it was very refreshing for me to read your paper to see like, oh, I... this is... And I keep telling her, it’s like, well part of the problem is that like these tools are rolled out and these tools are rolled out and like it will take time when, you know, we have actual professional academics like sit with them and develop, you know, tools and resources for someone like you at a high school. She’s... I think the verdict is still out with her but um, yeah I was curious like wha... what... why... what um, what compelled you to start to... to explore um, like creating um, you know AI-driven games that could be used uh, like in a educational, like in a pedagogical, you know, in a teaching... in a teaching context? Why do it that way as opposed to like Never Alone, right? Which is not a teaching tool in that way, right? It is a, you know, it’s a game experience that has like educational-ish elements. But yeah, why... why did you approach this as something that you could um, maybe teach or work with grad students or college students or anthropology texts? Michael Hoffman: Well it came like... it started out of a frustration uh, I would say. Like uh, you know, like I... I spent quite a lot of time in um, as an um, anthropologist doing fieldwork. Um, I think all together like something like three to four years in... in Nepal and India. Um, and like so you know just from my work like I... you know, like the... it’s... and in general, I would say in general that like it’s very hard uh, such uh... you know, such books are written um, when you write this up like you write it for an academic audience. That is also limited in a way. Um, and like you really want to bring this out to a... to a wider audience. And like so... and I think the... remember... I remember when I first had the idea which was actually in this sand mine setting in um, Western Nepal. Uh, where I was just thinking like, wow, you know like this is a couple of hundred people um, families often like working there really hard and like um, digging sand and um, selling it. Um, and like there’s a whole economy around it that I looked then into it. And I was like, oh this is really interesting also like obviously for... for... for environmental reasons as well. Like why are people doing that? Like urbanization and like what does it do to the river? And so it’s like... it’s a very complex topic like sand mining. And like in different countries it’s also quite dangerous like. Sometimes like there’s mafia involved and so on. And like um, often like then when you come back like, you know, in the West... and now in my case in... in Germany or before I lived in... in the UK for a while. Um, like there is maybe like these stories like people are really um, interested like when I tell them these stories. So... but like would you read an academic article? Well Jamin, like you’re more the exception I would say! [Laughs] You know, like who... who really has the time also to... to read such long articles? So, you know, like it... it would be just um, yeah a new way like to convey like this kind of insights in a... in a different way. So like this kind of frustration that like academic books in general and... and reading is in decline like let’s be honest, you know. So the idea is like, well could you um... I’m not saying that like this should replace reading. Actually I’m a big fan of um, you know, reading books and um, I uh, I want to encourage students to... to read. Um, but like I think this could be like more like an appetizer for um, for a book. Um, and playing it maybe um, for... for a bit and then like you get more interested in like the writing and you’re like, oh like this could be... this could be something really interesting... interesting read. And so you just like then um, really start to... to... yeah, read the book. Jamin Warren: Yeah, I mean I think some of the anxiety that uh, you know, educators or skeptics have around something like this is that, yeah to your... to your point, that the worry is that it will supplant reading. As opposed to, this is a way to encourage people to take that first... that... that first step to reading something that is uh... Michael Hoffman: Yeah. No, I don't see it... I don't see it like that. Like actually I think like it’s an open experiment. Like you could also say like the... you know, like it gives people the chance, especially if you then start to build such games with AI to... to look at like the shortcomings of AI and like the real power of like human writing. Um, uh, actually if you see like then for example if cultures get misrepresented or they just get like um, uh, like the culture is sort of flattened, right? So like that’s... that’s also then like... so if you do it like in a more pedagog... what... pedagogical way um, and like um, teach students like well, look, play this, read this. And then um, let’s have a discussion about like what... what do you think about it. Um, I think in that way it’s... it... it could be... it could be done responsible. Jamin Warren: Yeah. Um, tell me a bit about the Malinowski as a text because we’ll... we’ll talk about it a bit in the context of the game. Um, you know... what is it? Who... who was he? What was the research? And uh, and... and why is this work a foundational text for... for anthropology? Michael Hoffman: Yeah. Okay, yeah. Good one. Um, so I mean like this book is... is written in 1922. Um, and so Malinowski went like I think 1915 to 1918 it was uh, to the Trobriand Islands. Jamin Warren: Where... where are the... where are those islands? Just for like a geo... Michael Hoffman: Western... we’re talking about Western Pacific. Jamin Warren: Okay. Michael Hoffman: Um, and um, yeah. So he lived there um, on... um, on this island. It was also... you know, the broader context was like First World War. So he couldn't... because of the war he kind of got stuck there and like he was really then inventing like this what is uh, called in anthropology "participant observation." So he lived with the people, he put up his tent there and like immersed himself into their life worlds and... and wrote um, this book "Argonauts of the Western Pacific." Which kind of describes like the politics and like the economic system and also a lot of like for example magic rituals that are happening there. The Trobrianders like um, would kind of... was immediately striking was that like they go on like these huge sojourns on these huge voyages with canoes. And like these... these canoe rides like they’re quite long. And um, they’re very dangerous also. And so the question was like uh, why uh, you know, why are like... are these huge uh, journeys with canoes undertaken? And like so that was kind of like his starting point and um, yeah and... um and then uh, for to... to write the book. And through writing it like he found out a lot of like uh, you know, like um, interesting uh, insights on uh, this culture. And it became like um, because he wrote a many... so many different aspects of it, then it became uh, later a classic in uh, anthropology. So he’s kind of like the founding... one of the founding fathers of anthropology. Jamin Warren: Got it, got it, got it. So this is a work that you would be assigned to like a Anthro 101 class or... Michael Hoffman: Yeah, and this would be a work that is still read like in undergraduate classes. And uh, you know, like I mean but maybe just for the audience who... you know, like there is also all other great uh, monographs by anthropologists who go to all kind of places and look at modern settings. Um, so let’s say labor migration to Germany or something like that. So it’s uh... it’s not like... I don't want to give the impression here that like you know the subject is more concentrated on what one might conceive as like more exotic places. Right? Um, but like this book is definitely because of the method and because of the richness of the text, and this very kind of thick description, um, it is still uh, taught in uh, university classes. Jamin Warren: Hmm. Yeah. Is there a particular scene that stands out for you that was like a... I don't know, like as someone who’s taught it and read it, is there like a scene that stands out for you as one that’s like um, emblematic of the work or is really, you know... just a moment that you when you were moving to a game context that you really want to make sure that you could... could capture that feeling of reading that in a text for the first time? Michael Hoffman: Yeah, I think like, you know, like the... even like the entry, the entry passages like the introduction when he describes how he arrives at the island and how it looks like. So it’s a very um, neat description um, and uh... of... of... uh... of... of the beach and um, where he um... where uh... of the villages. So like this... and like, you know, the social relations in the village. Um, I don't know, I think I... I sound maybe a bit abstract, right? But like uh, you know, like this is... this is... it’s... it’s... it’s very... it’s very well written. And um, and uh, yeah. So then I thought like this could be... this could be, if you do an um, Anthrogame, um, obviously I could have done like something that I’ve written. You know, but like who... who might be interested in that? [Laughs] Right? Like it would be... be very narcissistic I would say. So... so then then, you know, like I... I thought like well, why... why not use a classic in anthropology? And like, you know, like these... probably most people like when they read the book they’re very impressed by like these journeys that like these Trobrianders undertake um, through the oceans and like this seafaring. And also the magic that is involved, like certain rituals, you know, like to... to... to ensure that like you have a safe journey. Like this strikes probably also like a... a gamer’s heart so to say. And like you’re like, oh like this could be... could be interesting to portray um, into a game, right? Jamin Warren: Right, right. Now... now before we talk about the game itself, walk me through the process of like developing... developing the... developing the game. Because you chose texts that... you chose that... like text adventures or interactive, you know, interactive fiction as the... as the way to express the work. Walk me through that process of like so you have the text, the text is completed. Uh, obviously a lot of writing about Malinowski throughout the years. How do you go from that to like developing a game that might be playable for someone to experience the... the text? Michael Hoffman: Sure, yeah. Like so... like basically um, there was this ChatGPT moment as we all know. Nine... what was it? 2021 or 2... I don't even remember anymore... right? Like 21 November, right? Jamin Warren: It seems so long ago. [Laughs] Michael Hoffman: Yeah, yeah. Okay. And uh, so like, you know, like there’s also on the internet like there was uh, quite a lot of gamers like playing around with it and like some indie hackers like trying to make text adventure games. Um, uh, but not like with... with books or like... this was so the first publication then like was like, well can you actually use like a... a Large Language Model um, in order uh, to... to produce such a game? And then like choose like a... a text format. I think is more bet... is better than um, you know, like uh, the... there is like this uh, famous game like The Oregon Trail... Jamin Warren: Yes. Michael Hoffman: ...which the older ones will remember. And um, and so I thought like well yeah, some... something like that would... would... would be good like to start just like in a very easy way like to enter like this space and make such... such a game. Like uh, I mean if you want to uh, turn Malinowski’s book and turn it into a 3D game like that’s probably quite a lot of work. You have to produce the assets. The assets have to look like the Trobriands. Not so easy. You won't find that on the Unity Store and so on. So that’s why, you know, for the beginning stay without graphics and then gradually like go towards more the hard parts which are the graphics. Um, uh, but story narration is also... yeah, is also a challenge. Um, yeah so like that was... that’s how it started. And then um, from then on like I went more and more uh, technically into it. Because like, you know, like I... I have a background as an um... um, uh, anthropologist but like I’m also a computer scientist. Like I... I work in... my day job is like in a Supercomputing Center actually uh, here in... in Germany. And like so in... in my daily work I’m building AI models. So that’s kind of like a interesting way then like to apply this um, in your off-work time. Jamin Warren: Yeah, yeah. Very convenient, very convenient. That uh, that... that is a skill that you... skill that you have. Um, and so building... building the game, um, tell me about like the implementation like of... of AI? Because, you know, it’s... it becomes this term, it’s like, oh it’s an AI powered game. Like what does that mean like functionally in terms of how the game like expresses itself? Michael Hoffman: Mm. Yeah, so like uh, it’s a... I would call it a... AI-native games. And there’s sort of like a very new genre. So like the AI produces like takes the text and like um, it gets fed into a uh... uh... a Large Language Model. Actually like in um, through a Retrieval Augmented Generation uh, technique. Um, that means like that you kind of like force the... the LLM like to really just read from the book. Um, and then um, you uh, the output that like you want, that you envision is like a... a short description of a scene and then like you have different scene choices. And um, you can pick them. And like, you know, I... I developed like several iterations and... and wrote about it. So now it’s also like that you create then um, from the scene descriptions like create prompts in order like to create images. And like try to play around with that. Like how does that work? Can you actually... can also like image models like DALL-E... DALL-E 3, how can they produce then um, kind of images from the Trobriands? Which... yeah, it’s... it’s a challenge because like these models are mainly trained on like um, uh, high culture material you could say. Like, ah... um... US uh... or... or like West or like China is like there you have a lot of image material. Um, and like yeah, some settings like the Trobriands are more limited. But like there’s... there’s lots of way like how you can extend that. And then actually um, I choose like a... um, in the latest uh, version it... so it’s a pixel art graphic and also choose to portray the indigenous people in a more respectful way through uh, silhouettes. So um, in order also like to convey a little bit like the um, colonial context in which like Malinowski wrote. And um, bring that in like so making... making students or like whoever plays these games um, um, aware about like so these questions of representation that are... that are happening in the book and like that are also happening uh, in... in games or could happen in games. Jamin Warren: Yeah, yeah. Um, so someone would sit down, they are presented with a... like a text interface and they are just... they can just start interacting basically with... like as an anthropologist like inside of this... ins... inside of... inside of... Michael Hoffman: Ah yeah. Yeah, yeah. So... sorry I... I totally forgot it. Um, um, and yeah like so it’s like uh... I mean principally there are like three different ways like how you could do it. Like so you could play as like the anthropologist or like you could just chat with like Malinowski or like you could also be like the, you know, be the mentor of um, uh, Malinowski which was someone called Seligman and like uh, you know, step into his role. But uh, you know like um, many different ways like many different roles that you could take on. But like in this version it’s more like a uh, Dungeons and Dragons kind of setting where the LLM is sort of like the D&D master and creates like the scenes and then yeah, you are stepping into the role of the anthropologist. Jamin Warren: Ho... how would this be different? Uh, so what are um... like I could see a world where you make this game and it is more like a traditional text adventure, right? Where you scripted all of the individual scenes and there’s a branching pathways, you know, branching narratives but it’s fixed, right? There’s a fixed amount of things that you could do. This is much more open-ended. Um, what are... wh... why did you decide to go that... go to... dec... decide to go that route versus maybe the way that maybe other educational games have gone of just like having a more traditional, you know, kind of like story? Michael Hoffman: Hmm-hmm. I think uh, it’s... it’s an experiment. It’s still like an open question like whether like a hand-made system, whether this would actually something preferable or like um, whether like... because if you do it with AI then like every game is different. Um, and uh, but like the advantage of that of course is that like once you build such a structure, like you could theoretically, you know like build... build a system that like migh... maybe even do... does that with a whole different range of anthropological books or historical books. And um, and so that would be very interesting to... to... how far like that’s kind of like the whole experiment idea. Like how far we can stretch that also technically um, and build such games. And uh, that... in order then in... in... in the end like to build something that like educators could use in the classroom and just upload the book and then, you know, you... you have some um, Anthrogame. Um, but that is actually like it’s still... still a way to go and um, it’s hard to do. Jamin Warren: The... the corpus, like the body of work it’s pulling from, is it just that Malinowski text? Or is it pulling in other things about, you know, about him? I mean that’s partly what’s interesting also is like there’s only so much that you have, you know? And uh, um, obviously now like I’m sure the way anthropology is executed you have recordings and you just have like so much more material to work from. For this Malinowski text is it... it’s just the book? Or do you have other sort of material that’s in the... in... that... that you’re pulling from? Michael Hoffman: At the moment it’s... it’s just the book. Actually like uh, I had a conversation with... with an anthropologist the other day and like they were kind of more wishing that like it’s um, uh, like that the corpus is broadening because like there are quite a lot of others who have written on the Trobriands... Jamin Warren: Oh, I see. Michael Hoffman: ...on these islands. So like you could get like a different perspective on um, on the culture. But like also like there’s a lot of material on um, Malinowski himself um, uh, and so that could be integrated as well. It doesn't necessarily all have to be, you know, in the game and um, then pulled out by the AI. Um, but like it can also be done as a sort of like a companion material like that you give in the classroom and say like, well, you know like this should be... this should be read in parallel. Jamin Warren: Yeah, yeah. Um, what are... what are some of the things that are gained from this experience with the text versus... you know what are the things that are gained? What are the things that are lost by, you know, experiencing the... the text this way? Michael Hoffman: Mm. I think it’s um, it’s uh, uh, so far in the evaluations and like um, yeah, uh, I think it’s for... for no... like yeah, obviously the... the... the... the... those who are most critical about such games I think are anthropologists themselves because like they know... they know the text so... so very well and like um, they have a very high standard like how this system should work. And rightly so. But um, others uh, for example engineers when they play it, like they usually like they uh, really surprised, you know like by this world. They get really interested in it. So like it raises some uh, curiosity I would say into like the book, into the local history, which is um, very important. Um, to draw attention to um, specific places that like all these places usually like in anthropological books they’re sort of under-represented in like mainstream media. Um, and so I think this is... this... this... this... this is some very uh, important insight like that there’s some curiosity that like gets out of there. And like um, yeah and that, you know, some of these uh, all of the anthropologists that I’ve talked to, despite being critical, actually would like to try it out like in their classroom like how... how... how it would work then. Jamin Warren: Yeah, yeah. Well I want to talk about some of the... you know some of the challenges of working with LLMs. Uh, you know, you... one you mentioned earlier like some of the colonial context for... for... for Malinowski. And obviously for like a contemporary reader, they’re going to bring an understanding... and... and... you know as you said this is sort of... this is an early text. It’s also early in the field of anthropology. You’re missing, you know, it takes time to develop like, you know, standards for observation and having a code of ethics and maybe some of the things that anthropologists could do a hundred years ago or the way they could talk about their... the people that they were studying is very different hopefully from the way that anthropologists do that work now. Um, how did you handle that tension between like you want to keep it authentic, right, in a sense, right? You want a student to experience the text as you might have experienced it if you’re interacting with Malinowski, you know, himself. But at the same time, you know, there might be problematic framing or, you know, how do you make it useful? How do you like... I’m... I’m curious how you balanced that tension between like, you know, an older version of anthropology and the experience of like what you know as an anthropologist today. Michael Hoffman: Right. Um, yeah I mean like there... there are so many issues. Like for example one... one thing is like, you know, like to get the authentic tone right. Like there you can play with like, you know, the prompting for example of the... of the LLM. Um, then the... the other uh, issue is also uh, inference time. When you work with graphics um, in order like to produce uh, graphics in... in real time. Like uh, right now uh, if you use APIs then usually takes like 40 to 50 seconds actually before the... the ga... the... the image is loaded that like depicts like the... the scene that like you’re describing. And um, so that for example like we did like with a... with a little mini game um, where you have to learn sort of the... the local idioms. They fly around and you have to catch it. Um, you know in... in that way actually draw like use some loading time, like gamify that again like in order to convey uh, local idioms. Like okay, what do I mean by that? For example like uh, in this book like uh, there are... there are for example when they go on the canoe rides like there’s like this story of like uh, this giant octopus um, that has a certain name, the kwita. So like, you know, like these things like if you... these local idioms, if you um, use them in this spare time for loading like to catch it, then like you automatically draw their attention like also like to, you know, to facts like that. Um, the other thing is like to... to make, you know, in... a game essentially um, also requires quests. You need to kind of like have some sort of progression. And um, yeah and so then that’s also how do you make that um, that the game has also an end? Because like the LLM would do that like um, yeah in... indefinitely, right? So like that’s also like some quest generation part in it. Like we have different quests, four different quests. And like you force the LLM like to come back to that issue. Um, and yeah through... in... in that way that these are quite... it’s quite challenging. Jamin Warren: Um, you know for anyone who’s used uh, you know for anyone who’s used ChatGPT or Claude, um, you know that it can be wrong from time to time. They give the little, you know obviously the little warning. I... I mean I find that it... it like it does make mistakes but not as many as... given... given the numb... the amount of information it gives me, it’s like generally pretty right like most of the time. Michael Hoffman: No, no, no. I mean like the hallucinations uh, are a real problem with LLMs. And... and that’s why like we went for like a uh, so-called RAG structure, um, Retrieval Augmented Generation. So where you actually um, the LLM draws only from the book. Um, and in that way you make sure that like the information that you get is not hallucinated. Um, it... it can still be hallucinated, let’s put it like that, you know? Like nothing is perfect. But like uh, like you minimize hallucinations. And like we did a lot of experiments as well on, you know, how does this RAG structure, how does it improve then um, you know, like the fetching out of... of... of... um, of... of text um, and like the... or like to be more precise like the... the accuracy of um, descriptions, right? Jamin Warren: Yeah. That... Michael Hoffman: Are there hallucinations in it or not? Yeah. Jamin Warren: That’s very interesting. By like narrowing the focus and saying, just focused on a smaller body. Don't... don't go out into the world and don't go into your total corpus and don't search the internet. Just... just focus on this work that’s been provided for... for any of your answers. Um, that’s really... that’s really smart. Um, well a couple last questions for you. Um, you know you mentioned that this is obviously like a... a prototype. Um, mentioned this is like a... is a prototype. Uh, you know, one of the things I’m always interested in is, you know when um, you know when creators or artists return to the field of games over and over again, you’re clearly committed to this space. Um, but you know I get the sense that, you know anthropology as a field is... is this is new territory for them. What... what do you think it would take for like Anthrogames to become like a standard methodological tool um, rather than like an experiment? What makes this... what makes this a mainstream thing that... that maybe students are using on a... on a regular bas... students and teachers and researchers are using on a... on a regular basis? Michael Hoffman: Hmm. Well I think um, diff... I mean first I think the first uh, barrier is technical skill. If you want to do it on your own as like an anthropologist like, you need to have like some training in like how LLMs work, how RAG work, how like different kind of new structures that are coming out now, reasoning models and so on. Like how do all these models work and then put it together? So like you could go in that way like maybe like... I think in the humanities now like there’s a new trend like that everyone wants to work more with AI. It’s becoming more and more. And so like this could be a... a good way also just to practice like basic skills on um, AI. Um, and the... but like the other... so that’s kind of like from a more uh, producer side. From a... from a consumption side, um, in order to play these games because like they are AI-native, um, there is the issue uh, that uh, these games uh, rely on um, uh, API requests to uh, commercial providers. And they cost money. And um, and... and that also is at the moment a problem because, you know, like that restricts uh, like if you put this out and then a lot of people play it, like then a lot of um, eh... eh... the... the AI... the like uh, the inference costs money and like that might uh, go beyond your budget. So like that’s still an open question. Like well I still think like that prices will go down and like, you know, like that someone will come up with a model as well like how to do that in a more uh, commercial way in order like to... to reach a... a broader audience, right? Um, so... so that... that is I see like these kind of two... you know, two barriers like for the widespread of... of Anthrogames. And then there’s also like the question, you know like is this really interesting for... for... or is it... is it... will it stay more kind of like a... a niche um, and like even if it stays a niche it’s... I think it’s an inte... interesting niche. Jamin Warren: Yeah. Yeah, cool. Um, trying to think if there’s anything else. Um, I... you know was curious just in the... in... with students playing the game, was it... was there a particular interaction that stood out to you? Like something that the... that the game revealed in the... in the play structure that was um, a sign that you were moving in the... in the right direction? Michael Hoffman: I have to think now. Um, yeah. Um, you know, like there were all kind of unex... unexpected um, events that happened then when like we let uh, students play it and like colleagues play it. Um, it was uh, you have like since you have like... so the game is that like you have a scene description and like you have like these three options. But like you also have like an open input field for example. And uh, where you can just say like instead of these three options you could say whatever, you know? Like go to the center of the island or like try to go to a canoe and like go to the next uh, island. Something like that. And then um, people tried to uh, very quickly students tried to find a shortcut in order to solve the four quests that were given to them. So like uh, that was... that was for example something very um, surprising for me. And like uh, also refreshing that you know like that like you take like a more gamer’s approach and try to hack it like how... how do you come fast like to the... to the solution? Jamin Warren: Right, right, right. Yeah, that... that is often a challenge with games. Michael Hoffman: Yeah. It is, it is. Definitely it is. Jamin Warren: It’s like always a... always a challenge even with our... yeah, whatever. Yeah, yeah, yeah. Um, great. Well thank you Michael, I really appreciate you making time to chat with me about the work. Um, I, you know, I... I’m very excited about this type of work just because, you know, I... I think that, you know so much of the conversat... I think it... it can be hard for... I think it can be hard sometimes for the public to hear about a lot of the bad things about AI. And there are many... ma... many re... and you... you... I... elaborated on several of those in the paper as well, like the energy usage and obviously like the cost and, you know, there’s just um, and you know some of the... you mentioned you know just the... sometimes the... there are challenges around like getting accurate information. At the same time, I do think it’s like the hard work of, you know, academics like working with these tools to try to bend them and make them useful for ultimately something meaningful and more important. And the goal for you is to like how do you get this work that’s really important to... out to a wider public? How do you get students to engage with it in a new way? It’s not just like... it’s not just because it’s a novelty, it’s because like oh this could be a really... a way for people to deepen their understanding of um, their deepen their understanding of games which I... which I think is really exciting. So. Michael Hoffman: Mm-hmm. Yeah. Well thanks a lot uh, Jamin for... for having me on... on your show and um, was a very uh, interesting conversation. Thanks for your very interesting questions. I enjoyed it. Jamin Warren: Yeah, of course. Of course. All right, let me... let me press stop here.