Mason Grimshaw: Probabilistic Living, Indigenous Data Sovereignty, and Foundation Models for the Earth

In this episode of Floating Questions, we sit down with Mason Grimshaw, an MIT alum and mission-driven data scientist driving critical progress at the intersection of environmental conservation and indigenous empowerment in the age of AI. We explore how his work, life, and community view intertwine to thread a journey that spans from the Rosebud Sioux Reservation to the cutting edge of geospatial modeling.

• Parenthood and the Probabilistic Worldview: How Mason applied a "process over results" mindset to navigate the challenges of raising a family while completing his degrees at MIT.
• Language as Ancient Code: How the Lakota language uses redundant information—such as encoding kinship and physical direction—to ensure the integrity of oral histories, functioning much like modern error-correction technology.
• Building the "Pocket Botanist": How indigenous youth at the AI camp use computer vision and Docker to preserve traditional knowledge of medicinal plants in the Black Hills.
• Indigenous Data Sovereignty: The critical need for tribes to maintain sovereign control over their cultural resources to prevent models from training on sensitive data without community permission.
• Geospatial AI for the Earth: Using "Clay," a foundation model for the earth, to transform satellite imagery into semantic data for tracking whale migration and deforestation.

In our rush to build the future, how should we handle the delicacy of history, memory, and wisdom?

Transcript

episode 19-Mason Grimshaw
===

[00:00:00] 

Rui: Welcome to Floating Questions, the podcast where curiosity leads, we follow and stories unfold. My name's Rui, simply asking questions, shall we begin?By day Mason Grimshaw uses data to protect the environment and Ode partners. And by night he changes diaper for his four kids. He's an MIT alum who played life on legendary mode. Surviving rigorous problem sets while raising a toddler Now he bridges high tech and deep roots as the VP of IndigiGenius powering indigenous youth to master AI and data science.

Mason, welcome to the show.

Mason Grimshaw: Hey, thanks for having me. Super excited to be here.

Rui: Yeah, thank [00:01:00] you so much for accepting the invite. 

you mentioned, you started your family while studying at MIT 

For most students, maybe they're optimizing for GPA or job search. Um, how did becoming a father actually change your objective function? 

Mason Grimshaw: I did MIT undergrad and then I did the master's right after. this is still pretty young by certainly Northeastern standards and certainly by, um, other MIT student standards. I, met my wife in high school and I think she's great and I'm so lucky to be married to her. we got married in undergrad, going into junior year. So she moved out to Boston. We got this little apartment that. Um, it was only slightly bigger than a shoebox, maybe. we come from the Midwest, um, from South Dakota.

So it was actually a little funny in that everyone in the Northeast was very shocked [00:02:00] to we were having a baby. Um, then everyone, here back home where I'm sitting right now was like, yeah, obviously you're having a baby, like, it's time. I think just the different cultures is sort of interesting there.

we're not so far ahead of our friends really. many of them had kids either right after us, or very soon following us. So. We were only barely early. But that's kind of where I'm coming from. being good at my job and, and doing well in school, all of that is still very important to me. that was always a part of my objective function. but I grew up where your community like, so important it's kind of impossible to do anything without that. and so when it became time to change that part of my life, it just felt very natural. We didn't stress about it. Jules was home with  Augie then, she would go to bed like or whatever. [00:03:00] I would stay up with Aki and take the next shift. 'cause I was doing psets and homework and stuff. he would sit on my lap and then we'd do, I'd do homework, and then I'd go to bed around whatever, two or three. Um, you do lose a little sleep. This is unavoidable. for the most part, things stayed mostly normal. it's a hard pitch to people who don't yet have kids and are very, career oriented, which is totally fine. but it is hard to sort of get across how meaningful those moments little baby Aki who's seven now, um, like laying on my lap while I'm doing homework and I don't know.

Those are just really, really, really nice moments that I look back very, very fondly on. Um, it did complicate things. It didn't, uh, wasn't super smooth sailing, um, it was just a very natural thing and I, I had a feeling that it would [00:04:00] lead to more good than bad, and it absolutely has.

Rui: That is very moving for me to hear. I don't know why. Just the way that you described the journey. I don't have kids yet, there's always a question in the back of my head of like, sometimes I'm a kid, am I ever enough to be a good parent?

Whatever that means.

Mason Grimshaw: I think everyone who's having that particular conversation with themselves or asking themselves that question is probably already ready. it's the people who don't ask themselves that question I get kinda worried about. the truth is, um, and maybe this is more just portraits how I sort of look at life. there's not really a perfect time. but everyone always figures it out. It just, it just happens. So I think if you just trust that commit to like, being intentional about it, there's nothing to worry about really. 

Um, there are parts of me that wish I had, um, you know, moved to [00:05:00] San Francisco or something.

You definitely have to like sort of pick and choose. But honestly I think it is the picking and choosing it is the prioritization. It's the, um, willingness to sort of. Give up a little control and go where life is taking you. at least for me, has led to a much more fulfilling life overall. 

Rui: where does that confidence come from? Meaning the confidence of giving up some control to flow with life. Because when, you know, especially people who are used to grind really hard in life, um, and also really smart, usually also come, come with a sense of like, I can control my fate and I can control my path, and I would try to go to the direction that I set my mind to, for better or worse.

where does that willingness to give up control come [00:06:00] from?

Mason Grimshaw: that's a really good question because I think a lot of people fall into that, um, trap. Trap maybe isn't the right word, but it feels like a trap because I don't know that life is truly controllable for anyone on like. The grand scale of things. Um, but for me, so I grew up on a reservation, so I am, uh, I'm Ong Lata or enrolled member of the Rosewood two Tribe. when I grew up on the res, life is very much like we take it as it comes. sort of vibe. Not a ton of industry, not a ton of grinding on the res. You know, you do your work for the day but then you're kinda whatever. 

when I got into MIT, I had a bit of imposter syndrome, and I'm not supposed to be here I need to like, really hunker down and make this happen. in the context of younger Mason, in that moment, I'm a kid who grew up on the res. I, I didn't, [00:07:00] go to a high school with like a bunch of, we had a pretty limited set of AP courses and they weren't teaching us how to code or anything like that. and I go to MIT and I kind of prove to myself that I could hang a bit at that point I was just like, yeah, I think things are gonna work out. So I have had these moments where I've had to adapt and change and in all of them they were difficult, but it worked out okay. it hasn't been perfect. but again, I think once you just sort of take on this perspective of life is this uncontrollable beast and I'm a very, probabilistic thinker. So all I can do is increase the odds of good things happening, but if the bad things happen. it's just a bad poker hand, right? So there's only so much you can do. And this bites me in other ways because, in situations where you really need like perfection I'm sort of, well, it's good enough, you know, we have like a 90% chance of doing well and that's good enough for me. [00:08:00] that probabilistic way of thinking is something that I've always sort of had intuitively I have words for it now. Um, but it's this idea that I'm playing the process, not the results. if you play poker people say things like that, and I just apply that to life.

Rui: um, a huge part of what you have talked about, In relation to building a family and your confidence about like being able to let go of control, it all traces back to the community that you grew up in, the reservation that you grew up in. and I know that a core part of your mission seems to be reframing indigenous knowledge, um, like weaving.

I think you mentioned it is also a form of computation or coding. Can you, walk me through that translation and how did you come to the realization that, the heritage itself is technical and you can help teach kids maybe through that lens.

Mason Grimshaw: Yeah, so there's a lot of really fun thread to pull on here. I think, one of the more interesting ones. Is that [00:09:00] indigenous languages, certainly, um, is they encode, they have redundant information in them. Um, that, that essentially works like this, self-reinforcing, code for important information within a sentence. 

Rui: Can you can gimme an example?.

Mason Grimshaw: Yeah. Well, English doesn't really work like that. it is just like in the way that you speak about and their relationship to you and, in some indigenous language, you also encode like direction. So you might refer to a person in, like if, if they're sitting to the north of you, you talk about them a certain way and, whatever the other directions too.

But you also have their, their sort of kinship relationship to you and their name obviously. But you have all these different ways of referring to a person that like very concretely determine the person redundantly as an example.

Rui: Oh.

Mason Grimshaw: there's, there's other sort of things like this, but, but this is important because like in English, you try to be very concise.

I'd have to talk to a linguist. My hunch is that this redundant information is really important because a lot of our [00:10:00] history is shared orally, so you have to make sure that. When you're sharing the story, you get all the information correct. And a way to sort of en ensure that is to, is with this redundant information. Because if any one of those bits flip, if any one of those words flip then the rest of the sentence you can sort of figure out that is incorrect. it's like this net of things that if you lose any one of the threads, you know, that like there's other this stuff to reinforce it.

So to me that was always very interesting and I'm not linguist, but when I learned that I was like, holy, holy cow. that's a really cool, um, technology, honestly, really. And we have lots of technology just generally speaking. we live here in South Dakota.

It is negative eight degrees today. and it would've been, negative eight degrees, you know, back in the day before electricity. So. Surviving that understanding, where you need to physically be at certain times of the year, but also how you need to clothe yourself, how you need to create the clothing, tan, the hides. all of that is technology. and it's important because [00:11:00] it becomes a part of the culture and it keeps you alive. so I've always viewed ourselves, as a very technical people, and we just need to understand for ourselves what that means in terms of 21st century like technologies. We have to figure out how to do it, like with respect to our traditions and our cultures and the earth. Um. What does that look like? And so this was a very important, um, aspect of me.

I mean also like having come from the res, having come from a high school that, that didn't have any code. I thought we needed to get this back in the community also. So just like making sure we're all talking about it and, and have the awareness that this is happening. Um, it began as more code I guess, and now it's more ai.

But, but the same thing is true. Like we just have to have an opinion. And our opinion might be that we don't like it, but we need to have an a well-informed opinion, and not just let these things happen to us. So [00:12:00] all that as the context, so I run a code camp for high school students. We bring 'em over to a college, Um, we house them for three weeks. a real like practice, college sort of environment. They're by themselves. They have chaperones and TAs and things, but, 

they're away from their parents, They're in class all day, and then they do extracurricular stuff in the evenings. we teach them how to code from zero experience.

we teach 'em the first week how to, just basic, basic Python stuff. end that week with a big presentation on some kind of data set that we change up. The first year it was NBA player salaries, 'cause they thought that was interesting. Um, ball is life obviously. And then, um, the second year we did I think was the Barb Inheimer summer. So they really wanted to look into movies. Um, one of the students had a model to predict box office scores. And one of the important variables was like Adam Sandler, uh, which I thought was hilarious.

Like, is [00:13:00] he, is he acting in the movie? And then the, and then this year we did weather. So we grabbed a bunch of data from local weather stations around the region, explored some trends, things like that.

they got to do a little more local. Sort of environmental, uh, data science, which was super fun. And then the second and third week are really about building their apps. the first few years we did, computer vision apps. So we have, uh, Linda Black Hill, she's an ethno botanist who helps.

We go on a hike through the Black Hills. The Black Hills are very sacred to us. And, in the Buck Hills, there's lots of medicine, so lots of like plants that you would have use for traditionally. and she'd show us, she'd say, there's one that you can like. Sanitizes your mouth. It's got like this, alcoholic kind of property to it. Anyway. The students take pictures of 'em. We, we create a label data set, and then they train a model to recognize those plants. So if they were ever hiking, they could, they could do that exercise for themselves. They, they'd have like a pocket Linda, so to speak. and then the third week is really building, putting all that up into an app. we use hugging face spaces for that. And then [00:14:00] this year we had, which is super fun, we had, uh, a language teacher come in every day and work with the students. and they built language apps. So instead of, um, classifying pictures, they were classifying their phrases.

So they all got to pick like their set of phrases, but they had to say them, practice the pronunciation and make sure they're saying them correctly. And then create these little labeled sequences of, or labeled phrases. and then they all built apps for them. So one of the students an app, um, so is, uh, is rainy.

Raining. And, um, so he'd say that, and then it would, like a little ring cloud would show up on his screen or, or he'd say, uh, well, key on thunder would, like, lightning would show up on his screen. so there's just like a lot of, of fun things like that where you're doing the integration, you're doing the weaving, right?

They can be technical people, they can be like deeply technical people. they use Docker, they use Git, they use all the industry standard stuff. by the, second week we're like, okay, do a Docker poll, update your GI or pull the [00:15:00] GitHub repos and we'll start the class.

They're in, like the terminal and they're just going, it's really cool. It's really fun to see. we're providing that space that is, um, technical and we're asking them to do difficult things, in a environment where they can be fully themselves, so fully Lakota youth, we see a lot of really cool like of that where they just are engaged and they're also willing to do hard things. and I, I just think there's like a calmness that seems to, to hit them, which is really nice to see. So where we're at with that. I mean, indigenous genius is great. We also have a research lab called Flair. first language is AI Reality Initiative, which is about. Building sort of the, the fundamental research required to do chatbots with indigenous languages. 'cause they have different structures than a lot of Western languages, certainly English. but yeah, it's been a really fun journey. we've had four camps so far. This summer we'll be our fifth 

It is a labor of love. 'cause it is a labor, but [00:16:00] every time we're done I'm like, wow, that was like really and like healing. Yeah.

Rui: Uh, this is amazing, first of all. Also thank you for just giving a overview of like what that, uh, nonprofit is really about 

when you talk about how your language encodes information in so many dimensions in order to make things so concrete to the point where if something is potentially by accident a mistake, people can know it's a mistake 'cause it just doesn't follow that context for logic sharing 

that to me is like perfectly set up kids for some type of programming. Sort of like, thinking, To clarify, I'm not a programmer, but for the very limited experience I've had with programming, one thing I realized is you need to be extremely precise about what you want to do, and then breaking down the task to the bits that you can actually logically sort things through and allow you to [00:17:00] debug at what point things go wrong that has a lot of parallel with the characteristics of the local language that you just shared, which is to me amazing.

I was just thinking. Have you ever tried to teach the kids, um, from the angle of the language, your language itself to make a lot of like parallel for like, Hey, this is how we talk, think about this in programming. It's equivalent of you doing that. Um, because I think it's very cool that this language so precise.

Mason Grimshaw: I we haven't, no. but that's a really interesting way in, 

I think we're predisposed to being good at programming. it's a hypothesis. I'd love, I'd love to test. and I do hear from a lot of people, if you like code, It's easier to learn Lakota actually we don't go from angle of here's here are the rules in Lakota and then mapping those to sort of like Python ideas. [00:18:00] I think that's a really compelling, uh, idea. Um, I think what we'd want to figure out is, is that too much to hit them with it once? So most kids, just because of our history of assimilation we're very lucky that we have lots of schools that will teach these days. But in general, most of our students might know a few words, but they don't fluently speak. But, um, no, we don't. I think that's really interesting though. I wanna, I wanna let the team know that maybe we should consider some lessons like that.

Rui: Yeah.

I'm super excited to see potentially what angle that you might end up, uh, with maybe just work with a linguist, um, especially someone who is also interesting help preserving indigenous language, but someone who also really understand cutting edge frontier language modeling. 

Um, I don't know. Right now I'm just spit bowling, like language is how we project our thoughts and reasoning onto this world. [00:19:00] And there are different reasoning mode and different reasoning mode could give you different advantages just depending on the context. And today's large language is mostly dominated by the most spoken languages around the world.

But potentially missing out interesting benefits from smaller languages, uh, meaning by the amount of people that speak that language. And so by modeling it out, we might discover even new model architecture that represents a different information processing architecture. So I think that's like fascinating.

Mason Grimshaw: you're Yes, yes, yes. So, so a thousand percent, this is a lot of the work that Flair is doing.

they have a linguist in-house.

they're building out the different architectures and the different things that would be required to, to sort of capture this different way of thinking, [00:20:00] which I think is so important because, um, turns out. Uh, most of the languages in the world are poly synthetic, technically boils down to this vocabulary problem. Um, and so we're missing a tremendous amount of humanity and we're missing out on the way that they process information. Um, I sort of tend to think that if people have been doing things a certain way for a long time, there's like something to look into there. And so yes, we're doing this, uh, building out some infrastructure at the moment, we should have some experiments hopefully by the end of this year, maybe early next. But yeah, this is a ongoing area of research that's like really interesting and compelling.

Rui: I know that one thing, um, that you also care about, it's the data sovereignty, but can you actually explain what does that word the phrase really mean, um, to you, and how do you define that? And how do indigenous [00:21:00] people really feel about their own data? Not just the part, like how do you correctly ask a language model to produce, you know, phrases and sentences that actually also make sense in cultural context, but the online data trail that they leave behind if they engage with this, 

Mason Grimshaw: and there's so many things here. So our history is, is one that like. Um, so through colonization, the lands that we used to live on and hunt on and pray on and gather on, um, were sort of taken by the United States government in lots of ways. so we have reservations. That's the land we were allowed to keep, and that colonization obviously has like a tremendous amount of impact on how we live today and how we view the world. I honestly think Lakota people are very open and we share a [00:22:00] lot of stuff, a lot of tribes don't. Um. And I think some of those tribes might've before, but they don't now.

A lot of 'em have always been secretive about stuff, and that's okay too. But basically the idea is we want tribes to have the ability, sovereignty over their data, the ability to make sovereign, uh, internal community driven decisions based on they want from their data. Um, they want, I mean, all the way up like, do they want it on chatGPT? Bt do they not? Do they, want children's books? Do they not?

Do they want repositories online on dictionaries and things that are freely available to everyone? Or do you need to prove that you're a tribal member so you can log in? there are so many different. things about it that you can make a decision on and it's just becoming increasingly important [00:23:00] these days because if any of the large language models train on data that gets leaked somehow or they get ahold of somehow, then you can't put it back in the box.

You know, like if a model is trained on a dataset, it would be difficult to, not impossible, but quite effectively impossible to remove it from the model. And That's why we think it's really important that tribes are having these conversations. it's one of the reasons why, a lot of our work is in community because it's most important that they can see these ideas and start to decide for themselves. that's to me, like what data, it's just allowing them to make their decisions. And our data is usually. a recorded elder telling a story to someone. Um, this is very common, perhaps a spiritual story, perhaps a, a meaningful [00:24:00] story within their family or through their community to a trusted person. If that's your data set, then I think you think about it wildly differently than you do about the tweets. Um, it comes from a completely different place. has a completely different amount of meaning. And so we want tribes to be able to make the decisions on what happens with that data. And also this is sort of Mason talking. But I just sort of don't trust that the big labs will handle it with the care that I think it deserves. I don't think it will be treated, responsibly. So this data sovereignty thing is super important. And again, tribes are gonna make decisions that are very open. They're very, very, very welcoming to this. Um, some tribes are gonna wanna stay closed off. Some [00:25:00] tribes are gonna stay closed, but they want the tools for themselves. So they have to build the tools internally, keep their data sets internal. They're gonna need capacity to do that. some just want to be able to license it out and be able to, reign it back in if something gets outta hand. What we just want people to understand is what is happening, what levers they can pull and. why we would like this control. I think there's groups out there. like no, everyone should share everything. It only makes the models better. It only makes all of us smarter. But if you're dealing in the context of a native community that has this history, that has this valuable cultural resource that, you're not entitled to, that argument starts to break down a little bit. And to me, that's all wrapped up in the umbrella of indigenous data sovereignty.

Rui: I really love. Hearing this perspective, because I've been thinking about that a lot. To [00:26:00] me, I think data sovereignty is something that everyone should care about. Whatever background that you're coming from. 

another guest on the podcast, um, Julia. she used to be a fashion designer, lives in New York. And doing a lot of art related work. now she curates, substack featuring avantgarde artists, interacting with technology.

And one thing that she mentioned, is there is a group of professor in University of Chicago, they invented a tool for the artists to poison the data. If they realize that their artwork displayed on public, could it be used for training. 

the technology they insert into their artwork, would mess up the training data in general. So, interesting how there are right now so many different responses to what do we really do with the data, I predict at some point there has to be some, almost like a iPhone moment of new device that really give people the choice back.

And I think that would potentially change a lot of [00:27:00] industry. 

Mason Grimshaw: no, I agree. And I think we're going the wrong way, which is like putting cameras on glasses or something. I, I feel very weird about that. I'm pretty chill about trying new things and I like new technology. but I just worry back to sovereignty. if you got your ray bands with your camera in them and they're scanning every face that you walk across, you've sort of opted those people in the AI ecosystem and they didn't get to make a choice in much the same way that someone whose, whose art is ripped off their website, didn't get to make a choice. Um, and like, is everyone gonna close up? Like websites become private? I don't even know. And there has to be a better way. This is like ridiculous. Um, and the hyperscalers, the big labs are only like incentivized to get more data and I think it creates these incentives that are not aligned with the general public.

Rui: I agree. Um, I guess there's just so much complexity and nuance when we talk about [00:28:00] data privacy and sovereignty in general.

Mason Grimshaw: Yep.

Rui: The example that you gave, like, you know, recording data on, on your glasses, whatever. On the flip side, uh, 'cause right now the domain that I'm building for is in risk, right? Like having access to data really helps us to understand whether you are a legitimate customer or whether you are just a bad actor, trafficking money, or using stolen credentials.

So like in those use cases, having permission to use your data, you will online digital footprint is really, really, really helpful. So there's always like two sides of the coin of like how much data you share and how much information you gain and how do you tell apart legitimate versus i illegitimate users in different contexts.

Um

Mason Grimshaw: Yeah. A thousand percent. And I think, but that's what I think ultimately comes down to like having the conversation. I think the AI labs are so far ahead of everybody. They're so far ahead of like regulation and governance and the conversations people are having. Hey, if we all got in a room somehow, or elected officials got a room, it's like, [00:29:00] actually we want this, this is best for everybody. Um, we provide clear ways to opt out somehow for the people who don't. Awesome. But I think that's the issue. It's like we're not having the conversation.

Rui: hmm. Yeah. everybody is chasing right now as if it's a new gold rush. It is a new gold rush, but I'm not sure.

Mason Grimshaw: Yeah.

Rui: The gold really lies in the current a few years instead of potentially decades on down the road. Anyway.

Mason Grimshaw: Yep.

Rui: Let's talk a little bit about your work also outside of IndigiGenious and data sovereignty. 

Mason Grimshaw: ODE is a consultant agency for the environment that I think that, fits the model quite well. People bring us problems, um, and we solve them. And these problems can be very different. From each other. but they're all having to do the environment in some way. So we just finished up a big project towards the end of last year, um, called Blue Corridors with the WWF, where [00:30:00] we helped them get some, whale tracking data out to the world. And the interface that, our front end team and designers made us just beautiful.

my role in that is to do a lot of the data processing. So I was working with the scientists, um, and the WDF to make sure that they're gonna send in the data in the way that they have it, and then it needs to be kind of cleaned up, minor tweaks here and there, and then made ready for the web, in a way that is responsive and, and useful to people. That's one. a lot of those are geospatial in nature. And so what you'd really like is foundation model for the Earth that allows you to make predictions that basically provide you with a set of features that are sort of ready to go.

And by features, I mean, um, your input variables to your statistical modeling problem. So they might bring in the target and that target might be something like, um, presence or absence of aquaculture above ground biomass [00:31:00] in a particular square. And you want features that help you predict that thing. in the old days, it would be something like, The greenness would tell you how much trees there are. 

It would be kind of a hard problem. but with foundation models, what they learned to do is take, chunks of the earth and create those features of a way that is, general and allows you to, to go from picture to, a potential target much easier. Clay does that.

So clay is this sort of foundational model for the earth that allows you to do geospatial calculations faster. Without these custom, setups for every location. I did a lot of what are called the fine tunes of clay. So how well does Clay predict, um, the presence or absence of fin feeding operations?

How well does it predict above ground biomass? How well does it predict a absence or presence of aquacultures? There's other ones that I'm forgetting right now. [00:32:00] Um, but just trying to get a sense for how generalizable that technique is to all these things that people care about. 

Rui: just to be clear, that foundational model for Earth in general is trying to train on all kinds of data about Earth, whether it's satellite data, images, geo Yeah.

Mason Grimshaw: Yeah. it is, um, clay. Which I can speak to most specifically is trained much like, uh, image models of training. 

so you have this image, is essentially all these like squares of numbers stacked on top of each other. You mask out some of those squares, you provide that to the model and you say, okay. Predict that image, but fill in the squares this time. And so it learns what that image should look like. 

in learning how to fill in those gaps, it creates um, what we call these really good representations, semantic representations of the location and the image. And that is the thing that allows you to work on downstream tasks, [00:33:00] like map that representation to, whether or not your crops are growing well or whether or not you had a forest fire or whatever. 

Yeah,

Rui: Interesting. So let's say I'm a farmer. I care about whether my crop is doing well this year or not, compared to maybe other locations, uh, similar years in climate, should I just ask this foundational earth model Hey, I, I'm located at this, place, can you help me understand how is crop doing this year?

Mason Grimshaw: my take is farmers know very well how everything is doing. but let's say you're in charge of maintaining the Amazon for example. it's a massive, massive area. Lots of different jurisdictions, very complicated, right? keeping track of where things are being built, where mines are showing up, where deforestation is happening, scale is sort of a, the way that you would think about using these. [00:34:00] and you raised another great point, which is like the user experience and interface. So they're not to a point for the most part, where you can ask them. there's not a text bridge, at the moment, but people are working on this. we're in a research phase of that.

you could imagine, at the moment the interfaces are usually something like provide, some examples of deforestation, provide some examples of mines, and then the model can help you find other pictures that look like those pictures. you could imagine, know mining in this particular region, and you might click on those points on a little map, and then you say, okay, model, find the rest of them. And then light up the other areas that, are high probability of mines. 

Rui: Interesting. So for example, even forest fire, you must have tons of satellite images about forest fire that has occurred in the past around the world. And train a model that predict different various stages of forest [00:35:00] fire. I don't know which governmental body would use this, but they would want to set up alerts using this model to be like, tell me where is potentially having this early stage of fire and we need to attend to it.

Mason Grimshaw: Yes, I think you're right on the money there with like early stages of fire. Um, my mother in-law's fiance is a fireman and a lot of the work that they do is understanding where fuel is located. Um, where if a fire would occur, like where is it gonna be bad the most, like what can burn. and so yes, I think satellites could probably help with this. 

Rui: Is there a particular application that you're super excited about to try? 

Mason Grimshaw: we've trained a lot of models on it. one of my favorites, which is finding all the playgrounds. Which I think is fun. 

But then what happens when you, when you blow that out to everybody and what are they gonna wanna find? And I think a really interesting, proposition because people are gonna ask for funny things that you didn't even think about.

[00:36:00] Um, we have a lot of the resolution of the images isn't high enough for this at the moment, but eventually, maybe we'll get there. There's a lot of, little dinosaur sculptures in my town. I'll send you some pictures. Dinosaur Hill in Rapid City, South Dakota. Look it up. Shout out to Dinosaur Hill. but then we also have some in like different places. It would be funny to just find all the dinosaur sculptures because that would be a funny little dataset to have. I do think there's like an opportunity to do more and that only happens when more people are are involved, 

Rui: It will be very curious for me to see how this foundational model for Earth would evolve, especially what kind of product you can build on top of it to give people fun or help people learn more things. Um, I could even imagine it becoming part of the education tool. Um, maybe starting from your AI camp.

To help kids discover new things and especially if you marry the image, uh, information with the language or [00:37:00] textbook information to help contextualize whatever the textbook is talking about. I think that will be really fun on.

Mason Grimshaw: Oh, you just gave me an idea. Okay. So you just gave me an idea for a camp. So like taking our, plant locations and like predicting where we might find other versions of those plants.

Rui: Yeah. And where, like the new locations, do people use those plants Similar way? Because when you talk about, use it for medical, you know, treatment, I I'm from China and then growing up I, I mostly consume Chinese medicine actually only when the symptoms is severe enough that I have to, 

Stop it. then I would go to a, what we call Western Hospital 

I'm wondering have shared latitude, longitude, and also climate and if they have similar plants, are there similar use cases or how they're different? Um, I don't know.

It's just, uh, interesting.

Mason Grimshaw: Yeah. Very interesting. Like do all of your, um, plants grow in a certain [00:38:00] latitude band or certain climate regions? Uh, could you model that? 

Rui: I should, uh, say that there's lots of really good weather models too, which are like a different class of model, but we're sort of taking steps towards these models that understand earth really well, which is super fun.

Mason Grimshaw: Well thank you so much for your time.Thanks, Rui. Thanks for having me. This is really great.

Rui: This could be the last episode of floating questions, or it may not be either way. I hope you enjoyed flowing along with us today. If you liked our journey, please consider subscribing. Thank you for listening and may the questions always be with you.