How Skip made motorised pants that make you feel 14 kilos lighter | Kathryn Zealand

Your first product is called the Mogo. What is it? It's like an e-bike for hiking. So it's a pair of powered pants that have in them a motor that physically helps you move. Walk me through how hard that problem actually is. It is so hard. This is a whole new product category. We were just talking at lunch today about how easy it would be to work for a humanoid robotics company. What you all you have to do is make a fully specified well-modeled robot walk. Instead, we're trying to kind of build something that can almost dance with the human. Catherine Zeland, today's guest and founder of Skip, is in the Moveware game. She describes what Skip does by comparing it to an e-bike, put on their robotically assisted pants, and you feel about 14 kilos lighter. And if she's right, then watch out. Australian sport, something in the order of 50,000 e-bikes in 2019. This year, the industry expects some of between 250 and 300,000 bikes to be sold. In Europe, the industry projects 17 million e-bikes will be sold a year by 2030. E-bikes started as something for people who struggled on hills and turned out to be wanted by nearly everyone, including bike enthusiasts. And Catherine's betting, a similar thing will happen with clothing. Her team at Skip designed their exoskeleton pants to help people stay active, particularly the millions of older folk for whom the battle with hills, steps, and getting in and out of chairs is the biggest driver behind their increasing ill health and social isolation. In this episode, Catherine reveals the fascinating story behind their design and goes in depth on the curious machine learning and data journey that they've been on. You'll soon hear that they have amassed the world's richest dataset on movement. And what predicts what you're about to do next? Think of it like the movement equivalent of a language model. I also talked to Catherine on an even more personal topic, and it's what I want to give you all a heads up about. In 2025, Catherine and her husband tragically lost their baby Ziggy at birth. As she shares in this episode, many of these deaths are preventable with a little more monitoring in late-stage pregnancy, something that she didn't get because she wasn't considered to be a high-risk mother. It turns out that still birth is still far more common than cids as a cause of death, though you wouldn't really know it from the way they get talked about. What the two of them have done with this devastating experience is really, I think, a testament to their fortitude and their commitment to impact. They started a new company, Project Start Ust, which is going deep on designing lightweight, non-invasive at-home monitoring technologies. That could help detect risks while not stressing out pregnant women. I bet that a couple made up of a wearables robotic CEO and a machine learning professor might just be able to crack this. This is an episode packed full of insights from one of the greatest Aussie and Kiwi minds on data, products made for real-world impact, and design intention. I give you Catherine Zeeland. Catherine, it is such a joy to have you on wild hearts. Thank you for joining us. Pleasure to be here. OK, so I want to start with Skip, which is the business that you have founded. You say that you're in the business of building moveware. What is moveware? So in the same way that you might have active wear that you were playing sports and sleepwear, that you were for sleeping, this is garments that help you move. Specifically, we kind of invented this new category, which is a little bit like exoskeletons, but we wanted a new category because it's different. When people think about exoskeletons, they think about big, heavy machines. That's not at all what we're building. We're trying to build products that fit into everyday lives, that you think of like garments. And when you're deciding what to wear in the morning, you can decide to wear one of our garments, which are powered garments that kind of help you move. It's very much anchoring people into the new category. It's type of clothing, but it's clothing that's actually helping you move, helping you achieve a task. It's quite descriptive there. So it's better than a wearable or wearable is like to what end, whereas moveware is like I'm telling you the end that we're aiming for. Exactly, yeah. And I mean, there's a lot of buzzies days about wearables. And I'm actually quite skeptical of most wearables. Why is that? Because most of them are passive. So most of the time, it's giving you an insight. So it's saying, here's how you slept, right? Or here's some information about your activity. What you see is people are curious about that for a few weeks, but will eventually kind of stop looking or stop using it because it's not as actionable. So we're all about making products that are not just telling you something, but they're actually helping you achieve a job to be done, or helping you achieve a task, or helping you achieve something that you really want to be doing with your life, which in our case is often they're getting people out of the house, helping them move around their home, their community, their environment. Yeah, it's interesting with wearables. I don't know if you're wearing one of the rings that everyone in our sector wears, but it's always interesting to me what people pick up from those. And it seems like the kind of most common one is the impact of alcohol on sleep or sleep in general. And I'm always curious about whether or not more information actually makes people feel better or not. Or you see these stories of people being like, I thought I had a great sleep, and then I looked at my data and it turned out that I didn't, and now I've completely repositioned in my own mind what my own experience has been. So the idea of building something that also helps you do the theme rather than just learn about the things seems pretty sensible to me. So talk to me, Katherine, your first product is called the Mogo, what is it? - So it's like an e-bike for hiking. So it's a pair of pants, it's movewear. So there's a powered pants that have in them a motor and some cuffs that kind of hug your leg a bit like a ski boot or a knee brace that physically helps you move. So in particular this helps around the knee joint. And so it's most useful, you know, if you're trying to go up stairs, up steep kind of hikes and climb standing up out of a chair, movements like that where, you know, it's kind of a very knee joint heavy activity that we do in partnership with our tariffs. So for those who don't know our tariffs are kind of a really premium leading outdoor power brand. I feel like I don't necessarily have the best sense of fashion style. So it was clear from day one that I wanted a partner who didn't bring a little bit of that energy, but then also got a great team that, you know, do a lot of work with back facts and climbing harnesses and kind of had you attached in the body really comfortably. And so together we've made this product that, yeah, called the mogo, and it's all about getting people outdoors for helping people, yeah, access hiking, backpacking, trekking, things like that. - And why needs, like why was that the first place you went when you thought like I could build anything that helps people move, why is it the knee joint that you started with? - It's a great question. And for anyone who's somewhat familiar with the broader kind of exoskeleton space, there's now a lot of hip exoskeleton on the market. And they think that's actually the run value proposition. So the reason we pick knees is if you look what movements are most common, it's level ground walking. Most of the time people are walking on the flat. That actually uses a little bit more of the hip joint. And so that's why people think I should make hip assistance. This is a wrong question you ask. You don't want to ask what's the most common movement? You want to ask what is the thing people most want to help with? And so that's with what we do in a lot of our user interviews. And then you start learning the vast majority of people in this kind of missing middle demographic, where they can move a little bit, but they're slowing down, they're struggling with movement, they need a bit of support, but they're not kind of like a full spinal cord injury that you might picture in a traditional exoskeleton. So in this missing middle category, if you ask people what do you want help with, what they'll say is I can walk on the flat, okay, I can walk indoors, I can go for a 20 minute walk, but I really struggle with that big hill at the end of the driveway, or I really struggle with knee pain, or I really struggle going downstairs. It's a very difficult movement for some of our population. That's what they want help with. They actually don't want necessarily assistance on the flat. That's a one bit they can do. For a lot of our users that are trying to kind of unlock almost more activity, more exercise. So the goal is it necessarily to make all movement a lot easier. The goal is to say, hey, what's your constraint? What's holding you back from being more active, from being more social? And how do we releasing constraint? And for the largest population, that's the knee. I mean, that obviously people are very different and there are lots of other segments you could draw, but that's certainly why we started with knee assistance. And so why even that, like, you know, you've talked about the first product being a sort of hiking assistance type product, high-end associated with doctorics, but clearly you also care a lot about the folks who are struggling with movement as much as, you know, want the enhancement. Talk me through sort of skips approach to building product form movement. Yeah, okay. Our goal, long-term, is that the one in five people in the world that struggle with movement, have a skip product for them, right? And for some people, that'll be something like the mogo, for some people that might be something more medical, like we do a lot of work in the Parkinson's space, that'll be a kind of more medical, if you approve device, for some people will be a whole new product line we haven't made yet. That's a goal. One in five people with a movement disorder should have a product that allows them to live full and happy lives. There's a huge amount of variation in that. And you're right, like probably there's a lot of people who, you know, would need more support or more assistance and they'll get with this kind of first hiking product. There's two main reasons that we started there. So one is on the data machine learning AI side, which is, I'm sure we'll get into later, but we essentially need to understand what the human is doing in every minute. And humans are really messy. Like we would love to be working with robots, But we love we work. you know, with enterprise workers, where, or the military, where they're only doing the same task again and again. But if you think about, you know, an older adult who just wants to kind of walk their dog or kind of move around that community, they're doing lots of different things with their body. The way everyone sits is different. You can take weird, like, side steps if there's, you know, something on the curb you're avoiding. But we have to cover a huge range of movement. And so we wanted to pick an early activity in an early market, which gives us as much variation in our initial data sets as possible. I think it's great for that because it ain't like everyone takes different sides of the step. It's just a lot of natural variation, whereas if we did a first market that was in a warehouse, you would just get the same level ground walking flat, kind of straight lines, you would get a lot less information. That was one reason. I think the other reason is, this is a whole new product category. There's a lot we're going to learn. We wanted to start with a population where we're going to get great feedback, where the use is a bit more episodic. So where we can really learn it to rate. Yeah, for example, hikers typically will put on gear for the kind of two hours that they're out and about. There are fairly robust population that from what used to, you know, well, if you have a backpack, sometimes it rubs a little bit here or if you have ski boots, you know, sometimes they're a little bit uncomfortable at the end of the day. We want to be collecting all of that feedback because what we eventually want to serve a population where maybe you're wearing this product for eight hours a day, not two hours a day, maybe you're wearing every day rather than occasionally. And at that point, the bar for it being comfortable is incredibly high, right? Like what you tolerate in a hiking boot for comfort, you don't tolerate in every day. And so we want to make sure that we really had time to iterate and dial in and get all this feedback from this kind of early only user group who are often very happy to give us feedback. So that kind of as we get into populations that are wearing it for longer, we're really dialed in and nailed some of that comfort. And similarly that we've seen every edge case, right? So going back to not just as hiking got a lot of variation in the data that we're collecting. But it allows us to be confident in what the model is doing and all those edge cases. And for the mogo, we can make choices like, okay, if for some reason we see a new movement that's not in our data set, we could decide to not provide assistance. We could say the same thing in this moment is to do nothing. It's actually incredibly rare. We don't really do that these days when we're starting to learn and explore us a choice you can make in hiking. It's not a choice you can make if you're you're providing support that a user is relying on, right? If you're kind of taking a significant portion of their body weight that they're expecting you to help them, you can't just say, look, I don't want to do I'm not going to help. And so I think for safety reasons as well, we wanted to have confidence and be able to say, hey, we've seen every edge case we've seen a million steps have seen a lot of different variety. We really know that we're able to support people 100% at the time before we go into populations where that will be expected. So I want to come back to the database and also the go to market strategy because I think it's super interesting and very astute of you. But just coming back to what the actual product does. So like it straps onto your knee, but looks like a pair of pants and for any of our UK colleagues that doesn't mean underpants, that means actual full length pants. And what what's actually happening? So what's the what's the sort of mechanics of the robotic assistance and what's it trying to achieve for its users? So here's one I've prepared earlier. Cool. So this is an example of the bit that goes on the side of the leg. And so for someone watching, they'll just see a pair of pants have trousers with this on the outside. And normally we have a motor cap here, but I've just taken it off so that you can see the what's in that is a motor. And this is the same kind of motor that you might see in heavy drone, like a large quadcopter or you know, some kind of ultra light robotic applications. And so again, part of the reason this whole product is possible now, but wasn't 10 years ago. Is the enhancements in underlying kind of motor and technology that mean you can make something much lighter than you used to. So this is on the outside actuator motor. In here, there's a battery and all of the smart electronics and then the sensors both in this bit and down here so we can kind of know where you're like position is. And then underneath the pant are a few cuffs, which made of cloud fiber. And so they look a little bit like a new brace that's it under the pant and this thing kind of quickly connect and disconnect from back so you can take on and off this module. And so then as you're taking a step, so imagine our person here is going upstairs with the foot just stood on a step like this and now they're trying to kind of lift themselves over. Or if you're actually even better, imagine someone is sitting in a chair. And so this is a kind of bent like position that they lead forward to kind of stand up. So we'll detect the very subtle movement of them starting to stand up. And as quickly as they can perceive that they've started to stand up, the motor will kick in because we'll be able to fence and turn it on, you know, much faster than human perception. Then this will kind of the motor will start extending the leg like this. It will turn on it and kind of push you up. And so it pushes up this lever arm that then through the cuffs on the back of the thigh. It kind of feels like it's listing you from the back of the fly kind of up and over to help you stand up. So how much of the weight is it taking like what kind of feelings does it give to the user? Yeah, the best description is that it does maybe 30 to 40% of a healthy person's muscle forces. So it's not walking for you. So you still have to do kind of the majority of the movement for users who have lost 30 to 40% of their muscle forces or their abilities. The idea is it restores them to full health to give you another contextualization of that about the difference in energy between kind of a very steep slope and walking on the flat. So again, you still have to do it, but it should make some of these hills much easier for someone who struggles a little bit with movement. And talking about, is there a constraint why that, you know, is the 30 to 40% a technical constraint or is that a design intention? It's both. So the more power you want to provide the more talk. The larger the batteries and the motor have to be the heavier it is, you know, if you've got a larger motor, then you need more battery to power it to get the same battery life and so everything just gets bigger and heavier. Now, it gets you much closer to the kind of medical exit skeletons, which had existed for a while, but then it's like the road, it's like a humanoid robot that you strap a person to. And so it's a very different product. The most people, they do not want that. In fact, I would say the number one challenge of our current product is that it's too heavy, right? Like we're always trying to find ways to reduce weight and to make it feel more comfortable in the body. And so to provide more assistance, inevitably, you end up running weight. Is it also the case that the intention is to help people retain muscle strength and therefore like what you want to do is encourage the movement rather than the potential for atrophying or is that just a completely miss understanding of, you know, of what building a kind of robot for the body for movement looks like. I think it's two interesting concepts in that. So one is how much assistance do people want you right that the majority of people would not want more than what we're able to provide. We know this because, for example, our head of AI and controls, he did his PhD at Stanford, doing work where they would optimize the control algorithms of these huge access skeletons. So they had big off-board motor so they could do huge amounts of forces. And he said, you know, anything above 16 Newton meters is just really uncomfortable. Right, like it starts to feel like you're being married to the round. Yeah, and like no one really wants that in their average, you know, right? So going back to the resum populations where if you have really full swan with one injury, like you might need something like that. It's a solar minority for the majority of the population. It's not comfortable necessarily to apply huge talks. So yeah, if we weren't constrained by weight, I might say, well, a few more Newton meters here or there, but you wouldn't want to go all the way to full assistance. The kind of separate, I think, question you've had is about atrophy and is there something that's encouraging movement or discouraging movement? The short answer is, we don't see any muscle atrophy. The long answer is, no one really knows yet. Right, like this is a whole new kind of group product that we're inventing. There's no good longitudinal studies on what MoGo does for someone wearing it over several years. Again, that's part of the reason we want to start in, you know, slightly more recreation focused environment before we go into, you know, more mobility and pet populations. We have good reason to believe it's not going to be a big issue. And I think there are two strong analogies here. One is an e-bikes. So what we see with e-bikes is particularly to say older adults or for people who've started to see a reduction in how much they're cycling. If you get them an e-bikes, they cycle more, because what happened is, okay, there was that one big hill that stopped them from cycling because it wasn't fun. If we, you were the evening to tackle that big hill, they're now going to be able to cycle more frequently, go along the bike rides. And so by releasing the bindings and straight on someone's ability to move, you actually raise how much the rest of their activity levels or their muscles are contributing. And so the muscle groups that we offload with Mogo, they will probably get the same workout on one did before they started using an extra skeleton. But the person is now for that effort or that pain tolerance able to go further for the rest of their muscles get more workout. There is also more data from occupational exit skeletons, so kind of passive back braces or kind of shoulder supports for people working under cars. Those have existed for longer than the kind of product categorically working, the biomechanics similar and Karl Zalik who is out of the band about university. He's done a lot of work and found that in those situations again, you do not see. Instead, you typically see people able to do more work before they hit that point and so the muscles are getting a similar workout as opposed to less. And I imagine some of this. is also just about habit formation and the maintenance of particular sort of daily routines and habits to your point earlier about the person who's hit a sort of mobility constraint around walking their dog and that that becomes the beginning of a bunch of things that happen in their life, which might also be, I would imagine social activity and the hits to loneliness. I mean, I'm really interested in how you thought about the role that movement plays in well-being more generally if you'd be happy to share. Yeah. I mean, this is why I started thinking about the space to begin with. My grandmother was in her 90s and used to walk everywhere around her little village. She was such a cute old lady, right? Like she was walking to things like Slower Ranging Club, Bliss, Water Church or to like Puffy and Chap mornings, but she would walk there. And then once she started, you know, she had a couple of falls, she lost confidence and she stopped being able to walk, she'd just stopped going to a lot of these events. Although, in principle, she could have called the taxi, you know, the kind of extra barrier and the challenges associated with that meant that she didn't go. And so that was actually what got me started thinking about the space is how do we help people be whole active members of society, regardless of what their movement constraint and challenges are. And once people stop taking about a thousand steps a day, that's when you see huge reductions in their physical health and wellbeing, right? So there's a lot of associated kind of party basketball, but as you say, also their mental and emotional and social wellbeing, so movement is really critical to that. Amazing. And I imagine that that also is quite compounding and quite quick that once you sort of fall below that, it then is rare that people come back out of that unless they spend, you know, enormous energy on rehabilitation and have good incentives to do so. So I'm really interested to play in the space of your path to market, Catherine. We've already talked about how the mogo, which by the way, am I right that it stands for Mountain Goat? You are. You've done some deep research. Yes. Amazing. But I also like that it's kind of more go, you know, that's a lot of motivation. It's quite a fun brand. So, you know, was it always the plan? You've clearly have this like passion. You said that, you know, you started the company really inspired by your nine year old grandmother and keeping her moving. And you've got this interest in the sort of one in five people who struggle with movement. You've gone through this path to market via Arcterix, which is the kind of hiking population. Was that always the plan? Did you know that you were going to go via there as a sort of instrument to getting the data that you wanted or was that something you had to work through and discover along the way? A bit of both. So, it's always been obvious to me that we did not want to start in industrial or enterprise or military. That's partly because with very little amount of user research, you've even really different categories. And so, if we started in enterprise or warehousing, something like that, the problems you have to solve to make that product work, each and nothing about the problems you have to solve to make the product. We actually want to build, which is helping people in their communities work because a lot of the challenges left in the space are around how you deal with soft goods and soft humans. Right? Like, what is comfortable for someone who's skin bruises a little bit more easily because they're slightly older? Hmm. What is comfortable to find where every day? What is comfortable for the whole range of body shapes and sizes, not just people admitted to the US military? What do people want to wear? Right? So, you know, walking sticks are a very effective fall prevention tool. A lot of people don't take them because there's some stigma associated with taking a walking stick, taking a walking frame. People don't want to admit often even to themselves that maybe they need these assistive devices. You know, it was important to us to say, "Okay, those are the hard problems. It's like the human factor problems. How do you make your product like this desirable, easy to use, really quick to put on and put off very comfortable?" Instead, you're selling to a large logistics company where they care about, "Can you show that our employees are walking 20% faster and the filling 20% more orders, but they can just tell their employees to wear something like they would wear PPE?" It's almost too easy a problem. Like, this is, I think, part of my training at Google X coming through. It's like, you could build a product that and it teaches you nothing about what you need to solve to understand whether this category is going to make sense for the biggest market that we're really talking about. That had always been obvious to me. On some of the subsidies around, do we do hiking or do we do kind of everyday living as a first market? I think that was a little bit more as we started to think through, "Okay, when do we want to launch a product?" Right? Like, maybe for the data variability, it's good to start in hiking, but we could just pay 1,000 people for three days each to go hiking, which is very similar to the self-driving car companies, right? Like, it drives cars around the kind of data collection. So we could have done all of that before launching a commercial product, but again, it was important to me quite early to get market signal, right? Because you don't know really how products are going to fall in until you start giving it to people. Another interesting analogy I think is in the advanced prosthetic space, so you see a lot of fancy robotic hands, things like that, the abandonment rate for those after three months is in the high 40s, I'm going to say, you know, it varies a little bit. But actually one of our initial software engineers had an upper limb difference, and he had one of these fancy hands that would come out occasionally, but every day use, it was just quicker and simpler to put on a passive, much more kind of dumb, assisted product with passive ends to it. It's quick, it's easy, it's cheap, you can bash it around, that's exactly what he wants. And so some of this, like, how people are going to use your product in the real world? You don't know until you actually launch, and then you see, "Okay, people are not just users, but keep using it, what does it look like for someone who's integrating this into part of their daily life?" And so that push me towards, "Okay, we want to launch a product as quickly as possible, hiking has a very good context to do that, the post to, you know, maybe hiking has just updated flexion test ground, but then that will only delay, I think, us actually getting a product to market." And what has been the conversion rate? So how does it work? Like if I'm interested in using a mogo to do some hiking, what do I do? Go to our website, just skipwithjoy.com, and you can either pre-order the product now, or you can sign up for a rental or an experience, was started with rentals because it's taking us a while to kind of scale up manufacturing. And so if we ship these first few units to market, then you only get feedback from the end of 100 people you ship the unit to, whereas by doing rentals much like renting e-bikes, we can get hundreds of different users of different body shapes and sizes and different preferences into our product and get that kind of feedback. And so we're currently, we do rentals very frequently all around the US West Coast, and then we do occasional rentals and demos in other locations. So for example, you know, we had some of us go to Australia, and we did a handful of rentals around Sydney and Melbourne, last Christmas, you know, so it's a little bit like the timing for the international audience is a little bit more based on when we send someone to a particular location, but normally when there's sufficient demand, we can line up like three or four days of rentals, then we'll kind of send someone out to do them. And you said earlier that it's, you know, you've taken some inspiration from the growth of the e-bike market. I had to read somewhere that your early prototypes had this like 40% conversion from rental to purchase, which I think from memory was like two to four times an e-bike conversion. Is that right? Yeah. Is the answer, is the question, how, how are we that good? Well, yeah, well, why do you think that is? Because there's such huge demand for this category, and there's nothing that thirds it right now. You know, if you are living with daily knee pain or if you have some kind of cardiovascular fitness issue and you want to go further and you can't, there was just nothing on the market for you. And so I think when we came along and started offering demos and then people had a good experience, like there's this latent backlog of, oh my god, I've been waiting for this product. I get emails, multiple emails a day from people saying, can I get my mogul soon? I just pre-ordered, but you know, we've sold out of this six months of production. So if you, if you ordinarily, you're kind of at the back of the wait list, the movement is so important to someone and if you lose it, especially if you lose it earlier than you otherwise wanted to, if someone who's lived a very active lifestyle, you are desperate sometimes for a solution. There's so little out there that I think if you try a product and if it works for you, which, again, it doesn't work for everyone, unfortunately, particularly people with more but of severe or further along modality conditions, this is not the right product for you. But if it works for you, people are excited to get the product. So, yeah. What a good problem to have. I mean, interested then, you kind of alluded to this earlier, Catherine, about the kind of the human factor problems, right? So you've obviously, you've, you've got to tackle a bunch of technical issues around, you know, the weight of the product and how much talk you can put into it and how much you want to and how you collect the data from people. But you also alluded to this sort of adoption challenge around like, what does it look like? What does it feel like, you know, do people actually want it? You've had to make a bunch of design decisions around the trouser itself. Talk me through how you thought about the constraints there and the ways in which you've approached the visibility or invisibility of the assistance itself. That's a good question. In our early days, we prototyped and played around with, do we want the mogo to be totally under the pattern? Like, we want to basically hide it as much as possible. If we quickly think clearly, the answer was no, because by hiding it, you're telling people it's something that should be hidden, right? You're sort of implying that, you know, this is something you don't want to be showing off. You end up with some like, weirdly long beach levels, which no one wants. But instead of saying, no, we want to create an aspirational product. We want people to think and talk about their mogo, like they do their e-bike. Like, actually, my dad's a good example of this. I don't have somebody paying. He probably should wear a knee brace or, you know, treat someone for this. He doesn't, because he's, you know. a gentleman of a certain age, a dad, a dad. But he loves his e-bike, right? So like, when he got an e-bike, he started telling all of his friends, like, oh, you should see me on the trails, like I'm so good, you want to try my e-bike, you know, like, oh, you should get one too. There's something about the hiking community that you go community gear, where people get excited about gear, and they want to share their gear, and they want to tell people, like, oh, have you seen this cool new piece of gear that's come out? And so I think we want to tap into that rather than leaning away from it. So instead of saying, hey, this is something that you should hide, I want to say, what would it be like to get people to say, I just got my logo, it's really cool. Do you want to try it, right? And kind of lean into that. Now that puts an incredibly high bar on the aesthetics of the product. But if you can achieve it, it also means that you can encourage more of this organic virality, you can encourage people to be proud and excited about wearing it. And I think this goes back to the partnership with our paroxys really important, right? They have a lot more experience than we do, frankly, and how do you make something beautiful? Like, how would you make a product that people really feel drawn to, and they're excited to show off? And so they've been really helpful, I think, in my side of things as well. It's interesting, though, to think about the kind of overlapping the populations of your test, you know, using the word test population, but actually it's a legitimate population to go after hikers and very, you know, able-bodied active community all the way through to your ultimate market, which is the sort of movement constrained population that may tend older, may tend toward needing it for clinical reasons or assistance rather than kind of enhancement, as it were. How much do you think the kind of aesthetic decisions you take around people who do love talking about gear versus your 90-year-old grandmother? Like, how would she feel? How have you thought about, like, her aesthetic choices around using the version of Mogo that will evolve for her? Well, my grandmother was very cool. She was very proud of her, like, sparkly gold shoes. But I think a more serious answer is we look to eye glasses as an analogy here, right? Wow. Not everyone needs glasses, right? So they are an accessibility product in that sense. You and I think today are both wearing quite cool glasses, if you don't mind me saying. Thank you. Thank you. But, like, these are kind of, so these are both somewhat statement pieces, right? We've gone for, like, big, bold, like, the something that I'll personality here. You're as a blue, they are epic. Yes, I'm into them. Well, I mean, you've got the cat eye going, I'm into yours. This is maybe how I think about people's relationship with Mogo, right? Like, this is something you're kind of showing off and cool about. Now, the older adult, some of them will opt into the cool statement pieces, right? Some of them are really into that. Some of them may be not right, so you also see people with the much finer, kind of rimmed spectacles, which do disappear a little bit. They're kind of aiming to be able to want to scrape. Even for that population, the existence of the people wearing the bold statement glasses means it's a lot less noticeable when someone is kind of wearing a thin rimmed wire. You know, no one, no one comments on glasses anymore. You might comment on the type of glasses, but you would never say, well, that's since wearing eye glasses, they must have a problem. So I think, yes, like having this bold, first, inspirational product only makes it easier. Even if eventually we make products that are a little bit more discrete, and we'd hope to have a range of brand partners eventually, right? So this isn't kind of a one and one, but in the same way that your glasses are different, my glasses, there's MoGo, there might also be kind of other products for, you know, other demographics. And all of that, I think, just adds to normalizing the category overall. Yeah, that makes a lot of sense. So Catherine, we touched on it earlier, but I really, really want to come back to it because I think one of the most interesting pieces of the development of MoGo is the data itself and a lot of machine learning that you have put into being able to build a product that sits on a person, that they feel as an assistance, not a, you know, something that gets in the way of them achieving the movement that they are looking for. It is so hard. Like we're just talking at lunch today about how easy it would be to work for a humanoid robotics company. All you have to do is make a fully specified, well-modeled robot walk. Instead, we're trying to kind of build something that can almost dance with the human, right? So it's several things. And the first step you've got to understand what the user is trying to do. Like what are you, as Kate, you know, are you trying to, for example, pick up a heavy object where you want assistance standing up with your big bag of groceries, or are you trying to tie your shoelace? And if I'm doing the right thing, I'm bending with the knee, not with the backs. Correct, correct. Or are you trying to tie your shoelace, which might look similar from a biomechanics perspective, but if you're tying your shoelace, the last thing you want are your trousers trying to force you upright halfway through a bow tie. So the step one is trying to understand what is the user trying to do with very limited kind of on-body sensing and compute. Yeah, and obviously no cameras, right? So it's all picking up movement data. Correct, full movement data. And then even if you understand it conceptually what someone is trying to do, you've got to understand what is the right assistance to provide and how will they respond to it? So we're saying one of our biggest challenges is, you know, you can apply the same amount of torque to different people, and they will respond differently. So it's not just a like, great, I can, I know you're picking up a heavy machine therefore, like, you know, apply A or apply B, it's almost every kind of millisecond or ten, you're re-evaluating how is this kind of cooperative movement going? Do I need to back off? Do I need to add more and different people and very different preferences and responses? So it's a, I guess, order of magnitude harder than most robotics problems. I think another good analogy is it's like, okay, you're, you're trying to do a self-driving car. You're trying to build a self-driving car that's attached to a recruit, right? And so it's like, the use that can just randomly do like totally different movements and you're trying to like help it move in a certain direction, but there's a lot of uncertainty and chaos here, which is very different from how traditional engineering and kind of modeling approaches are defined. I love that analogy. I was thinking about self-driving cars because obviously this sort of my understanding of the pathway to getting to where Waymo now is in the city that you live in in San Francisco is, you know, you start with a very constrained set of environments and you build confidence around the ability to turn an exact 90 degree or to stop after a hundred meters or to pause when something comes across the road and then you introduce greater and greater complexity as the machine learns such that you can then deal with the chaos of the streets of the mission, for example, but you know, adding a raccoon where it's like, well, not only that, but it doesn't have there's no road rules and you're suddenly in a three-dimensional or not a two-dimensional plane is a good one. I think, you know, you alluded to this earlier, but, you know, using hikers because of the variance in their own movement is part of the value that you're getting out of getting all of this data back from the use of the rentals themselves. Maybe give me a sense of like, maybe this is a, it's too hard a mathematical question, but like, is there more variance between people or within the movement of an individual that you have to account for in the algorithm? It's even worse than that. There is also variance in an individual over time. Oh god. So I'd say in order, the variance in movements is the most, right? Because basically, if you are just standing still and doing nothing, you'll want zero assistance. And if you're picking up something heavy, going upstairs, standing up at the chair, you want max assistance, right? And so understanding where you are from zero to max, that's kind of the variation of what type of assistance you're pointing at different movements is number one problem. And that's where I think we've actually done a lot of interesting work. We collaborate with Georgia Tech, for example, and so we've got some nature papers with them on using machine learning to work in these really unstructured environments, but to still apply assistance across a huge range of different movements, including things that you would never meet in like a warehouse or a military setting like very kind of ad hoc non-cyclic movement. So that's difficulty number one. Then we go into number two, I think, is over time. And so in the academic research so far, we find that the number one, the terminant that saved how much talk someone wants is how familiar they are with exoskeletons. So for a while, you had this problem where the research is only done on grad students who work in exoskeleton labs. And they're like, yeah, we love high talks, like aggressive routines. And then you put that on someone who's wearing an exos for the first time. And it feels way too strong, very uncomfortable, they don't understand, and they're not liking this feeling, it's distracting, they'll end up almost turning the kind of assistance of a way down. And so we're having to do a lot of thinking, and this is, you know, we're currently experimenting, iterating using some of these demos, with how do we onboard users to this? Yeah, the kind of academic research says, well, it takes about six hours of exoskeleton before you kind of converge to an optimal preference. We don't even know, is that per week per month? What, you know, resume is not per setting, is it can we accelerate that? Right, like if we deliberately step people through, you know, okay, we're starting a little bit faster, and then adding, you know, a little bit more support over time, or we say, you're at a high level of support, but try these four movements, and get a sense of how it will feel to do these different things, so that you feel that when you're expecting it for the first time, rather than the first time you squat is like, when you're trying to do it for real, and then it might surprise you. So we have to be thoughtful around what that onboarding experience looks like, and how we pull involved with people over time, when you might kind of reset that, when you start again, so that's kind of one thing, and then underneath it as well, individual by individual variation. Those basic examples here is, if you're someone with knee pain, then we find that you enjoy more support going down stairs or going downhill. If you're someone who, maybe it's a cardiovascular issue, you've got no problem going downhill, you almost don't want much support then, that might [BLANK_AUDIO] bit distracting, but you really like support on the way uphill, right? And so those types of things can vary your preference. There's a more subtle layer of preferences which is just like what feels good to some people doesn't always feel good to other people. I had an AI and I have like an ongoing bet around what percent of the variation in preference is true psychological preference, which you can never know without asking someone to prefer A or B kind of giving them back to like test. What percent of that with sufficient data would we estimate based on how you walk, right? So at the moment, the field thinks it's a preference. I'm like, oh, you take just prefer more knee flexion because you like that and I don't. We have a hypothesis that with sufficient data, we might be able to say, well, as measured as how you walk, okay, and based on something, you don't know how to articulate, but the model can kind of learn. It means you're more likely to prefer additionally a flexion compared to me. More and mostly known in the world right now has data to answer that question, but we hope to be the fast. Yeah, I mean, I was even thinking like, it seems obvious to me that if I would use the mogo for a long period of time that you would learn things about me, I didn't know about myself. Potentially that, you know, most of us don't know that we carry our weight slightly asymmetrically, right? And I imagine that's one of the first things that you learn about a person when they're in the mogo or maybe that we are walking with a certain kind of gate that is actually again asymmetric or we lead with a particular foot. Most of us don't even know until you do that sort of test where you push a person and see which one comes out first. What have you found intuitively or, you know, what have you been able to be surprised by with users in using the product? It's a good question. I think the honest answer is we're still learning some of the stuff. So, yeah, I think the first time we spoke, we spoke about can we predict someone's going to need osteoarthritis treatment before they know it? I don't have enough confidence, yet, to say, "Oh, yeah, that's something we can do in the immediate time." But await the next nature paper for that one. Exactly. Exactly. Because at the end of the day, we're measuring how you're walking right now. And so, yeah, some people aren't aware, yeah, to your point, maybe that they're leaning more to the left of that, right? That said, a lot of the people who find up to try an extra skeleton for the first time tend to be somewhat thoughtful and already thinking about they know they have a movement problem, they've seen their PT for this, they're interested in how they walk. So, I don't think we've got really good answers to that scale, but I've got two kind of related anecdotes. One is that the very first batch of models that we trained, it were for much better on people turning left than turning right at the top of stairs. And we were like, "What is going on here?" And the answer is that a lot of the training data had come from the stairs in the stairwell and our workplace. And people were like, "You'd go up into the left, can you go down into the right?" And the model had just not seen enough examples of people going upstairs and then turning right. It's like, "No, you've got upstairs, you've turned left." Like, every stair has been trained on you've got up in your turn left. And so, I think there's a lot of that sort of thing that we're still a little bit working through before we'll get to the population level clinical information. The other thing is really interesting, so I don't know just like the right moment to talk about it, is our Parkinson's and full prevention work? Yes, please. Long term, as we said, wants to serve the one and five people with a movement challenge. So far, we've spoken a lot about Mogo, which is kind of one end of that. People who have quite a mile, they're just starting to hike a little bit less, but they're still fairly active. Okay, but for me to be credible when I say, well, also going to serve someone who's had a stroke, someone who's had a sports injury, someone who's an older adult, we wanted to start building out our core thesis around using machine learning to reuse a lot of the core components, a lot of the hardware, adapt to these very different conditions in the software layer, run in the hardware layer. That's how you unlock our ability to serve what's currently seen as a bunch of tiny fragmented markets, and instead have like a huge town, right? So that's kind of the other half of this platform thesis. I think that's quite similar maybe to how a single smartphone has replaced what used to be, you know, like dial up phone and then a stacks machine and then kind of a computer. It's like used to have different tools for lots of different use cases, and then you can just have one piece of hardware that's pretty generic and maybe comes in smaller and large form factors and maybe used to have some different configurations, but it's a lot smarter and able to do it more. So we want to do that for the movement space, right? Now to be credible in that claim, we have to show that we at least start to generalize to a very different, very difficult gate. And so I've chosen to work with Parkinson's, for people living with Parkinson's disease, partly because they have a very salient issue that we really want to help, partly very tactically, because they have something called freezing of gate, which is where they'll be halfway through kind of a movement and maybe they're going through a doorway or they're trying to turn, they're trying to do something where the brain has a little bit of a, oh, I can't quite figure this out moment and they get almost stuck, like where you see them, it's a bit like a starter or a tremor, kind of in their walking. And not that everyone with Parkinson's, but again, it's certain for these progression stages, this is quite common. And it's really salient. It's like we've had users who in their home trials without assistance, it'll take them 15 minutes to kind of stand up out of the bed, walk to the living room, turn around, walk to the kitchen, go back, because they'll get almost stuck in place multiple times in these movements. That can be very debilitating for your quality of life. And that's something that we know exascultants can help with. So, you know, we did some of this research internally. It's recently been done by Harvard, as well, showing a very similar thing where exascultant gate assistance can help people like break out and prevent these freezing moments. So, we think they're good population to work with, because they have a very acute, conditional experience that we know with high confidence we can help. But Catherine, just on that, how does that work? So, when you described how the mogo works, it's picking up the subtle signals of movement. You use the example of someone coming out of a chair and it might be like almost before I'm conscious that I'm doing it, you're picking up the data that I'm clearly intending to do that because you've seen the sort of biomechanics. If someone is freezing and it's a miscommunication between the brain and the physical body about what is trying to happen, is it that you know what the tremor freezing looks like and so you're sort of overriding that and saying, "No, no, let's keep moving," or like, what's the prediction piece there because presumably the body itself is not sending the signal of the intention because it's the miscommunication. That is such an astute question. Let me come at that. So, just to finish a quick thought before, sorry, that's the reason why someone with Parkinson's would buy one of our products. The reason we pick Parkinson's is as well as that, they experience fall and bounce issues at a higher rate than average. In fact, one of our participants says Parkinson's won't kill them, it'll either be choking or falling. That's how pop of mind this is to people. Long term, I would love to make a real dent on fall prevention. What would it take to actually improve balance to reduce falls? And to get, I should say, we're quite a waste of doing that. What's helpful about working with Parkinson's patients is it gives us a chance to study that very generalizable issue, which we can't promise that we're going to help them with that in the next year or two, but we're collecting all the data in order to learn and iterate and hopefully help with that while giving them a lot of immediate value in the freezing of gate. What's going to worry you pick Parkinson's? The answer, you're very smart question around freezing of gate. Correct. So, I've blessed over another hard part of this problem, which is when should an excess skeleton or when should a moveware product be opinionated and when should it be a bit more passive, right? So, with MoGo, it will always desire to the user, right? So, whatever you want to do, it's just there to support you. It will never, currently, with the current controls, override kind of something you're trying to do. For Parkinson's, there are moments where we do want it to have an opinion. We want it to identify, oh, there's something happening now that the user physically is doing, but doesn't really intend to do, let's help them out of it. Or, falls are another example of that. If the user starts to have a fall, we need to intervene strongly and immediately in something that is not necessarily user initiated because we have to respond faster than the user. Yeah, the user's indication is to fall and the intention is not to. And we don't want to do that. Yeah. And so, now there's a whole layer of complexity and it's partly why these more opinionated, you know, which is often more medical product will take a little bit longer because there's another layer of models on part, which often there's one layer that's running identification, right? So, is the Parkinson's patient experiencing a freezing of gate moment now or not? And so, we basically run on the edge, on in all times, a probability model, which is a probability of having a freezing event in the next few steps or kind of in the next X milliseconds. We can do slightly better than clinicians at that. As in, it's funny because the ground truth labels are a clinician watching and saying there's a freezing event between this time and this time. And then we can watch our probability models that's trained on that information, but you can see the probability of a freezing event start to creep up before the clinician have identified the start point. But this is maybe closer to one of your examples around, okay, like it's not quite obvious enough yet for someone watching the video to make the decision, but there's enough social signals in someone starting to slow down, maybe even within a step, you're starting to see them, oh, they're failing a little bit. So, then you say there's a certain probability of their being a freezing event either in this moment or in the immediate, you know, next few milliseconds. And then you run a different model on top. So, then you can have kick in a more opinionated decision maker, which will say, okay, the right thing to get you out of this problem that you faced, though it could be a freezing of gate, it could be a fall, right, very similar thing is this person, they're kind of running all the time model of saying, what is the balance state of this individual? Do we think this person's becoming uncertainty? And then instead of threshold, you say, okay, yet, now we want to have an intervention, like what is of the intervention. we're still figuring out the right intervention. So one of our clinical trials is comparing, okay, is it better if we actively throw the left leg out to help them initiate a step? Or are we better to just exaggerate the previous rhythm? Or do we want to kind of do more subtle versions? Like there's a lot of kind of baby testing behind the hood. But a lot of that has to be done with patients or with users because again, it's about how they react. So it's not good enough for us to say, or if someone's losing balance, we want to help them stabilize their core. We want to get their leg out under them. We need to know the right talk to apply for every individual in rare instances where their response in combination with the assistance we're providing gets a foot in the right place. And so for both of these examples, kind of freezing of gain and fall prevention, we've started this work, but I should also be honest with it, there's a lot of clinical work left to do before we have the confidence to roll this feature out. I can floor on scale, but that's where we're heading to. - Extraordinary. And you do that with the Michael J. Fox Foundation, I think, is that right on the Parkinson's work? - Yes, so the work's funded by the Michael J. Fox Foundation and then in collaboration with researchers at UCSS here in Parkinson's Disco, Georgia Tech University and at Stanford. - Yeah, extraordinary. So I mean, I think you glossed over this actually, Catherine, but you built foundation models to predict movement, right? Like, you know, there's so much discussion these days about the consolidation of foundation model design, but actually you've just gone ahead and built it. I'm really interested in you've been at, you've been at this five, six years with Skip sort of before, I mean, you even use the language of machine learning, not AI, which suggests that you've been doing it long enough that, you know, that has meaning to you. Many people still don't know what machine learning is, but we've all learnt the vernacular of AI. Talk to me a bit about the sort of evolution of the field around you and how that has shaped your sort of product direction and where you've needed to put energy as a team. Are there things that have suddenly become easier to work on because you've been able to benefit from the sort of leapfrogging in the market around you or is it just as hard a problem as it was when you started? - It feels like it's a harder problem. - Okay, there's a few questions in that. So one is we do benefit from some of the, you know, the externalities or frequencies, things like edge compute, right? The chips needed to run the kind of models that we run are becoming, you know, more, anything more freely available, but people are working on them and they're getting better and better. And I think that's, you know, because there's a whole industry of physical AI that's kind of growing up around this. Similarly, going back to the self-driving car analogy, a lot of how we think about model distillation, safety validation training, we can kind of pull that from the philosophies that have come out of self-driving cars. That said, no one else is actually working with our data modalities. And so a lot of the model training, we do in-house and is very specific to this kind of unique data sets that we're generating. So yeah, so I mean, we benefit in the ways that all startups benefit of just everyone's working our code is a much more effective because they're using AI tools. So there's a lot of that. But in general, I think the ecosystem of physical AI gives us a lot of infrastructure and tools and lessons to pull from, but a lot of what we have to do in-house, we were really early in that. So maybe another kind of angle in your question is, we spun out of Google X. Google X is all about, could this be a moonshot? Like what would it take for this to be, not just a unicorn like, but the next Google size company? And for that, you don't bother with making a product. Like none of our time that Google X has been actually building a product, instead you said, well, what would have to be true if this huge moonshot vision to be true? And one of the fundamental questions is, is there enough information in an individual's gate to be able to solve these, these models? So we didn't try and do them on the edge, we didn't try and kind of do an optimization, we were just looking at what would it take? Like is it possible, even? Who with the information that we can get from wherever sensors, understand and predict human movement to the fidelity necessary? I mean, spoiler it, spoiler it down for us, yes. (laughs) What we had to, like the models we could run then, we were running on, you know, Google scale, data center, centers, computers, all run offline. Then now we have to distill and run those models in much molar chips, using much less battery and energy, running them on device on the edge very quickly. And so I think the process of doing that, we benefit a lot from the broader kind of ecosystem that's developed around us. Yeah, so let's come to Google X, because you were there for, is it five years before, you know, spinning out skip? In many respects, it sounds like a job that blends venture studios, venture capital, you know, founder, led experimentation, talk to me about what the job actually was. Yeah, so my job at least within Google X was on the rapid evaluation team. So I was responsible for thinking what should be the next project that X kind of incubates? Like what will be the next way more? What's going to be the next big Google? It's a bit like venture in the sense of, there's hundreds of things, right? And the job is to say no to 99% of these, but to figure out what are the one or two that makes sense. It's more like a kind of studio or entrepreneurship in the sense that we'd build most of these in-house. You know, sometimes occasionally people would pitch ideas and things, but we would tend to run mostly investigations from the project internally. And so what that could look like on a day-to-day basis is to begin with a lot of just like getting smart about fields that I didn't necessarily know a lot about. So maybe to walk you through the skip journey, how do you decide which of these 100 different ideas could be a moonshot or which is only venture scale returns? There are a few criteria we look at. So one is, is it solving a big enough problem, right? If you're not going after a big problem, it's not going to be meaningful, it's not going to be a moonshot. We were looking at aging, aging in place, the one and five adults of the movement challenge, shifting demographics, like it's clear that something in the space is going to be huge and there's a huge problem with other. Another criteria is, is it the right level of technical maturity? And this is a funny one because the most common reason we wrote things out is that it was too easy. The reason you roll out things that are too easy, is actually because if it could be done by two grad students in a Stanford garage with AI coding tools, they're going to win, right? Like we want to instead find things which are complex enough that it would take something like Google. You know, Google is going to be a little bit more expensive and a slightly slower place to incubate things like I think everyone will acknowledge that. So it almost has to be worthy, like a problem worthy of that kind of technical excellence and the depth of experience that I'll for bet. Sometimes things like the convening power, et cetera, where it makes sense for Alphabet to do it. So it's often things that are really difficult, where if Alphabet doesn't do it, it might take years for anyone else to be able to, but at the same time, physically possible, right? So again, the other type of thing, you're all products out there, okay? Not actually. - Nice little narrow window you have there. - Not a little more, whatever. - You know, I think it's also a window that's nearing over time because when I joined Google X in 2017 and '18, something like that, there wasn't as big a deep tech ecosystem. Like when you wanted to fund a nuclear fusion project, there are VCs that will do kind of deep tech investment, was much, much rarer back when we started this. So when we started, I'd Google X was like, if you wanted to fund some of these kind of crazy technical instruments, this is one of the few places that would do it. So huge market to skip, obviously yes. Difficulty of the technology, right? So this is a good example where we'd had that inflection point in the motors and the batteries, where it was physically possible to provide this kind of assistant that we hadn't yet had the like full AI revolution and no one had shown that you could do anything beyond a treadmill. So at the time we started doing this, the state of the art was it is possible at a fixed walking speed in a lab in constant conditions. And I was like, okay, to go from that to 90-year-old granny, that is an incredibly hard problem. - Yep. - But physically possible, physically not impossible. That's kind of another criteria. You know, and then there are some others that we'd look at around, is it a good fit for Alphabet's kind of resources and do they have right skill sets to do it? And then is there enough value created somewhere? So we didn't have to know the business model yet. Famously Waymo was started as a project before they knew how they were gonna pay for it. And then Uber was invented kind of and the business model's obvious. But you had to show that you were creating economic value so that you could imagine a viable business model. So that's kind of how would evaluate projects? - What do you think you learned? So you went in there like you're a very, very unusual person, Catherine and I say that with the utmost respect. But like looking at your CV is as intimidating as it is confusing. So you started a PhD in theoretical physics. You then went and spent a bunch of time in Africa working on development economics challenges and big challenges for everything from general mutilation, how to address that to big public health concerns. And then you went to Google X and you know spent some time as this sort of in-house VC trying to move the needle on Alphabet's overall valuation. And then you started a robotics company. It struck me that two things were clearly true about you. One is that impact means a lot to you and you've got like a pretty grandiose idea of what's worth your time. And also that you mentioned this earlier like if it's too easy a problem, it's not worth doing from Alphabet's perspective but quite clearly that also applies to you as a human and what you've spent time on. What else drives you? Like what else has given you the kind of signal of go here even though it doesn't look obvious in each of your career moves? So I don't think it's a lot of work or problems that are too easy but I do think about where does it need me specifically versus the next in line smartest person? What's a good example of this? Actually part of the reason I left in the work in Africa is I was like, hmm. The end of the day I think this job is a really impactful important job but it could be done. By the like local country manager that I can hire to replace me, you know, there's nothing specific about my skill set That's like really impactful here. That's also the reason why I ended up picking skit. So when I was at X I had a huge initially huge portfolio and then we kind of down selected the hands of projects that I've spent six to nine months on Some of them still really successful kind of all those secret projects within X right so there are things that I worked on that Arguably a kind of more likely to become the moon shotman skip, but all of them had a more obvious natural owner Right like one was a pure software play and I'm like, I'm not really a software engineer That doesn't make sense for me to do like I could tell that skip or the project that became skip It wasn't going to be solved Like kind of a typical quote unquote person because to get this right you needed yes the machine learning story and a really hard data I play but you also needed someone who was thinking about what a typical user would want like you know Thinking about grandmother having that kind of user perspective really center There was like some soft goods, you know, there was a lot of like physical mechanical engineering I'm not particularly good at any one thing, but I think I'm quite good at having that more generous Based line perspective of like if no one like if I didn't take this project I didn't think it would do as well, whereas all the other projects are better handled by other people So I do think a lot about where can I incrementally drive impact? That's one thing you can also got myself a show of a time It's like that idea of having grandios impact when I was in that I was like my goal is to maximize my personal impact on the world Right, and that was how I ended up doing a lot of this economic development work Someone I was doing a job that I think ticked that box So this was with what at the time was called the Clinton Juicester enterprise partnership But they would do small angry business investments So if like a million dollars they'd build a processing facility someone that didn't have kind of the agricultural processing And then they could pay about 10 to 20,000 farmers Double the income that they'd be getting from buy of alpha food raised amount of poverty eventually For a million dollars a huge economic impact like yeah I've like found my like maximum kind of value per dollar the algorithm looks good. The algorithm looks good I just wasn't happy, you know, and I think it's a kind of job that you can do in your 20s And then you hit there and you're like Just a bit lonely like it was a really hard it's like my quarter life crisis I tell people because they've been so driven by maximizing impact But these were very rural populations. So often they didn't speak English My job was very transient because I'd kind of go and help diligence something But then we'd hire a local manager to actually stand up the factory and then I move on But I didn't really have a build that sense of community part of coming to X and coming to the US was like Yeah, I really did love that impact I really wanted to keep something that felt really meaningful But is there a way to do it which A is more unique to my perspective leveraging my science background which kind of has how the Google X felt like a good fit B is at least in one location like somewhere I could imagine Having a sense of community a sense of family and it took me a while for me to admit to myself that that was important But it probably is obviously like it's not sustainable to just optimize the impact How dare you choose to have community around you as a part of your decision making matrix I know I have to come back to this notion that some of what's Made decisions for you is about why me Like there are no end of big problems that deserve solving in the world But you clearly have this like quite honed sense of It has to be Some combination of my unique experience brought to bear on it that's worth me doing it versus any other next smartest person to use your language You've recently made a decision To take on another big project on top of skip Call project status which you know you feel has come out of the why me but not necessarily from a good place Which is that about a year ago you lost your son Ziggy Atful term which has ignited a whole Area of exploration for you bringing together robotics and machine learning And impact can you tell me a little bit about project status? I would love to and thank you for asking You know, I've been happily living my life as kind of a founder of skip and I didn't think I was missing anything in the world But going back to I selfishly moved here partly to raise a family And so we started trying to have a family and got through the first my master no problem What I was free of miscarriage risk right that's what everyone says went into labor At term, you know expecting a healthy baby Instanted hospital at the time they kind of hooked us up to the monitors We already had lost a heartbeat and so the tragic thing is like I've done had passed In utero probably at some point in the hours You know leading up to kind of going into labor. I hadn't even known And it's I mean it was just awful But it's also one of those things which I'm so sorry Actually, I just like wasn't on my radar, you know, I think I'd been Been worried about other things about being pregnant I was like how am I gonna cope as a mom, you know, I was like the idea that I could be in the world's best hospital Have not even known that it was already too late was really upsetting and I think You know, we went through a bit of a grieving period and then thought okay, what's something we can do to honor our fun Figgy and you know, we'd set aside a little bit of money for childcare and things like that and we said Let's maybe fund a PhD student or let's you know do some angel investing in the space like find something in his memory That would actually help address the situation And so started just trying to get smart right like I feel like I knew nothing really about, you know Feet of medicine before this The more that I learned the more it was kind of shocking Both how big the need is and how far behind we are so one in 150 pregnancies and in stillbirth So that's either kind of like in our case you maybe you you lose a baby after it was viable So kind of that's the different thing miscarried but before labor or or during childbirth and you know Often kind of the during labor during childbirth. These are maybe fragile babies to begin with so it's kind of a related Challenge because it's really common and it's just not talked about that much. Nice partly because it's really tragic I think for everyone involved including for the doctors, you know, like they were also crying right like this is a bad day for everyone I think what's even more sad is it's actually quite preventable and we know this because some of the existing Monitoring tools mean that when we do detect basically that the baby is thick then you often will deliver early Or you know deliver by c-faction or you can kind of give a medication From quick start to kind of show you how how easy this should be in the u.s There's a cutoff where when woman turned 35 like considered advanced maternal age, which we won't comment on or geriatric pregnancy As I have heard right that was being the older term to be the less laughing So everyone past that age gets extra monitoring that means a few more ultrasounds you might go for a weekly kind of heart rate check Everyone post-35 will get that The stillbirth rate actually drops That kind of moment right so the difference between the 34 and a half year old or 34 and 3/4 35 year old you see a reduction in stillbirth. Wow, which is and find that we know that this monitoring works like it catches Babies that are looking sick or are looking tricky and then there are kind of various clinical pathways We know that this monitoring works Take know I'm even monitoring is really basic So our chief medical officer is Aris who's the head of fetal medicine at Oxford University So this is by the way an incredibly smart and credential team working on this his analogy is After a baby's born you wouldn't imagine a world where you put a newborn baby in a room and then only open the door to check on it once a week That would be crazy, right, but that's basically what we do for prenatal care, right? Like, you know the babies in utero and maybe once a week of your lucky you go to the doctor appointment and they'll check the heartbeat It's okay. Our case is such a good example of that. He was really healthy a few days before we went into later But there was nothing to live in between that And in fact because we were considered low risk We didn't even have basically any ultrasounds in the third trimester because in the US at least That requires a synographer and a hospital room and so they can only offer it to a small number of people So the bare minimum what we're trying to do with project startups is take that kind of monitoring where it's technically already feasible But because it currently needs a human like a synographer and a hospital room It can only be offered to a tiny fraction of you know high risk moms and say what would it look like to make that universal So it takes some of that technology But make it possible to be prescribed or bought by any mom who's kind of interested in once one and the news Essentially AI and algorithms to do some of that early triage right so we might say okay We can tell from the heart rate tracing everything's fine or okay, go see your doctor But the bare minimum extending the current standard of care to you know to everyone Um, and we know that that will save lives because already dealt with the people who get it Then you stop say well actually imagine if instead of Checking in on the baby for 20 minutes every week We actually had these longitudinal continuous measurements of things like the fetal heart rate the fetus movement Other babies growing and because I'm evolving We would actually be a lot better equipped over time to understand the baby's development and then went into vain and what to do And a good example of this is like for a second so actually everything I've got a second child. He's fit and healthy Congratulations. Thank you But him because we were considered high risk, you know, it goes for these heart rate monitoring kind of they call them a non-stress test appointments Maybe just 20 minutes and they look so do we see enough Accelerations in the heart rate that we confident the baby's kind of moving around But it's 20 minutes and at one point we didn't and so they had to admit us into hospital We kind of get another four hours of monitoring before they said okay. We've now seen enough movement We think you're okay you can go home. That was incredibly Stressful of course nerve-wracking expensive thin was probably sleeping happily and it just happened there was the 20 minutes that you were like trying to make sure that he was happy and well. - Exactly, exactly. And on the flip side, I've got a friend who, you know, was the lowest pregnancy, but at some point, wasn't sure when she'd lost about movement, didn't know, was it a bit worried? She also took half a day of work, broke to her hospital, got checked out, everything was totally fine. And so, for people who were anxious and kind of seeking the three assurance, if we had just continuous measurement of how the baby's doing, the individual trends, the lot like what we're talking about was skip. Okay, we need to understand how frequently does your baby typically move? What if it's average movement happens? Does it move more in the morning or the evening? And some moms are really in tune with that. I was not, right? Like I was bivvy, I was not thinking about it. You're on stage with hard fork, you know, showing off the ex skeleton. - Exactly. - Yeah. Exactly. Yeah, so building the hard vote, eventually be the form of a wearable, which can provide some of this monitoring technology in something that looks and feels maybe like a glucose monitor, like a CGM, that gives moms, and their doctors, both information for the current standards of care and then over time really furthering our understanding. And we think we can avoid not just kind of up to half of skill births, but also a large number of pre-temperth, pre-eclampsia, things that end up in kind of nico-stays and complications for the mom or the baby, even if they don't kind of end this kind if we have asked what we did. And draw gold as to over the next few years, launch those products and to try and overt, you know, what happened to us happening to other families. - I mean, Catherine, it's extraordinary. I'm, as you know, eight months pregnant at the moment with my second and so it's very top of mind for me. The, as you say, the kind of things that you hear doctors say, which is like, are they moving well? And you're like, what is well, you know? And what was yesterday? And even when you think you're being in tune with your own body, like, it doesn't always seem obvious how you would answer a question like that with confidence. I'm really interested if I was a, you know, I know your early stages of thinking about what the productization of this looks like, but what do you imagine some of the design considerations that you have in there? One of the things that comes to mind, of course, is that some wearables, and we talked about this earlier, like, sometimes data is not helpful, particularly in the mind of an anxious or worried potential parent. Have you, have you had to think about the role that information plays for a non-expert user, where the stakes are so high? - Yes, we spend so much time thinking about this. And actually going back to, you know, why have we done this? So it's obviously a little bit maybe my background, I make wearables, I make techier approved devices. This is kind of an actual product. My husband happens to be a professor at Stanford in machine learning, focusing on how to hate decisions with noisy data sets and how human plus algorithm decision makers can outperform humans. It's his area of specialty before any of this happened. You know, he hadn't applied at the woman's health, but he's kind of from the theory side. So that was part of the reason, like, those are the two hard parts of this problem. And we were looking around to fund another startup and I was like, well, hang on. Like, it not us, who would do this? - Statistically unlikely to find the overlap. - Exactly, exactly. Eris, he's done a lot of thinking about this. And we've got Rachel Cummings, who's a professor at Columbia has also been thinking. - Yes, I think one thing is to push back a little bit on I think some of the paternalism that is inherent in the space, where it's like, woman, I think, should be entitled to this data. It's about their bodies, their health, their children. Like, yes, we want to be thoughtful around how we show it. But I think we should also trust people that, you know, this isn't going to be a product for everyone, right? Like, if you don't want this kind of data, you don't have to take it. But there's a lot of one where the lack of any data is an idea and choosing. - Yeah. - Right, particularly for my second, it was a time between appointments that was really stressful, right? It's kind of not finding any information that's really, really tough. And I think we should give women the credit where they'll be able to cope with a little bit more information. - I don't know, you know this, but when the pregnancy tests were first introduced, there was big pushback, particularly from overdue weigh-in, and thinking, oh, women are going to freak out, they're not going to be able to cope with having this information without the doctor there. - What, knowing that they were pregnant without the doctor, right? - Correct. - Correct. - Okay. It turns out people are fired and coping with this data. That said, we do want to be thoughtful, right? And so I think one thing is staging a bit of a product rollout, where we will not be diagnosing medical conditions with the first product. Instead, we'll be tracking your baby's movement and vital signs, and I've got a garment watch. So everyone would give me a sleep score like we're talking about kind of the rings, did the same. And it will sometimes say you have an outlier metric. All right, we might do something similar where we'll say, hey, talk to your doctor about this, or oh, we think you might be at risk for some of these factors like go talk to your doctor, but the first product will not necessarily be diagnosing anything. I think we also think around giving parents kind of reassurance that everything is okay, is a much easier thing to do. And we're comparing to the current fans that have cared or stayed at the art, which is very little. So it's totally that either as you buy, you can get what they call them at home dopplers, which are like cheap heart rate fences. They're really bad. There are tragic examples in both directions of someone not able to sign the baby's heart rate, finicking, you know, go to the doctor and everything was totally fine. It's just these are hard kind of to use tools. Or, you know, someone who found the mom's heart rate, thought it was a baby thought everything was fine. Actually, unfortunately, the baby had already passed. So the current tools are terrible, but most of them don't even buy that. So what's more common is that if you start to worry and you're anxious, you go to the doctor and you can have to get a checkup. So again, this is, I think, just empowering woman where we know everything is okay to be like, yep, you can put on the start after you've got it 24/7 in your app, you can have a look and it will say, you know, by the signs are normal. Everything is within range for your baby. There's no reason kind of for more concern than normal. Well, we might say, okay, where, you know, detectings are not lying metrics. You know, clever telehealth call with our doctor or go to your own OBGYN, ask these questions, kind of get some follow up is kind of what we'd expect. And yes, we, you know, there will be some false positives. There'll be a hell of a lot more, I think cases that we catch where we can do something early, which would otherwise go undetected and much like the self-driving cars. You know, you're not necessarily able to have that everything locks in it, but if we can overall improve safety and save lives. So you work that even if you can't kind of save every case and even if occasionally you apply the emergency brake unnecessarily, it's much better than kind of not having these products at all. Completely, I mean, I think Catherine, it's extraordinary that you've been able to take what is such a tragic experience with Ziggy and build something extraordinary out of it for the benefit of many others. How have you thought about that putting you in the line of being in the position of talking about him and your own experience so often in order to get the traction and the beginnings of pulling a team together and I know that you've handed over to another CEO, but you still have a role to play in the organization. You know, that must be quite a strain on you emotionally. - Yeah, I mean, it can be, but I think silence is also really hard. And so I talk to a lot of other lost parents, right? Because we're so public about it. You know, people who've had similar experiences will find us and some of them will say it's been years since anyone asked them about their baby. Or, you know, been years since anyone said their name, right? And I appreciate you've been deliberately doing it the last few minutes. And so I actually think for us, it's maybe being of anything a little bit more therapeutic to not just be able to talk about him and our experience, but have something positive that's come out of this, right? Like there's some legacy he's leaving, whereas for most people, it's just a tragic event. Like there's, you know, when an adult dies, you can say, well, they lived a good life, here are the things I accomplished, you know, they leave behind kids, all this. When a baby dies, there's very little kind of redeeming about the situation. It's just sad. But I think for us to be able to bring something positive out of that experience isn't itself a little bit healing. And then to meet other parents who've gone through this and talk about their experiences and our experience, what's been different. So, yeah, this is a little bit vulnerable sometimes to kind of be out there. It's gotten easier over time. Like I think, you know, as a podcast I did quite soon afterwards, I started crying, half of these are podcasts, but I think we're able to talk about it a bit more. Now, it's also been really, you know, now that we've had a second baby in some ways, that's been healing. But in other ways, every milestone that's in our current sun hits, is a reminder of the milestone that we should have had with his older brother, right? And that kind of loss and pain doesn't go away, but yet the silence around pregnancy loss persists, kind of for a long time afterwards. So, yeah, I mean, I appreciate you being when they talk about it. And I think if anything, we need a little bit more because people acknowledging that this is happening and kind of being supportive and more awareness around it because it is both surprisingly common. - Yeah, I mean, your start about one in 150, I think we'll be surprising to many people who are listening to this and will cause them to query who do they not know might have gone through this in their own network. Catherine, it's totally inspiring and I thank you for that work and for your capacity to take what is a horrible experience for you somehow pair that with your unique skill sets, the skill sets of yarn, your partner and build something extraordinary out of it. My final question for you, Catherine, for the podcast is what something you've changed your mind about? - Oh, that's such a good question. I mean, just because we were talking about family, I was not sure that I wanted children actually before we had a figgy 'cause I'm a very technical nerdy, like a personality and another big fan of babies, ironically, given we not working projects out of us. You know, once we had even having Ziggy, I think what's especially tragic maybe about kind of still birth rather than kind of an early stage pregnancy loss is you still go through labor and so you still have all of those hormones that make you think like, oh my God, it's just like the cutest human, the best thing I've ever done, like this is the love of my life. and you're at the same time, kind of does a lot of grief in that moment. I hadn't been sure that I wanted children. I think until we went through that, and although it ended in tragedy, the one thing that was really clear to me is, okay, I definitely want kids, like this is, but I feel like that's a bit of a, that's not a very good answer to your question. What have I changed my mind on? - No, I think it's quite a counter-intuitive answer for many people that may be enough, certainly if they were uncertain to say, and now I've experienced the worst version of attempting that path for many people, I would imagine that that would have pushed them in the other direction. - Yeah, it hasn't changed. I'm still not a baby person. Like, our neighbor comes over and she's like, "Oh my body is so cute." You know, like she's like such a baby person, and I'm like, I find him cute because he's my son, but I have the assistance of hormones that make me think that my slightly alien-looking child is the most gorgeous thing that I've ever seen. - Right, they're all kind of ugly potatoes, you know? - They are. - So that still builds the same. - Yeah, amazing. Before I let you hang up, is there anything that you would like to cover that we didn't in this conversation? - I always like to give a shout out to Anna, my co-founder, who came with me from Google Apps, and actually in those early days, going back to when we started, and I didn't know anything about robotics, I went to a lot of conferences met with all the best academics, and one of them, Elliot Rouse, who gets sabbatical from his full-time professorship to come and help us get smart about horrible robots and start kind of what became skip. So, it's not been just me on the journey, there's been a lot of blood slain tears for many people. - Catherine, thank you so much for joining us on Wild Tarts. - Well, thanks for having me. - Thank you all so much for joining us for another episode of Wild Tarts. If you want to learn more from other ambitious people building, designing, and creating the world that we all want to live in, then please hit the subscribe and follow button. It would mean the world to us, the founders, the operators, and the investors, who join us on Wild Tarts. This podcast is a labor of love from Blackbird and our production team. The show is produced by Joel Connelly of Blackbird, our marketing genius is Laura Coford, and our editor is Andy Jones. Thank you all so much for listening, and we'll see you all next week.