Osmo is Giving AI the Ability to Smell — Alex Wilschko, CEO of Osmo, Explains

Announcer (00:02): Data Masters is the go-to place for data enthusiasts. We speak with data leaders from around the world about data analytics and the emerging technologies and techniques data savvy organizations are tapping into to gain a competitive advantage. Our experts also share their opinions and perspectives about the height and over height industry trends. We may all be geeking out over join the DataMaster podcast with your host Anthony Deighton, Data Products General Manager at Tamr.

‍

Welcome (00:38): Welcome to another episode of Data Masters. Today's guest is Alex Wiltschko, chief Executive Officer of Osmo. Osmo is a digital of faction company on a mission to give computers a sense of smell to improve health and wellbeing of human life. They're bringing an unprecedented combination of hardware, software, data, and capital to the historic challenge of giving computers a sense of smell. They believe that scent is the last sense left to digitize. Alex is also an entrepreneur in residence at Google Ventures, an organization that supports innovative founders and typically invests across the life sciences, consumer enterprise, crypto climate and frontier technology sectors. He helped to found the deep learning team at Twitter after they acquired his company Wetlab in 2015 and then moved to Google where he worked on Google Brain. He has a PhD in neurobiology from Harvard University. Welcome Alex.Alex Wiltschko (01:32):Glad to be here. Thanks for having me.

‍

Anthony Deighton (01:34): Great. So let's start a little bit with Osmos founding and according to an article and Wired, you've been obsessed with scent since you were a teenager. And what about bad obsession led you down the path of neuroscience and then ultimately I opposed to Osmo,

‍

Alex Wiltschko (01:52): It was a slow compounding process of an obsession that wouldn't let go. It just kind of got worse over time. I guess. First it started with just being very sensitive to smell. It's just the window in my senses that's the widest open. And I found this book Perfumes the Guide by Luca Turin and then later co-author Tanya Sanchez. And it was just review after review of perfumes and they were like poems. And I got obsessed with these perfumes and I started collecting the ones that I could find and afford when I was a teenager, which made me of course very popular in the small Texas conservative town that I grew up in. But I just couldn't let go of it and I didn't necessarily even think that I could be a perfumer. It just never even crossed my mind. But I knew I could be a scientist cuz I was raised by two scientists.(02:32):So I did the logical thing to me, which is go study how the brain constructs scent. It just seemed like this incredibly potent mystery. And so I did my undergraduate in neuroscience and then I went to Harvard to study neuroscience cuz that's where all olfactory neuroscientists were and still are. And just took it all the way to its very extreme, which is kind of my nature, which is, you know, just do it all to the very maximum that's possible. And what I learned at the end of that process, I had kind of gotten a of this throughout my training, is we actually don't know that much about smell. We don't know why things smell the way that they do and we don't know how to organize the world of smells. And uh, that really, really bugged me and it that's been the journey that I've been on professionally for the last 15 years.(03:14):And you know, I started in academia and kind of hit a ceiling in terms of what I felt I could do there and got sidetracked and you know, entered into the world of entrepreneurship and artificial intelligence. I just got really lucky I think of if I'd have made that transition. Now the people doing that right now are so much smarter than I am and then moved into industry, uh, first at Twitter and then at Google and Google's where I founded the digital Olfaction team. And I had a lot of leeway to try really interesting new ideas because Google Brain at the time, who knows now today given that it's um, now Google DeepMind. But uh, we tried a lot of interesting small things and I got to try out digitizing smell and we hit our milestones, right? We actually built AI models that can predict why a molecule smells the way that it does and we've published those papers.(03:59):Uh, one just came out in eLife yesterday. We've got another one in review and it works. We built an AI system, a neural network that can predict why a molecule smells the way that it does at superhuman accuracy. So we've passed a kind of an odor touring test and when we hit these milestones internally at Google, we asked ourselves what's next? How can we take this further? How can we go bigger? And in looking across all of Alphabet, working with the office of the C E O, working with the head of research Jeff Dean, it was clear that the right route was actually through GV or Google Ventures. And we also found a wonderful partner in Josh Wolf at Luxe Capital in all of us together, alphabet, GV, Luxe, myself, and then a fantastic founding team. We started this thing, we started Osmo, we spun out out of Google Research and raised a series A and we've been off to the races ever since it's been seven months. So we're still very, very young, but I'm thrilled to work with the people that have signed up for the mission.

‍

Anthony Deighton (04:54): So those are a lot of great themes I think we'll pick up on. But I want to go back to something cuz start in academia then you've also worked, it's really some of the biggest names in tech, you know, Twitter and Google and you've sold two companies, so you've sort of done both sides, the startup route and the big company route. And now you're doing maybe, I don't know if you would characterize it this way, but this kind of middle round, uh, with Osmo maybe from a career perspective. I'm sure many people listening are thinking about, you know, is it better to go work in Google, you know, some of the big tech names to do my own startup to work at a startup. So how do you think about that distinction and then related to taking the plunge to start Osmo several months ago, how do you think about that from a career perspective and from a listener's perspective as opposed to yours?Alex Wiltschko (05:39):It's a great question and you're forcing me to reflect and, and maybe what you're gonna get is sanitized and a more intentional sounding version of this story arc because it really wasn't the whole time I've just been chasing my interest and the main interest has been sent. That's been the flowing thread throughout my professional and also in some sense my personal life. So that's always been dragging me into the future. But I've also been interested in lots of other things. This whole AI thing when it started out and was blowing up just seemed like the coolest thing in the world and I just wanted to be a part of it. It broke my heart a little bit to get away from pure science and to move into more of like a product and research and engineering blended role at Twitter. And then at Google I had imposter syndrome.(06:22):And so the first, you know, two years were, was really about me proving myself like can I engineer systems at Google that get shipped? And I did that. I led one of the compiler teams in tensor flow too and we got that out the door and that really helped me feel like I could, you know, I could exist in that world but after all those pieces that I just got dragged back to my personal passion, which is sent and really I just would do it to the logical extreme that was possible in the context that I was in. And what I kept learning is that sometimes you have to change your context in order to keep following your obsessions. And so it's not been planned as in, you know, I'm not looking at my resume or CV and saying what's the next hole to plug? What I'm thinking about is how can I get this thing that's burning inside of me trying to get out, which is this obsession about why can't computers smell, what can I do to make that real? So that's been the the overarching, that's how I think, not necessarily planned, but that's how I add.

‍

Anthony Deighton (07:17): Yeah, and I think that's a really interesting insight. This idea of plugging holes in one's resume is something I hear from a lot of young people kind of earlier in their career and the really thinking about the job career arc, almost like as a game, like I've gotta figure out I got badges as I go along.

‍

Alex Wiltschko (07:33): Look, I get it, you know, like we all want to collect our merit badges, we all want to play Candy Crush and you know, get all the high scores like we're built for that kind of gamification. The really, and I think there's some value into experiencing that and winning that I can build some self-confidence, right? So if you get that really cool job, awesome, you know, that really can build a layer of self-confidence that you can rely upon later in your life. But really hard things aren't gamified, right? They're really interesting. New things have no reward structure around them because they don't exist yet. And there's nobody else that's built this, you know, casino for you to tell you that you're doing a good job. That's the hardest part. And the true emotional reality of doing something new that you're obsessed with or that you really need to bring into existence is there's nobody that's gonna tell you you're doing a good job, but there's plenty of people that are gonna tell you that they don't recognize what you do and maybe you should try doing this other thing. So like the reward structures completely flip when you reach the end of the road. And so I don't know how useful that is to somebody hearing that, you know, still wants that first win, but that's been my experience. It's a very, very different emotional reality once you're trying to do something new.

‍

Anthony Deighton (08:40): I would say that the motivation there becomes internalized as opposed to externalized. And to your point about getting the merit badges or getting, these are external validations of a step you've taken as opposed to following an internal passion and interest. And this goes to a broader point, which is I've seen real success in careers and in more generally in life come from taking on hard problems and kind of working those problems to success. And as opposed to getting merit badges,

‍

Alex Wiltschko (09:11): I think that there's, I mean always grain of salt when somebody says there's X kind of people. But I've noticed that people's motivation, there's like three kinds of motivation phenotypes that I've seen. One, and this is the rarest is people that can create structures that motivate other people, right? People that can spin a web that other people can live their lives upon. And then there's people that are internally motivated, these are, you know, folks that go home from work and then start on their hobbies, right? So you're just telling me about your raspberry pie in your basement. Like you have obvious internal motivation in that regard and I'm sure you create motivation for other people at work as well. But like that internal engine is kind of type number two. And then there's folks that really just want to go to work and they want to know that they're contributing.(09:53):And so they need to be in a structure where they know what the game is and people can transition between these different types and people also can transition over time and people can be all three types in different aspects of their lives, right? I mean there's some areas of my life where I don't want to create new structure. I just wanna be told like for exercise, like I'm not internally motivated to exercise, I need somebody to tell me what to do. I just completely abdicate all structure or motivation. I totally externalize it. But in my professional life it's my job to create, you know, narrative structure to create incentive and to create a mission that other people can sign up for. So I just see those three types throughout life and then people swap between them, you know, even within a day.

‍

Anthony Deighton (10:33): I love that framework. And in that same spirit, let's shift a little bit to Osmo since my understanding of what you're doing at Osmos is yes it's about digitizing scent, but fundamentally it's about categorization, cataloging and mapping. And that was sort of an aha insight for me as we were talking a little bit about Osmo before, maybe share a little bit about the relationship between a map and small, which may not be obvious to everyone.

‍

Alex Wiltschko (11:00): So this was one of the aha moments for me as well was that's what we need. We need a map for everything that we want to engage in where we want automation or we want engineering, we want systemization, we need a map, we need some simple thing that we can put on a piece of paper or a computer program that compresses the complexity of the real world that has correspondence with it, right? So for color we have RG B, it's one of the most beautiful maps in the world, right? It is three numbers that describe any color that we can see with our eyes. It's incredibly powerful. And without it, you can't make a camera, you can't have this conversation we're having over a video chat, you know, I can't see you, we can't understand how the brain structures color vision, we need the map.(11:42):It's the same for sound as well, just a different map. It's the Fourier basis or low to high frequency smell has never had a map though. And smell needs a map. The problem is that while color vision is three-dimensional and we know that in our eyes there's three channels of color information that roughly correspond to red, green, and blue. Our nose has a hundred times more channels of information, at least a hundred times more. How are you gonna fit a 300 dimensional map on a piece of paper? You need a different mapping technology. And that's why we've needed to wait for machine learning and statistics, artificial intelligence, the whole, you know, toolkit of modern data analysis to build a computationally supported map of odor. And that's what we did at Google Brain. That's the core engine of what makes Osmo hum is we've mapped odor and it turns out it's about a two or 300 dimensional map depending on which version of the model you're working with.(12:30):The current one is 256 dimensional, lovely power of two for the engineers out there. <laugh>, I won't speculate on the true dimensionality of odor, but you know, what we've seen is about 200 to 300 seems to work really well. It's also about the number of active olfactory receptors in your nose. So you know, perhaps there's a conclusion to draw. But this map has been incredibly powerful because every point on the map is actually a molecule. It's a a molecular structure that has a smell and only about 5,000 or 6,000 of those points on the map have been smelled and described in the literature. But there's billions and billions and billions of points on this map. And so our first beachhead market is exploring this map and finding new olfactory molecules and bringing them to market. It's a very interesting business to be in the fragrance industry and frankly we need better, cleaner, safer, new fragrance ingredients. And so we think that this is the first stop on our very long journey to give computers a sense of smell.

‍

Anthony Deighton (13:23): So before we jump into the commercial opportunity, just wanna go back to this mapping idea. It feels like one of the big trends that we've all been a part of over the last, you know, 10, 15 years is maps making our lives easier. I mean, not the least of which of course is Google maps and digitized two-dimensional and three-dimensional maps. So is it fair to say that what you are doing is sort of Google maps for smell, where instead of a two or three dimensional, depending on how you wanna think about Google Maps, whether it's two or three dimensions, you have a 250 dimension map?

‍

Alex Wiltschko (13:56): That's exactly it. And we're not done making the map. So the first version that we have is simple. It's a map of single molecules in how they smell. There's more maps to build, there's more complexity to add here. There's uncharted territory in the map that we have currently. If this map is built, it's going to give us a framework for thinking about all of the sense in the world. And that means a very large portion of all of the feelings and memories that we have because scent is so emotional. So not only is this a map of what we perceive in the world, it's potentially a map of our own lived experience and our own very deep-seated and wordless emotions, very powerful emotions. And beyond that, the other things that are on the map include signals that we might not be able to smell the equivalent of infrared or ultraviolet light, but for smell.(14:41):And we know lurking in that domain of the chemical universe that's barely smell. Some of it is our information about disease and wellness state. We know that people, you and me can smoke Parkinson's disease. That's been an amazing arc of research. Dogs and bees can smoke covid 19 and who knows what else. So if we complete the map, it's not only going to include our memories and our emotions and the things of beauty that we can experience out in the world and the opportunity and danger the chemical world tells us, it's gonna tell us about our own health, our own wellness, our own longevity. And so if we build that map, but then if we can learn to act on it, people will live happier and healthier lives.

‍

Anthony Deighton (15:18): Yeah. So in the same way that a map in the physical world has points of interest, these are the points of interest in the small world. Is that a fair analogy?

‍

Alex Wiltschko (15:27): That's really aptly put, you're actually giving me these lines that I don't ever use in the future because you've been thinking about maps so much. So I actually wanna learn about you. What's your reaction to this notion of a map of smell? Like how do you think about this?

‍

Anthony Deighton (15:40): So I think I have this vivid memory as an early driver. I have a child who's just gotten his driver's license and I have this vivid memory of being an early driver and being stuck in the west suburbs of Chicago in my car with no idea where I was. I mean, I knew vaguely I was in Chicago and my only solution to that problem was to literally stop the car, pull over, go into the trunk, find the map that was in the trunk paper map, locate myself in space, and then, you know, try to work out a way across multiple pages of coming back to something that I could then sort of recognize and navigate eventually back home and how difficult that process was. And it feels like the scent industry is working in a similarly arcane sort of set of tooling. And now we think, you know, my son who's just starting to drive the idea that he, you know, wouldn't be able to key in a way point in his phone or and navigate to that space.(16:33):Like that's incomprehensible to him. And it opens up all kinds of possibilities for the way he navigates in space. Many of which I think we wouldn't have thought about when maps were fir like physical maps were first being digitized. And I think in that same sense, that analogy is pretty apt here. Like think about point of interest databases or you know, even Yelp reviews. Think about all of those sorts of layers that sit on top of that. That's the sort of interesting space. And fundamentally that is a data question. So, you know, I have this thing, I often will say that at its core every business is a data business. You could imagine Osmo is really a technology company focused on hardware and software and the stuff that sits, you know, at that layer in the stack. But you know, your story of maps suggests that actually Osmo is a data business. Is that fair?

‍

Alex Wiltschko (17:22): I think that's fair. The caveat that I'd make is, you know, a map without a means to navigate it is an interesting thing you put on your wall. And so what I'm very intent for Osmo to be and to become even more is a company of adventures with a map, right? With a means of locating where we are on the map, a compass is sextant, you know, all the tools, but then the means of conveyance, right? A ship could go explore and find new molecules or a great perfume team that can create new formulations or interesting hardware and software that can store and understand and recreate sense in your life and store them so that you can return to those memories later. So a map and a ship is how I think about Osmo.

‍

Anthony Deighton (18:05): Speaking of that, you've built, uh, an interesting set of technology, hardware and software. You have the beginnings of a map, a map of significantly better fidelity than existed previously, but arguably one that will improve in fidelity with time. And you mentioned this before, so you're going after the fragrance industry, it's $30 billion year industry. Maybe talk a little bit about that decision and also what opportunity you see in the fragrance industry.

‍

Alex Wiltschko (18:32): I have a belief and it's supported by planning and evidence that what we're doing to digitize smell's gonna take a while and it's gonna be difficult. And as c e o one of my first jobs is to never run outta money. And you know, I'm very sensitive to what the cost of capital is and right now it's high, you know, two, three years ago is lower, but it becomes significantly easier if you have actually built a business that's generating cash flow. So this is something that maybe isn't so popular in Silicon Valley, but I think we can build a meaningful business by using an unfair advantage in an existing market. And that means building better fragrance molecules faster in the market of fragrance. And we like the market of fragrance because we know it, we love it and it's high margin, it's cash generative and frankly it needs an update, right?(19:18):We need to build better ingredients and we need to do it faster. The average time, the development of a fragrance molecule is like 5, 6, 7 years and there's two or 3000 molecules in the fragrance palette. About half of those are under threat in the sense that our standards for cleanliness and for safety are justifiably growing. And about half the molecules that are out there might not meet those standards. And there's six companies in the world that make one or two molecules per year. And if you do the math, you know, there's 10 years for half of the perfumers palette to somehow come under threat. We're not keeping pace, not by an order of magnitude. And so we see this as a, you know, present and clear opportunity for us to build a business that has value to people and also increases human health and happiness by making better ingredients that are in all aspects of our daily life. If you've washed your hair or washed your hands or put on a perfume, you've used something from this very large but secretive industry. And if we can build that as a commercial foundation, then we can build, you know, incredible organizations and technologies on top of that basis. That's, that's how I think about it.

‍

Anthony Deighton (20:20): And that's a always an interesting challenge for an entrepreneur to think about. The big opportunity digitizing scent and to your point, can have impact in health and wellness and commercial uses, probably industrial uses. And then this super narrow idea of focusing in on the fragrance industry. And my guess is you're not really focused on the fragrance industry per se, but within that a narrow sub-segment of what you can go attack in the short term.

‍

Alex Wiltschko (20:50): Yeah, and that's the 7 billion ingredients sub-segment and there's a sub sub-segment that we think that we can directly address. But yeah, we're, we're narrowing things down into where we actually have an advantage where we can help folks in the value chain by building better ingredients faster. But uh, yeah, we're tightly focused on that. The sales cycles aren't short like software as a service, but they're durable, right? So if you come up with something that's valuable, it's protected under patent and you can produce that ingredient for, you know, multiple decades to come.

‍

Anthony Deighton (21:20): So this creates a, a nice stable foundation on which to go after some of these other ideas, potentially bigger ideas or certainly different ideas.

‍

Alex Wiltschko (21:30): That's the idea. In all the while we're building out the map, we're improving that. We're also trying new experiments in digital olfaction. So we've got our main line of what we do, which is the next generation of safe and clean fragrance ingredients built using our AI platform. But while building out that AI platform, there's all these other very interesting experiments that we're doing and we'll have more to share as time rolls on.

‍

Anthony Deighton (21:51): So I thought I would end with that actually and give you an opportunity to tease a little bit about, you know, the rest of 2023 and 24 and what you are working on with Osmo. Any sort of big projects, sort of cool thoughts, something to leave the audience with to wet their appetite about where Osmos going.

‍

Alex Wiltschko (22:10): I think what folks will see is we're establishing ourselves and there's some interesting partnerships that we're going to be able to announce soon that show that our ambitions aren't just in scent for humans, but also scent for insects, which has a massive public health impact. So there's an exciting partnership that I'll be able to announce shortly, but also when we launch these ingredients, you know, I think folks are gonna have these new scents in their lives and we're thinking very hard about how we can develop ingredients, do that safely, do that quickly, but then actually get those beautiful experiences to people as quickly as possible. So I don't have any details to announce yet, but I think stay tuned. There'll be some beautiful, there'll be some impactful and there'll be some inspiring things that we're, we're gonna share this year and in the years to come. Well,Anthony Deighton (22:56):That's exciting and Alex, appreciate you taking the time and sharing some great thoughts with us around both career and how to think about moving between big and small companies doing startups, challenge of data and mapping and how to focus from a startup perspective on attacking a foundation technology and industry, what I hope will be a great business, uh, in the future. So thank you Alex.

‍

Alex Wiltschko (23:20): Thank you so much for having me. It was a pleasure to have the conversation. All

‍

Anthony Deighton (23:23): Right

‍

Announcer (23:24): Cool. DataMaster is brought to you by Tama, the leader in data products. Visit tama.com to learn how TAMA helps data teams quickly improve the quality and accuracy of their customer and company data. Be sure to click subscribe so you don't miss any future episode. On behalf of the team here at Tayer, thanks for listening.