Teaching Computers to Smell?! The AI That's Digitizing Scent

0:00So imagine if you could smell a photo, not just see it, but actually smell it. Like someone sends you a picture of fresh coffee and your phone releases that exact aroma. Sounds like sci-fi, right? Well, a company called Osmo just figured out how to teach AI about smell.

0:25Welcome, humans, to the latest episode of the Neuron Podcast. I'm Corey Knowles, editor of The Neuron, and we're joined, as always, by the writer of The Neuron Daily AI newsletter, Grant Harvey.

0:36Alex Wiltschko:Today, we are talking about something truly wild, teaching computers to smell. Our guest is Alex Wilchko, founder and CEO of Osmo, the first company to digitize scent. They've created an AI that can predict what a molecule smells like, teleport scents across rooms, and design brand new fragrance molecules that have never even existed before. And this isn't just about making better perfumes. Alex thinks this could actually help us detect diseases earlier and even fight malaria. That's wild. Alex, welcome to the Neuron. It's a pleasure to have you. Corey, Grant, it's awesome to be here. I'm excited to chat.

1:11Awesome. Well, I guess first question that everyone's going to be wondering, what does it mean to digitize smell? So, I mean, let's talk in analogies. What does it mean to digitize sound, which is what we're all experiencing right now? There's three steps. Take the physical world, which is vibrating airwaves, and turn that into a digital signal. So that's what a microphone does. That had to get invented at one point, right? Initially, that was literally changing those pressure waves into grooves of a needle, carving into a piece of physical material. Step number two, encode and reason about and then decode that signal, right?

1:54So now we have MP3. It's a way of compressing and understanding audio. We've got audio editors. So first you got to read the physical world, then you have to map it. And then you have to write it back out again, right? So that's what a speaker does, which is it takes those originally just grooves, but now digital signals, and then it moves a surface to recreate those airwaves. The important thing about all that is those three steps are wildly different. There are actually three different, totally different steps. Those are three different technology stacks that each operate under different physical principles, but you have to have them all.

2:30And so we're looking at digitizing the sense of smell, and we have to do all those three steps. We have to read the chemical world and turn it into digital signals. We have to map it, and then we have to write it back out again. When we started doing this, I mean, I've been thinking about this problem for about 20 years, and Osmo's been going for three. And before that, I was working on this problem at Google Brain for about six years. What's a little crazy is the ability to read the chemical world already existed. And it's really good, actually. There's these things called spectrometers. It's what like CSI uses to figure out, you know, what are the different trace samples at a crime scene.

3:07And then the ability to write the chemical world back out again is also pretty good. There's these fluid handling robots that will move around ingredients and mix them kind of like an inkjet printer can mix inks. The part that was missing was the map. And that's been a super hard problem to crack. And again, let's think about vision. RGB, three numbers that are the map of color. And in fact, we roughly have three different kinds of color receptors in our eye. They roughly correspond to RGB. We have over 300 types of olfactory receptors in our nose. So the sense of smell is going to be, just based on that, 100 times higher dimensional, 100 times more complex.

3:52Whatever map exists for smell is not going to fit on a flat piece of paper. We're not going to look at an RGB diagram and be like, oh, that's why lavender smells close to this other floral but far away from rotten eggs. Not going to happen. The map exists in data, though. So that's the core insight that we've been operating around is like, well, now we have algorithms. and we have huge data sets. And we don't need to wait for our own visual cortex, our own visual systems to lead us to the map. We don't draw it on pieces of paper. We extract it from data. So that's kind of the core insight. And then we reuse a ton of amazing technology and we stitch it all together.

4:30So when did you realize that this was even possible? So I got really into this problem. Let me back up. I got really into smell when I was very young. Like me and my dad would like guess the spice when we went out to eat and try to figure out like, hey, what's the smell here? What's the aroma here? I got really into perfume and perfume collecting when I was about 12. I was already a big computer nerd and computer programmer. So adding like perfume collecting in like a very small Texas town does not like make you the most popular boy. That's fair. But you don't choose your obsessions. They choose you.

5:05and then i watched a ted talk by a guy named luca turin uh when i was in i think i was in college and the ted talk said two things one is you know we really don't actually know why things smell the way that they do and then number two is he had a solution for it i i turns out i think that it's mostly wrong the solution but oh my gosh is there value in asking the right question and this idea that like you can tell me rgb and i can i can figure out what color that is it seems trivial we can't do that for smell like we can't talk about why something smells the way that it does it's just random and then i ended up going completely down that rabbit hole and i ended up and you know i did my undergraduate neuroscience i got my phd in olfactory neuroscience and i learned that actually nobody knows the answer.

5:59Like you go to the very edge of human knowledge, at least in academic science, and there's no answer forthcoming for something that seems very basic. It drove me absolutely bananas. And that's basically been my life's work since then is trying to figure this out. Wow. Hey, everyone. Today, I'm excited to tell you about Flora. The AI canvas used by creative teams like Base Design and Pentagram, it makes it easier for designers to generate campaign visuals, treatments, and final deliverable assets in just minutes. Unlike MidJourney or Runway, Flora gives full workflow control so creative teams can ideate, iterate, and refine with precision, all in one canvas.

6:40If you're a designer or an agency looking to scale ideas with a high degree of creative control, check out florafauna.ai. That's F-L-O-R-A

6:49Alex Wiltschko:f-a-u-n-a dot a-i check them out today and tell them the neuron sent you so in 2024 you achieved scent teleportation that's right uh it was a fresh cut plum uh from what i understand and what happened that day what was that that day like how did the project go tell just just kind of walk us through it if you would one of the wildest most excellent days in my professional life was that first breakthrough. And it was really just a first step. Like it's not the top of Everest. It's like getting to base camp, but it's still so, so powerful and so vivid in my memory. So the team had been working for a while on the technology and the processes to do this.

7:34And one day they said, Hey, look, I think we're ready. And they sat me down in the morning in front of 30 amber vials and inside of these vials, blind to the entire team, been set up by somebody else, lying to the entire San Teleportation team, inside of it were fruits and vegetables and some flowers. And so they didn't know what I was going to choose. But these were real things from the grocery store. And I think what I picked was a banana and it was a tropical fruit. I think it was a papaya. And this is just by smell and then I verified it later. And then a plum. And we put each three of these through the teleportation process, which is you show that scent to a device that was specially tuned and set up to analyze the molecular content of that smell.

8:27The data was uploaded to the cloud and analyzed by the AI systems that we built. And really it's AI for scent. We call that olfactory intelligence or OI. And then it was reprinted on our smell printer, which we had custom programs to work with the fragrance ingredients that we see in the real world. and the first smell that came off was plum and it was absolutely engrossing and transporting that i was smelling this a blotter dipped into a vial of clear liquid and i was having the visceral sensory experience of biting into a fresh summer plum like the kind with the snap when you bite into it it's still firm it's a bit tannic but like my whole kind of consciousness was filled with the experience of having just bought into a snappy, just having bought into.

9:18Bought into. My sensory experience. Bought into. Yeah. My sensory experience was filled with the experience of biting into a fresh summer plum that still had a snap to it. And I was like, wow. Oh my gosh, it works. Like no human was involved in actually looking at the data. It was all instrumentation and robotics and software. And we actually transported a plum from one room to the other. It was incredible. That's so cool.

9:49Alex Wiltschko:That reminds me of reading stories about Alexander Graham Bell and people in the early days of figuring out the audio and how you send messages to each other. It reminds me of that sort of level. I have to ask as well, just this is a slight side tangent, because the way that you're describing this is really, really visceral, really draws me in. Have you ever seen the show Drops of God on Apple TV? So I read the manga because I'm a big nerd. Okay. Fellow manga nerd. Yeah. And I watched the first couple episodes, but that like the intensity with which they portray the sensory labeling is like so extra and I love it.

10:30Yeah.

10:30Alex Wiltschko:For people who don't know, it's a great show about essentially a famous wine. What do you call it? A sommelier is the person who can smell and describe and knows about the world of wine. Yeah, who gives away, like basically dies and wills his fortune to his protege and his daughter. And they have to compete in a test to see who is worthy of his$150 million wine collection. And they just capture smell like so excellent. They almost have like a spiritual experience when they smell it. And she, in the show, did what you were talking about with the spice testing, where, you know, as a young girl, she was supposed to taste stuff and try and smell it.

11:11Alex Wiltschko:Her way, I guess, it was tasting as opposed to smelling. Yeah, and 90 % of flavor is smell. So there's just a few things on your tongue, and they're important. But the experience of wine, much of that is going to be through your nose. So, like, if you have a cold or if you plug your nose, like, it's just going to be sour or bitter or tannic. You're not going to get— The reason things don't taste right when you're sick. Exactly, exactly. Because, you know, what happens is you drink something or you eat something, you chew it, it creates a chimney of basically steam and aerosol that actually goes back up your nose.

11:46It's called retronasal olfaction. So, you're smelling as you're eating. Wow.

11:51Alex Wiltschko:I didn't know that. So, like, you mentioned the read-map-write framework, right, where you're reading it, you're then mapping it. and then you're writing it back. Could you just explain, like, in terms of, like, your system specifically, like, what is actually happening? Because you said there are, like, three different, like, systems that work together, I guess. Yeah, so reading is about using existing chemical sensors to turn molecules into digital signals. And that's a really old tradition. I'm more than 120 years old. Basically taking a piece of chemistry and turning it into electrons or photons.

12:26And there's a bunch of different ways of doing it. Many of them fall under the label of spectrometry, which just means measuring, you know, measuring physical things. And the kind of spectrometry that astronomers do looking at the stars, you know, the fundamental principles are kind of the same of what we do when we really, you know, dial in on ascent. We use different physical principles, but the idea is we're trying to figure out what's the matter in the cone of perception that we're pointed at. And so we use spectrometers in order to turn chemistry into digital signals. Normally, before OSMO, very, very well-trained, specialized people would look at that data.

13:12And that's still mostly the way that that data is interpreted is by people. and what we've developed is a suite of AI tools to either augment in some cases or automate the analysis of the data that comes off of those devices and that's that creates a firehose of data for us so we know what different physical things smell like we know what different market products smell like and that's our core business is we actually design fragrances for brands that's the way in which we're building the business to continue funding the curiosity to move to the next step of Shazam for smell and then to move to the next step of sniffing cancer.

13:53Shazam for smell. I love that. We sell perfumes to get the Shazam for smell and then we build both of those businesses to ultimately detect disease which is that's the holy grail.

14:02Alex Wiltschko:That's our North Star. That's amazing. How close are we to then like sending scents or detecting scents on like our phones like we send photos. Yeah, I mean, I can see a path to detecting scent already today. We do that for some customers. So there's an article about this online, but we've sniffed fake Air Jordans at StockX. So fake shoes smell different than real shoes. And that has to do with how they're constructed. So we already have a team and we already have a technology that can smell as well as our nose can. That's not our main business today. So we want to continue to develop the technology to take it from the size that it is, which is, I mean, it's not small, but it's not big.

14:48You can hold it and move it around. We want to make that something that is moving towards being pocketable and more affordable. It's not quite yet affordable, but it already solves the real world problems today, but we're going to continue to develop it. Getting it to something the size of your cell phone may require scientific breakthroughs that we can't even describe today. So I want to be real, like that's why we're taking the strategy that we are. Like I've looked at all the Enos companies. I've looked at all the olfaction companies that I've been able to find at least. And every single one of them that kind of aimed for that Star Trek tricorder vision, but didn't chart a path through kind of a reasonable business, all of them that went right for the mountaintop, they're all dead.

15:37And it's not because they weren't working hard enough. It's because it's so freaking hard, right? Like to solve that problem may require hundreds of millions or likely billions of dollars to build a device that can do what we really want it to do, which ultimately is what a dog's nose can do, which is detect disease and detect, you know, produce at borders or drugs at borders. Like to keep people safe and to keep people healthy, we want a device that you can hold in your hand in your pocket that smells as well as a dog's nose does and like i actually don't think anybody knows how to build that in short order um the world can change in an instant so who knows but to really do that well what the tack that we've taken is like you have to look at that with complete humility and understand that like man before that thing gets small it's going to be big as with all technology and before things are even in your home they're going to live in a like a warehouse or a special room which is like think about computers like mainframe computers in the 1960s filled up the whole room and had specialized staff on it huge then they got smaller by the time bill gates came around like you still there are no personal computers right he had to break into a company or a university to use a computer um and then they became personal right so like we're in the realm now where the ability to create smell and to read smell exists but it has to be enclosed in a laboratory and in a factory, which is why we have both of those things.

17:07Because we can't fit a mass spectrometer into our phone yet? We cannot fit a mass spectrometer into our phone yet. But we're going to get there. People are ingenious. It might either come from one of the folks at our company, from an Osmonaut, or it might come from outside of our company in an academic institution or another company. Somebody's going to figure this out. And we'll be there to help make sure it happens faster. But like, man, we're going to get there. 100 % we will. So you launched Osmo into the fragrance industry. What's the response been like? It's been overwhelming. And so we have more work than we know what to do with at this point, meaning we're using OI to design beautiful fragrances and we're working our absolute best to keep up with the demand.

17:54In fact, we're opening up a new factory in order to keep up with the demand. We learned a few things. Um, we, we've worked with a number of different customer sets and what we've learned is there's a lot of people out there who love smell and a lot of people out there who want to create beautiful scents and add it to their, their company, their brand. And not just like perfume brands, like you'd be surprised. Like there's a lot of individuals out there that have a following online and want to launch their own fragrance. Right. And they don't want to promote somebody else's. They want to design their own.

18:25They want to do it themselves. Exactly. That makes perfect sense. And so, you know, it used to be really hard to even make your own visuals or edit your own video. Everybody can do it now. And so what we're finding is that lots of people want to make their own scent and we can actually help them do that. So what we designed was this software experience called Osmo Studio. And it basically lets you design the scent. So you talk to it. You type into it just like a prompt in ChatGPT. you can select certain things like hey i really want this fragrance note or you know i want this to be unisex or i want to wear this at night it turns out there's some you know subtlety to how you construct the fragrance to make all those things make sense and then you hit build my scent and you hit buy and we send you a sample and actually three samples in a box of completely oi generated scent that matches the emotional experience that you're trying to create and we've gotten amazing feedback from it.

19:23Like we've, we, we kept it in a very small closed beta and we're going to be opening it up more and more over time, but you know, people are smelling it and they're saying, oh my gosh, this is exactly what I wanted. Some people are saying, uh, more and more over time are saying, I actually want to turn this into a brand, like let's launch it. And that's kind of where our demand is coming from. Uh, so this is really, really new and really, really special. So the normal way you design a scent at the industry could take you a year and a half or two years to get your scent finalized and launched. And what we're finding, even with the state that we're in today, is that we can give people their first scent in one week.

20:00And we hope to get that to two days and not too long. And then people can launch their whole brand in three or four months. And we have a wait list now. And so that's what we're able to do kind of with the current customer set. But we hope to take that down much, much less. So it's kind of turning the industry on its head where like it's there's not there's not a scent designer or perfumer hidden away somewhere it's you like you're creating the scent which is which is super beautiful and fun so you don't have to buy some celebrity's set anymore yeah you could be the celebrity you know like you could design a scent and launch it online and like totally capture the zeitgeist and then it could be the cory and grant you know neuron scent that really really captures the moment that's super beautiful and addictive.

20:48I think that's what we need, Grant.

20:50Alex Wiltschko:Yeah, we need to launch a scent, Corey. It smells like an orange cat. An orange cat. That's a pretty good name for a fragrance to start. Canvas your audience. Would you buy Ode to Orange Cat? And we can make that a reality. Actually, we should totally do this, Corey. That's awesome. I agree. We will take you up on that, Alex. Just mail on science. You know, that's possible. I have a funny question. So you mentioned like you can do this via chat now. So, you know, chats are going multimodal. Could you upload like a song or an image and say like, give me the vibe of this? Like how long would you could do that?

21:27We actually launched, we launched a product at the Museum of Pop Culture in Seattle. And if you go there, you go to the gift store and actually soon we hope bigger retailers, you can buy the scent of that museum experience. The only thing they gave us to create that scent was a picture of a 30 foot tall sculpture of like a swirling tornado of electric guitars. Oh, wow. So the input was this image of like a ton of like, you know, instruments, rock guitars. And the thing that came out of the scent that came out of that process smells like the metallic strings of the guitar. it's sweet and it's fresh and it's like you know lively and they put in we i mean they ultimately launched the product but we put it in a bottle for them we put it in a package for them and it's in the gift store right that's so cool wow what's another place that has like a really

22:31Alex Wiltschko:distinct smell that people oh i've seen this so there's a candle online that it recreates the smell of going on the et ride at universal studios okay have you seen this before no this is new to me who sells it like there uh i don't know who sells it but i've seen this uh perhaps on etsy or somewhere and uh people really like this because allegedly this place had a very distinct smell i think because of the smoke uh machines that they would use and okay that's so funny yeah and so that's like you got to find these places where they have a really distinct smell that people like and try and recreate that because that's an awesome.

23:08Yeah. I mean, it smells so powerful. It's super powerful, right? So that's, I mean, that's what we want to honor and tap into is like people want to create, people want to explore. And I think if we create the ability for people to express themselves, you know, even if it's some crazy new modality like scent, I think the empowerment that comes from that is really

23:32Alex Wiltschko:beautiful. actually i'd love to get your take on this because i have heard this before but i don't know the science behind it so i'm interested um i have heard that your uh however the the sensors in your nose your natural sensors work uh it because they die very quickly they have to encode uh information really fast i don't know if that's actually how it works you can tell me but um that that means that like when you have uh you can have a very strong smell memory uh to remember things and it's really powerful like to the point where it can take you back to a moment in time if you smell like a certain type of smell and so that makes me wonder if it's possible to you know have a picture and then that can maybe even like like from your past take you back to that moment camping trip with your grandparents as a kid or i mean yeah could you explain just how that yeah so there's there's three pieces to that like one is what happens to the olfactory receptors like how do they work um and then why is smell so emotional and can you can you tap into that and basically transport somebody back to a moment so the first is olfactory receptors do we call a turnover meaning the olfactory sensory neurons that are like the the cells at the top inside at the top of your nose that actually do the sensing they they don't live forever um very because they're and direct it's your brain directly touching the world right so like your brain leaves your skull that is what smell is so when you smell a rose your brain is literally touching the rose so those cells get subject to a lot of environmental you know change and so they they do die and the brain replenishes them and so this is why if you lose your sense of smell you're waiting like two weeks for it to start to come back that's the transit time for neurons to come to you know and regenerate the ones that are lost.

25:24It's happening continuously though. Then number two is like, why is it so emotional? And why is it also so tied to memory? And that's because it's directly jacked into your centers of emotion and your centers of memory. So every other sense before they can create memories or can create emotions have to go through a bunch of steps. And one of the way stations in the brain is called the thalamus. Smell is so old evolutionarily, it skips that. So it goes directly to the signal once it's kind of collected outside of the nose, inside of the skull in a place called the olfactory bulb. It goes right to the amygdala, which is the seat of emotion.