WEBVTT - Teaching Computers to Smell 0:00:15.356 --> 0:00:22.436 Pushkin, tell me what we don't know about how smell works? 0:00:23.196 --> 0:00:25.916 Oh jeez, be sure to tell you what we do. 0:00:27.556 --> 0:00:30.516 This is Alex Wilscoe. He's the co founder and CEO 0:00:30.596 --> 0:00:34.636 of a company called Osmo, and despite his protests there 0:00:34.756 --> 0:00:37.436 he did tell me some of the things nobody knows 0:00:37.516 --> 0:00:39.756 about how smell works. 0:00:40.436 --> 0:00:43.716 Why do things smell the way that they do. Why 0:00:43.716 --> 0:00:46.956 can we smell certain things and not other things. What 0:00:47.156 --> 0:00:51.396 is the logic of how molecules are combined to create 0:00:51.436 --> 0:00:56.436 beautiful smells? Why do some smells create incredibly powerful emotional 0:00:56.476 --> 0:01:01.276 associations instantly and others seem neutral? Right? Why do something 0:01:01.356 --> 0:01:04.396 smell different to people? I think we have a hints 0:01:04.436 --> 0:01:09.116 in all these directions, but we have nothing like musical scale, 0:01:09.356 --> 0:01:11.956 where we have nothing like a periodic table. We don't 0:01:11.956 --> 0:01:14.556 know any structure to why things are the way that 0:01:14.596 --> 0:01:17.876 they are. It's a ton of mystery, and that's what 0:01:17.916 --> 0:01:20.316 makes it so exciting to work on this topic, is 0:01:20.396 --> 0:01:22.676 like there's so much we don't know, and to. 0:01:22.636 --> 0:01:27.956 Be clear, like with light, we just know whatever. If 0:01:27.996 --> 0:01:31.876 you tell me the frequency the wavelength, I can know 0:01:31.996 --> 0:01:34.156 exactly what color you're talking about, or the same thing 0:01:34.156 --> 0:01:36.876 with a wave form of sound, right, and so but 0:01:36.956 --> 0:01:39.436 if I give you some random molecule and say what 0:01:39.476 --> 0:01:40.276 does it smell like? 0:01:40.796 --> 0:01:44.116 Do you know? So that's what I've spent a lot 0:01:44.116 --> 0:01:46.716 of my professional life working on. It is exactly that question. Yeah, 0:01:46.756 --> 0:01:49.636 which is a draw structure of a molecule on a whiteboard, 0:01:49.796 --> 0:01:51.756 point out it and say, hey, what does this smell 0:01:51.796 --> 0:01:55.316 like wood or flowers or fruits or whatever? And so 0:01:56.396 --> 0:02:00.156 there is no way to know that for sure at all. 0:02:01.076 --> 0:02:04.996 But there's no good way even statistically to predict that 0:02:05.036 --> 0:02:08.476 without using large data sets, and at least in our hands. 0:02:09.356 --> 0:02:11.916 You need neural networks, you need deep sorting in order 0:02:11.956 --> 0:02:19.956 to do that. I'm Jacob Goldstein, and this is what's 0:02:19.956 --> 0:02:20.476 your problem? 0:02:20.636 --> 0:02:22.516 The show where I talk to people who are trying 0:02:22.516 --> 0:02:24.436 to make technological progress. 0:02:24.876 --> 0:02:25.596 Alex Wilch. 0:02:25.636 --> 0:02:29.516 Goo's problem is this, Can you use AI to teach 0:02:29.516 --> 0:02:33.036 computers to smell? And once you figured out how to 0:02:33.076 --> 0:02:35.716 do that, can you build a profitable business around it? 0:02:36.676 --> 0:02:39.756 Osmo's fun out of Google in twenty twenty three. The 0:02:39.796 --> 0:02:43.276 company recently launched a fragrance house to develop new perfumes. 0:02:43.676 --> 0:02:47.276 They've also done some work using scent to detect counterfeit shoes, 0:02:47.796 --> 0:02:50.276 and in the long run, they plan to use scent 0:02:50.436 --> 0:02:55.556 to diagnose disease. Before he started Osmo, Alex worked at 0:02:55.596 --> 0:02:58.676 Google as an AI researcher. Before that, he got a 0:02:58.716 --> 0:03:02.876 PhD at Harvard studying how mice respond to scent. But 0:03:03.116 --> 0:03:05.596 maybe the most important part of his bio came even 0:03:05.636 --> 0:03:09.436 earlier in his life, specifically when he was twelve years 0:03:09.436 --> 0:03:13.196 old and went off to summer camp in his home state, Texas. 0:03:13.516 --> 0:03:16.876 I was from a small town college station, and then 0:03:16.996 --> 0:03:18.916 most of the kids were from big towns like Houston 0:03:18.956 --> 0:03:23.436 and Dallas and Austin and San Antonio, and I hadn't 0:03:23.476 --> 0:03:28.076 really been exposed to, like I don't know fashion trends 0:03:28.276 --> 0:03:32.316 or you know, what was cool or popular. But everybody's 0:03:32.316 --> 0:03:34.476 all lumped together in summer camp. And then there is 0:03:34.516 --> 0:03:38.596 this thing called perfume that some of the richer, frankly 0:03:38.716 --> 0:03:42.196 richer and more popular kids had, And it was just 0:03:42.236 --> 0:03:48.036 amazing to me that these boys could spray themselves with 0:03:48.076 --> 0:03:51.916 this invisible mist, a clear mist, and then for the 0:03:51.956 --> 0:03:54.316 next four to six hours people around them would treat 0:03:54.356 --> 0:03:58.956 them differently, and that just blew my mind right, like 0:03:59.196 --> 0:04:01.356 there's no I can't I can see the clothes. Yeah, 0:04:01.436 --> 0:04:03.596 I can see how they act and walk and talk 0:04:03.676 --> 0:04:06.116 and how they you know, posture and all that, but 0:04:06.156 --> 0:04:09.796 I cannot see the fragrance. Yeah, but yet it is 0:04:10.076 --> 0:04:13.156 obviously doing something magical. It's like an. 0:04:13.036 --> 0:04:17.716 Axe body spray ad what what does that cause you 0:04:17.756 --> 0:04:18.036 to do? 0:04:18.236 --> 0:04:23.316 When I get home, beg my parents to buy and fail. 0:04:25.996 --> 0:04:28.316 We shopped at TJ Max and I started to really 0:04:28.316 --> 0:04:31.156 look out for fragrances there and then it just kind 0:04:31.156 --> 0:04:33.236 of snowballs from there, which I just realized there was 0:04:33.276 --> 0:04:34.916 like a whole lot of these things. I guess what, 0:04:34.956 --> 0:04:37.996 you can just try them, and some of them are 0:04:38.356 --> 0:04:41.796 actually way better and more opinionated and more beautiful. I 0:04:41.796 --> 0:04:44.396 don't I didn't have the vocabulary then, but like it 0:04:44.516 --> 0:04:47.516 just it was clear to me early on that like 0:04:47.876 --> 0:04:49.876 I never really thought about who made the clothes, but 0:04:49.916 --> 0:04:53.036 I started to think about who made these perfumes, because 0:04:53.076 --> 0:04:55.236 it was clear that there were choices that were being made. 0:04:55.676 --> 0:04:59.036 And like I just remember trying, and this is years later, 0:04:59.196 --> 0:05:02.996 trying Bulgari black, which really kind of clued me into 0:05:03.036 --> 0:05:06.356 this world. Bulgary black is not necessarily a great fragrance, 0:05:07.156 --> 0:05:09.916 but you can experience the top, middle and base notes 0:05:09.916 --> 0:05:12.196 in like forty minutes. Forty five minutes is pretty short, 0:05:12.556 --> 0:05:15.756 and so like a bigger fragrance, like a creed, events 0:05:15.876 --> 0:05:18.316 will last on your skin for a day, and so 0:05:18.716 --> 0:05:21.756 the whole fragrance unfolds. I mean, top notes will last 0:05:22.116 --> 0:05:25.436 max Fifteen thirty minutes, but the heart might last for 0:05:25.516 --> 0:05:28.796 several hours, and the base note might last for ten hours. Right, 0:05:28.836 --> 0:05:30.236 so it smells different. 0:05:30.276 --> 0:05:32.716 You can still smell it, but it smells different whatever 0:05:32.756 --> 0:05:33.476 an hour after. 0:05:33.356 --> 0:05:35.796 You put it on and four hours after that, because 0:05:35.796 --> 0:05:39.196 a great fragrance is actually many different fragrances within it. Yeah, Right, 0:05:39.196 --> 0:05:41.116 There's the first one which peels off or that quickly 0:05:41.116 --> 0:05:43.716 burns off quickly. There's the second fragrance, which is the 0:05:43.716 --> 0:05:46.396 heart note, which will last for you know, sometimes hours, 0:05:46.396 --> 0:05:48.796 but in this case, like another twenty minutes. And then 0:05:49.036 --> 0:05:51.196 the base note, which is a third fragrance, which is 0:05:51.196 --> 0:05:55.036 what's left after those two burn off. And it's like 0:05:55.076 --> 0:05:57.676 three acts of a movie, right, I think it's quite beautiful. 0:05:58.036 --> 0:05:58.756 So how do we. 0:05:58.676 --> 0:06:04.996 Get from you being a teenager preoccupied with fragrance to 0:06:05.356 --> 0:06:09.796 you using AI to predict how malls will smell? 0:06:10.036 --> 0:06:13.076 Yeah, Like the computer part was always different from the 0:06:13.076 --> 0:06:16.196 fragrance part I just I love computers. We always had 0:06:16.196 --> 0:06:19.516 computers at home. I started programming around I don't know, 0:06:19.636 --> 0:06:22.076 eight years old. It was my life, Like, my entire 0:06:22.076 --> 0:06:24.116 life was computers for a long time, still lose in 0:06:24.196 --> 0:06:27.676 a way, and fragrance was not a part of it. 0:06:28.476 --> 0:06:31.476 I got into you know, statistics, which became machine learning 0:06:31.476 --> 0:06:33.756 around the same time. Again for totally independent reason. There's 0:06:33.796 --> 0:06:35.916 this thing called the Netflix Prize. It was like one 0:06:35.956 --> 0:06:39.996 of the first competitions to build great mL algorithms. I competed. 0:06:40.116 --> 0:06:42.716 Now, that's basically to tell me what else I'll like 0:06:42.756 --> 0:06:44.876 on Netflix, right, That's what that contaest was. Like, if 0:06:44.916 --> 0:06:49.956 I've watched whatever, if I watched Succession and the Sopranos, 0:06:49.996 --> 0:06:51.196 what should I watch next? 0:06:51.356 --> 0:06:54.116 Then you're gonna like another kind of dark, but you know, funny, 0:06:54.316 --> 0:06:57.316 kind of soap opera type of a thing, exactly. And 0:06:57.356 --> 0:06:59.636 so Netflix did a really bold thing, which is they 0:06:59.676 --> 0:07:01.836 released a data set and said, here's what good looks like, 0:07:01.836 --> 0:07:04.236 here's how we measure it. Have at it. And they 0:07:04.276 --> 0:07:06.316 paid a million dollars to the Winter, which was a 0:07:06.316 --> 0:07:08.916 combination of a few teams. But what they really did 0:07:08.996 --> 0:07:13.556 is they brought a particular kind of machine learning into 0:07:13.596 --> 0:07:17.556 the forefront called collaborative filtering. Really showed that this stuff worked, 0:07:17.596 --> 0:07:20.676 and by the way, other companies were already racing to 0:07:20.796 --> 0:07:23.996 use this, So like this recommended systems was a big thing, 0:07:24.476 --> 0:07:26.436 but Netflix was putting it out into the public and 0:07:26.476 --> 0:07:28.316 allowed a kid like me I think I was eighteen 0:07:28.356 --> 0:07:30.556 or nineteen years old to actually compete and pretty well 0:07:30.596 --> 0:07:33.356 in that. And so I just got exposed to this 0:07:33.436 --> 0:07:36.396 world through that and it was super fun. I mean 0:07:36.396 --> 0:07:38.956 they gamified it and had I had a blast. So 0:07:38.996 --> 0:07:40.756 that was my first exposure to machine learning. 0:07:40.996 --> 0:07:42.956 Turned out to be a good time to start working 0:07:42.996 --> 0:07:44.316 on machine learning, Yeah. 0:07:44.236 --> 0:07:46.156 Totally, because if I had started now, they wouldn't let 0:07:46.156 --> 0:07:50.316 me in because that's probably ten years ago. Yeah, ten 0:07:50.356 --> 0:07:55.476 years ago. Yeah. And then, you know, I was doing 0:07:55.516 --> 0:07:58.916 my undergraduate training in neuroscience, and I was studying more 0:07:58.996 --> 0:08:02.796 behavior than old action because it actually turned out that 0:08:02.796 --> 0:08:05.556 olf action was a hyper specialized sub field of neuroscience. 0:08:05.596 --> 0:08:08.196 I didn't realize how niche it was. I loved smell 0:08:08.516 --> 0:08:10.796 and I was doing neuroscience, and I knew I wanted 0:08:10.836 --> 0:08:14.476 to do smell neuroscience the fancy word for nactual factory neuroscience. 0:08:14.996 --> 0:08:17.236 And so there's really two universities in the world that 0:08:17.316 --> 0:08:20.876 like have a critical mass of these researchers. It's Columbia 0:08:20.916 --> 0:08:23.276 and it's Harvard, and I applied to both. I went 0:08:23.316 --> 0:08:27.516 to Harvard, and I realized nobody cares about this problem. 0:08:28.276 --> 0:08:32.116 Nobody cares about why molecules smell the way that they do. 0:08:32.476 --> 0:08:34.596 There's a much longer conversation as so why that's the 0:08:34.636 --> 0:08:37.316 case and why that's still persistent. Now that's changing. Well, 0:08:37.396 --> 0:08:38.836 let me ask you this. Let me ask you this. 0:08:38.916 --> 0:08:41.356 At that time, I mean I get it as a 0:08:41.396 --> 0:08:42.636 basic research question. 0:08:42.996 --> 0:08:43.636 I mean, I'll tell you. 0:08:43.996 --> 0:08:46.396 We was talking with the producer and editor of this show, 0:08:46.396 --> 0:08:48.276 and we were getting ready for this interview, and we 0:08:48.276 --> 0:08:50.316 had this interesting conversation talking about scent and what you're 0:08:50.316 --> 0:08:53.876 working on whatever. And then I went down and I 0:08:53.916 --> 0:08:55.836 saw my daughter, and so, what are you organized? Said 0:08:55.836 --> 0:08:57.516 this guy who's trying to figure out scent and teach 0:08:57.556 --> 0:08:58.316 computers to smell? 0:08:58.396 --> 0:09:01.396 And she said why, I said, I don't know. 0:09:01.596 --> 0:09:06.436 I ask, so why was it compelling to I get 0:09:06.436 --> 0:09:08.556 it as a basic research question, but at that time 0:09:08.716 --> 0:09:12.076 was like, there were there applications that came to your mind. 0:09:12.636 --> 0:09:20.676 Look, the the steps of development of this thing that's 0:09:20.716 --> 0:09:23.636 now OSMO went through those different iterations. 0:09:23.676 --> 0:09:26.276 You know, I started as an academic scientist and I 0:09:26.356 --> 0:09:30.156 was trained in that world, and then I left and 0:09:30.196 --> 0:09:33.356 it did some entrepreneurship, but it ended up in industrial research, 0:09:33.716 --> 0:09:36.116 and they're like being curious, frankly was enough, and the 0:09:36.356 --> 0:09:38.956 idea this is this is Google, this is now a 0:09:38.956 --> 0:09:41.276 Google Brain. Yeah, and there's a few steps in between. 0:09:41.316 --> 0:09:44.316 But basically as you're an AI researcher at Google at. 0:09:44.156 --> 0:09:46.836 This moment when you're doing industrial research, right, yeah, exactly, 0:09:46.916 --> 0:09:49.036 and Google Brain at the time now it's Google Deep 0:09:49.076 --> 0:09:52.396 Mind very much had like a thousand flowers bloom mentality, 0:09:52.436 --> 0:09:54.596 and so people were working on crazy stuff, including me 0:09:54.636 --> 0:09:56.076 working on something Bell Labs. 0:09:56.116 --> 0:09:59.196 It's basically like Bell Labs of the twenty first century, right, you. 0:09:59.116 --> 0:10:01.556 Have it, exactly, Bell Labs, Erox Park, that kind of 0:10:01.556 --> 0:10:06.396 avid and it truly was dreamy, sounds dreamy. It was awesome, right, 0:10:06.436 --> 0:10:08.316 And it was also a moment in time, and now 0:10:08.316 --> 0:10:10.356 I think that moment's going for better or for worse. 0:10:12.796 --> 0:10:17.196 The idea was pretty straightforward for Google, which was, are 0:10:17.316 --> 0:10:19.156 the products at Google know what the world looks like 0:10:19.236 --> 0:10:21.716 and know what the world sounds like, and that's useful. Right, 0:10:21.716 --> 0:10:25.076 that's information that Google's organizing. If we knew what things 0:10:25.116 --> 0:10:28.316 smelled and tasted like, that would be useful, right. Uh huh. 0:10:28.356 --> 0:10:31.676 The original mission of Google is organized the world's information, right. 0:10:32.036 --> 0:10:35.316 Exactly, and make it universally accessible and useful. And there 0:10:35.396 --> 0:10:38.956 was a whole slice of reality, huh, the chemical slice 0:10:38.956 --> 0:10:41.796 of reality that was invisible, right, not just to Google 0:10:41.836 --> 0:10:45.316 but two computers. Yeah. Yeah, and that felt really important, 0:10:45.436 --> 0:10:47.476 and we had agreement and buying all the way up 0:10:47.516 --> 0:10:49.756 to the executive level. They're like, yeah, let's go, let's 0:10:49.796 --> 0:10:50.276 go look at that. 0:10:50.756 --> 0:10:53.276 So you're doing a like basic AI research at Google 0:10:53.316 --> 0:10:57.076 and you decide to see if you can basically use 0:10:57.116 --> 0:11:01.196 AI to figure out scent, to say, here's a molecule, 0:11:01.396 --> 0:11:02.236 what does it smell like? 0:11:02.316 --> 0:11:04.956 Right, that's the basic endeavor. How do you do that? 0:11:05.156 --> 0:11:08.556 What is it that you actually do? Yeah, so first 0:11:08.596 --> 0:11:10.116 it starts to innovation, which I was like, let's figure 0:11:10.116 --> 0:11:13.756 out smell. But it actually was a lot more natural 0:11:14.036 --> 0:11:18.436 than I think it sounds, which is scent is just chemistry. 0:11:18.596 --> 0:11:22.476 It's molecules, and we got to do AI for molecules. Right. 0:11:22.516 --> 0:11:25.116 If we're going to do AI for scent, and the 0:11:25.156 --> 0:11:28.196 thing that had happened in between. You know, there's a 0:11:28.236 --> 0:11:33.916 five year period between my academic life and my industrial life, 0:11:34.076 --> 0:11:35.916 and what had happened in those five years is actually 0:11:35.916 --> 0:11:37.956 some of the people I did my PhD with and 0:11:37.956 --> 0:11:39.396 then some of the people I ended up working with 0:11:39.436 --> 0:11:45.276 at Google Brain really cracked machine learning or AI for molecules. 0:11:45.436 --> 0:11:46.956 But they didn't do it for scent. They did it 0:11:46.996 --> 0:11:49.156 for a few other things. They did for drug discovery, 0:11:49.156 --> 0:11:52.836 and they did it for like materials discovery, so like 0:11:53.196 --> 0:11:54.476 new materials for LEDs. 0:11:54.796 --> 0:11:59.436 Right, So you happen to be doing essentially basic research 0:11:59.596 --> 0:12:02.436 at Google at this moment when there is this new 0:12:03.836 --> 0:12:06.876 way to use AI that is well suited to molecules, 0:12:06.916 --> 0:12:11.156 and you say, we can do let's do it, Yeah, 0:12:11.276 --> 0:12:11.636 let's do it. 0:12:11.756 --> 0:12:13.556 Yeah, we can do it. The other pieces are great. 0:12:13.596 --> 0:12:16.356 You got the algorithm. Where's the data? Classic? That's the 0:12:16.396 --> 0:12:19.796 classic AI question, right, like, exactly where's the data? What 0:12:19.916 --> 0:12:23.076 I did know just from being obsessed and in this 0:12:23.156 --> 0:12:25.836 world for a long time prior to that, there were 0:12:25.876 --> 0:12:29.636 these collections of data sets that were honestly really more 0:12:29.676 --> 0:12:33.756 like magazine catalogs for fragrance ingredients, and so there were 0:12:33.796 --> 0:12:37.556 these catalogs basically saying this is the ingredient, this is 0:12:37.596 --> 0:12:39.716 the molecular structure of this ingredient, and here's what it 0:12:39.756 --> 0:12:42.276 smells like. And by the way, the rating of what 0:12:42.316 --> 0:12:45.356 it smells like was done by a professional, by a perfumer. 0:12:45.996 --> 0:12:48.236 And so the special sauce that we added is we 0:12:48.396 --> 0:12:50.876 we went and we got that data and we fused 0:12:50.876 --> 0:12:53.796 a few data sets together, and we cleaned it very carefully, 0:12:54.196 --> 0:12:55.636 and that that hadn't been done. 0:12:55.716 --> 0:12:58.516 And it's something it's like five thousand dish right, it's 0:12:58.556 --> 0:13:00.676 five thousand or so different molecules. 0:13:00.716 --> 0:13:04.076 Okay, yep, exactly. And here is this the one with 0:13:04.156 --> 0:13:06.316 the list. I love the list. Here I have it. 0:13:06.716 --> 0:13:11.076 This the one sweet fruity, vanilla, powdery, fluoral, barry, fermented, 0:13:11.156 --> 0:13:12.756 nutty ozone, buttery musk. 0:13:12.836 --> 0:13:15.596 It's it's that lispright. Those are they? And there's one 0:13:15.676 --> 0:13:18.396 hundred and thirty eight of those descriptors I think that 0:13:18.436 --> 0:13:21.156 we used in that data set. Sometimes we use smaller subsets, 0:13:21.196 --> 0:13:24.276 but the full set originally is about one hundred and forty. 0:13:24.836 --> 0:13:28.356 So okay, so you have your whatever, your five thousand 0:13:28.396 --> 0:13:32.276 molecules labeled with one hundred and forty different sets. You 0:13:32.356 --> 0:13:35.836 train your AI model on this data set, and then 0:13:35.876 --> 0:13:38.036 you want to find out does the model work? 0:13:38.076 --> 0:13:39.196 Does the AI work right? 0:13:39.276 --> 0:13:42.756 If I give the model some new molecule molecule that 0:13:42.916 --> 0:13:45.916 wasn't in the training data, will it know what that 0:13:45.996 --> 0:13:49.716 molecule smells like? And to test that to answer that question, 0:13:49.796 --> 0:13:50.676 you actually. 0:13:50.316 --> 0:13:51.116 Do this study. 0:13:51.276 --> 0:13:53.236 So you get a bunch of people to smell these 0:13:53.636 --> 0:13:56.676 molecules that are not that your model was not trained on, 0:13:56.836 --> 0:13:59.636 essentially right, and say what it smells like? It's weird 0:14:00.076 --> 0:14:03.076 like what? You don't actually care what it fundamentally smells like. 0:14:03.116 --> 0:14:05.956 You just care what everybody on average thinks it smells. 0:14:05.676 --> 0:14:08.716 Like, because guess what, that's what it's what smell is? Yeah, 0:14:08.756 --> 0:14:10.876 that's what do you think of smell? Yeah? Ye? 0:14:11.236 --> 0:14:14.516 So you asked this panel to what do all these 0:14:16.036 --> 0:14:18.436 molecule smell like? And then you ask the model what 0:14:18.476 --> 0:14:20.356 do they smell like? And you compare the results and 0:14:20.396 --> 0:14:21.196 how does the model do? 0:14:23.156 --> 0:14:26.956 That was really the threshold of breakthrough in my mind 0:14:27.196 --> 0:14:29.676 was like are you worse than a person? Yeah? Or 0:14:29.676 --> 0:14:33.756 are you slightly better than a person? And we got 0:14:33.756 --> 0:14:36.276 slightly better than a person, which was a breakthrough in 0:14:36.316 --> 0:14:36.956 my view. 0:14:36.956 --> 0:14:39.796 Right, and so yes, so that paper you published in 0:14:39.836 --> 0:14:43.516 Science and you started Ozmo kind of around the same time, Right, 0:14:43.516 --> 0:14:46.436 you started that study at at Google, is that right? 0:14:46.436 --> 0:14:47.956 And then by the time it was published you had 0:14:47.956 --> 0:14:51.076 spun Osmo out of Google, right, that's right. So you 0:14:51.396 --> 0:14:54.356 have this map, you have this model that can basically 0:14:54.396 --> 0:14:58.236 given a molecule, predict pretty well what the average person 0:14:58.356 --> 0:15:03.516 thinks that molecule smell like. But there's still a second problem, right, 0:15:03.596 --> 0:15:07.196 which is, in the world, in the wild, you don't 0:15:07.276 --> 0:15:09.156 know what molecules are in the air. You don't you 0:15:09.236 --> 0:15:13.076 know what molecules somebody's smelling. And so for that second problem, 0:15:13.156 --> 0:15:15.636 you need to try and build some kind of automated 0:15:15.676 --> 0:15:18.716 system for figuring out what molecules are in the air 0:15:18.916 --> 0:15:20.156 at a given that's correct. 0:15:20.676 --> 0:15:25.956 Getting to one molecule structure is actually not trivial. So 0:15:26.116 --> 0:15:28.036 to go from a physical thing and know all the 0:15:28.116 --> 0:15:31.876 molecular structures like not a solveable. So there's a lot 0:15:31.876 --> 0:15:33.476 of ways to do that. There's a lot of chemical 0:15:33.476 --> 0:15:36.196 sensors out there, none of them will just tell you 0:15:37.276 --> 0:15:41.036 the formula, right, So that's hard, really hard. 0:15:41.156 --> 0:15:44.676 So there's like a chemistry problem of like isolating the 0:15:44.756 --> 0:15:49.316 molecule basically and deriving the chemical formula. 0:15:49.236 --> 0:15:51.716 Exactly taking a real smell and it's composed of a 0:15:51.716 --> 0:15:55.196 bunch of different molecules with different structures, and there's different amounts, 0:15:55.236 --> 0:15:57.956 there's ratios. You got to get that recipe out of 0:15:57.956 --> 0:16:02.036 the air. So that's on. That's hard. That was unsolved 0:16:02.036 --> 0:16:03.836 at the time to do it in an automated way. 0:16:04.396 --> 0:16:08.116 And by the way, if we're following this story chronologically, 0:16:08.116 --> 0:16:10.196 we hadn't done this yet, agah, but we knew we 0:16:10.236 --> 0:16:12.236 had to do that, right, So we knew that, Okay, 0:16:12.236 --> 0:16:15.436 if we wanted to actually digitize the world of scent 0:16:15.796 --> 0:16:18.516 and have a record of what the world smelled like 0:16:18.516 --> 0:16:20.396 and maybe even replay it, we're going to have to 0:16:20.436 --> 0:16:22.596 do this. We needed to automate that and have it 0:16:22.676 --> 0:16:24.196 be automatic, and that's what we did. 0:16:24.396 --> 0:16:28.116 So basically, you can put any smell into the machine 0:16:28.116 --> 0:16:31.236 and it'll tell you what it's made of. At this point, 0:16:31.276 --> 0:16:33.596 oh yeah, so you're setting out to start OSMO, Like, 0:16:33.716 --> 0:16:36.516 what are you thinking of in terms of the set 0:16:36.556 --> 0:16:39.916 of potential commercial applications. 0:16:39.516 --> 0:16:45.716 So we really had we had three in mind, and 0:16:45.756 --> 0:16:49.756 there's still very much present in mind the focuses has 0:16:49.836 --> 0:16:52.036 become a lot crisper though in terms of what we're 0:16:52.476 --> 0:16:56.436 concentrating on. We know the fragrance industry is huge and 0:16:57.076 --> 0:17:02.276 very profitable, and it's also something I personally love. That's 0:17:02.676 --> 0:17:05.916 a thing we want to automate and understand. And then 0:17:06.076 --> 0:17:09.476 we know that dogs can detect things, right, and so 0:17:09.516 --> 0:17:13.356 we know dogs can detect harmful substances like drugs or bombs, 0:17:13.676 --> 0:17:15.796 or things that just shouldn't be there, like produce where 0:17:15.796 --> 0:17:18.996 it shouldn't be being shipped. And then we also know that 0:17:19.036 --> 0:17:22.596 dogs and even in some cases people can detect health 0:17:22.796 --> 0:17:25.836 or disease states. Right. We know that missus Milner, a 0:17:25.956 --> 0:17:29.516 nurse in the UK, was able to smell Parkinson's disease 0:17:31.316 --> 0:17:34.156 and she's since been able to teach that skill to 0:17:34.196 --> 0:17:36.396 other people, which is really amazing. And then we figured 0:17:36.396 --> 0:17:39.796 out all the chemistry of what's actually being smelled. We 0:17:39.876 --> 0:17:43.356 know that there's many many instances where there is a 0:17:43.436 --> 0:17:45.836 scent signature to a disease or to a wellness or 0:17:45.876 --> 0:17:50.276 to a health state that hasn't yet been fully figured out, right, 0:17:50.316 --> 0:17:53.236 but we know that they exist. Those are the three, right, 0:17:53.316 --> 0:17:57.596 So fragrance industry really security and supply chain and health 0:17:57.596 --> 0:18:03.076 and wellness, and I view them in that order because 0:18:03.516 --> 0:18:07.476 that's like the order in which I think we can 0:18:07.516 --> 0:18:11.036 be useful to the world. Right, Designing fragrances is something 0:18:11.036 --> 0:18:14.156 that's much more attainable technically, and frankly, it's just a great, 0:18:14.556 --> 0:18:17.156 much faster sales cycle to be business to be in 0:18:17.836 --> 0:18:20.676 than ultimately diagnostics, which are so hard. Right, I mean, 0:18:20.836 --> 0:18:22.916 it is my north star. It's like where I want 0:18:22.956 --> 0:18:25.596 to take the company. But I also have no illusions 0:18:25.636 --> 0:18:27.556 about how hard that is, and I just I've seen 0:18:27.596 --> 0:18:29.556 all the failures of the companies that have attempted it, 0:18:30.076 --> 0:18:33.356 and I think I've learned from what hasn't worked, and 0:18:33.396 --> 0:18:35.916 so I'm incorporating those learnings into how I want to 0:18:35.916 --> 0:18:39.276 build the company, which is, build a great business in fragrance, 0:18:39.356 --> 0:18:42.716 build beautiful fragrances for the world, and then strike out 0:18:42.756 --> 0:18:46.876 from that position of strength and to even more ambitious frontiers. 0:18:49.716 --> 0:18:51.276 We'll be back in just a minute. 0:19:00.276 --> 0:19:02.396 When I first heard about the work Alex was doing 0:19:02.396 --> 0:19:07.236 at OSMO, I understood why it would be useful for sensing. Basically, 0:19:07.356 --> 0:19:10.516 you might be able to build automate sniffing machines that 0:19:10.596 --> 0:19:14.196 could say detect cancer in a person or detect a 0:19:14.236 --> 0:19:18.116 bomb in a suitcase. But I couldn't figure out truly 0:19:18.556 --> 0:19:22.676 what the business case was for perfume. And in fact 0:19:22.756 --> 0:19:27.196 Osmo has recently launched a perfume business. It's called Generation. 0:19:27.836 --> 0:19:32.356 So I asked Alex, why is using AI and fancy machines? 0:19:32.636 --> 0:19:35.396 Why is that better than just designing perfume in the 0:19:35.436 --> 0:19:36.196 traditional way. 0:19:36.836 --> 0:19:41.276 We can go from the first kind of client demand. So, hey, 0:19:41.276 --> 0:19:42.836 I want to create a fragrance, and here's who my 0:19:42.836 --> 0:19:45.156 brand is, here's what I want to do. So just 0:19:45.156 --> 0:19:49.116 that description to a starting place of a fragrance in 0:19:49.156 --> 0:19:49.876 a minute or two. 0:19:50.316 --> 0:19:53.796 What happens at a traditional perfumer when somebody comes in 0:19:53.796 --> 0:19:54.476 with that request. 0:19:54.596 --> 0:19:56.756 Well, so let's say you're an emerging Let's say you're 0:19:56.756 --> 0:19:59.196 an emerging brand, right, So you're starting out or you 0:19:59.236 --> 0:20:00.596 have your first product and you want to add a 0:20:00.596 --> 0:20:02.996 second one. But you're small, right, You're not making a 0:20:03.036 --> 0:20:05.916 billion dollars in revenue. You're making less than that. So 0:20:06.116 --> 0:20:09.716 if you want to make a new custom fragrance, good luck. Right, 0:20:09.756 --> 0:20:11.436 you're not going to be able to get the attention 0:20:12.116 --> 0:20:15.276 of the big fragrance houses because they want a service 0:20:15.436 --> 0:20:20.196 business that's like millions and millions of dollars and you're 0:20:20.196 --> 0:20:22.476 not big enough yet. So if you want a great 0:20:22.516 --> 0:20:25.636 custom fragrance that your consumers are going to love, and 0:20:25.676 --> 0:20:27.916 you want to do it quickly so you're responding to trends, 0:20:29.636 --> 0:20:32.876 you aren't going to be able to get it done. 0:20:33.236 --> 0:20:35.556 So you have to make compromises, right, So if you 0:20:35.596 --> 0:20:37.156 want to move fast, you're gonna have to use a 0:20:37.196 --> 0:20:40.476 regurgitated fragrance. It's also called a library fragrance, which means 0:20:40.596 --> 0:20:42.276 somebody else in the market has your spell. 0:20:42.356 --> 0:20:44.396 I'm imagining that people who sell it call it a 0:20:44.436 --> 0:20:47.436 library fragrance rather than a regurgitated library fragrance. 0:20:47.476 --> 0:20:49.876 They do, they don't say regurgitated, but that's what it 0:20:49.996 --> 0:20:51.916 effectively is, right fair. 0:20:51.956 --> 0:20:57.796 Regurgitated does have a particular old factory connotation, so it's 0:20:55.956 --> 0:20:56.636 a level. 0:20:57.796 --> 0:20:59.756 It's vic or all sticks in the mind. 0:20:59.596 --> 0:21:03.596 What like, And I'm not I just genuinely don't understand, Like, 0:21:03.596 --> 0:21:05.636