Today, Explained - No one nose

Episode Date: March 12, 2021

Believe it or not, scientists still don’t know how the sense of smell works. But that isn’t stopping one guy at MIT from trying to reverse-engineer it. Sounds like a job for Vox’s new podcast, Unexplainable. Transcript at vox.com/todayexplained. Learn more about your ad choices. Visit podcastchoices.com/adchoices

BetMGM, authorized gaming partner of the NBA, has your back all season long. From tip-off to the final buzzer, you're always taken care of with a sportsbook born in Vegas. That's a feeling you can only get with BetMGM. And no matter your team, your favorite player, or your style, there's something every NBA fan will love about BetMGM. Download the app today and discover why BetMGM is your basketball home for the season. Raise your game to the next level this year with BetMGM, a sportsbook worth a slam dunk and authorized gaming partner of the NBA.

BetMGM.com for terms and conditions. Must be 19 years of age or older to wager. Ontario only. Please play responsibly. If you have any questions or concerns about your gambling or someone close to you, please contact Connex Ontario at 1-866-531-2600 to speak to an advisor free of charge. BetMGM operates pursuant to an operating agreement with iGaming Ontario. Today, today explain.

What you're about to hear in your Today Explained. trying to say here is subscribe to the show if you haven't already, because they are going to be bringing the goods, people. Enjoy. This is Unexplainable. I'm Noam Hassenfeld, hanging out here with our senior science reporter, Brian Resnick. Hey. So this is a show about unanswered scientific questions. And Brian, I wanted to talk to you because when we were initially talking about topics we might want to cover on the show, you told me there was this mystery you wanted me to look into about the nose. Yeah. I haven't done any research on this myself, but I do remember seeing a YouTube video that it was going on about like how smell is just still a mystery. Yeah, we still don't know everything about how smell works. And it's kind of shocking because it's a pretty basic thing to not fully understand. Yeah, it's like one of the five senses you learn in elementary school.

Yeah, and as I looked into this over the last couple weeks, I realized this isn't just a fun mystery to solve. If we could solve this mystery, we could actually save lives. Because cancer has a smell. I've never smelled it. Well, we can't smell it, but dogs can. I'm sitting here with actually four biodetection dogs. I talked to a scientist who's a particularly big dog fan, Dr. Claire Guest.

If you hear them in the background, it's because they've been for a nice long walk and they're currently drying out and scratching their paws on the carpet. Oh, those sound cute. So in the early 2000s, Claire was training these dogs, her bio-detection dogs, so that they could smell disease. But then her work took a turn. She was training her dogs for a big upcoming study. And one of my dogs, a beautiful dog who lived with me called Daisy, who's a fox-red Labrador. She started to look upset.

That was probably the best description, a bit upset by me. She was just kind of sitting there. Just staring at me and then nudging at me and staring at me and nudging at me. Claire felt the spot where Daisy was nudging and she could feel a lump. And I actually went on to be diagnosed with a very, very deep-seated early stage breast cancer. She got it treated and luckily she's okay, but she might not have found it in time. Because the tumor was actually very deep-seated. If it wasn't for her dog, Daisy. Wow, so like Daisy, she was trained to do this?

Yeah, yeah, so it's sort of like a game. Uh-huh. So you teach them to recognize an odor. When they sniff that new disease, they say, whoa, well, that's interesting. That's what you capture. You give the dog a reward. The dog starts to realize you want me to find this characteristic odor, and then

you build from there. Once the dog learns the smell, Claire can use a bunch of different things, urine, sweat, even pieces of socks worn by people with a certain condition. She takes some positive samples and some negative samples, and she puts each one on a

stand. Okay. They're sort of slotted into a grill on the stand and the dog goes along the grill sniffing each in turn. If the dog smells a positive sample he'll stop and wait for a reward. If it's a negative sample he moves from the one sample to the next so the passing on means it's a negative sample. So is cancer the only thing they can smell? Can they smell other diseases? Oh, it's not just cancer. It would seem from the dog's behavior that every single disease and condition has a unique odor. And dogs can be trained to reliably find this odor and tell us about it.

What diseases are we talking about? They can basically detect everything they've tried so far. So they've done various forms of cancer. Bladder cancer, prostate cancer, and breast cancer. Parkinson's. Bacterias. Malaria. And what I think might be the most exciting part of this whole research is that now they're starting to test this ability on COVID.

Like really? Like dogs can smell people who are carrying the coronavirus? Yeah, and they can do a pretty good job, actually. Claire says the research is looking really promising. They should be publishing the results soon. And just think about the potential here. I mean, it's so much quicker and less invasive than your average swab-in-your-nose PCR test. This is with a single sniff. Now, this sniff takes the dog about half a second to do. You could just have a dog smelling tons of people and send the ones that the dog thinks have COVID for a PCR to confirm. So, like, what do you want here?

Like, an enormous dog army? Yeah, that would be one way. And some places are actually doing that. There have been COVID-sniffing dogs in airports in Helsinki. Here in Helsinki, they're ahead of the pack. In Dubai. Now the folks in the Emirates claim

they have trained dogs to detect the smell of coronavirus with a 92% success rate. They were even used to screen fans that went to a Miami Heat basketball game. COVID-sniffing dogs will now greet fans before they enter the arena. But the idea of seriously scaling all of this up, it would be really tough to manage. It would take lots of time and money and expertise. The training isn't easy.

And aside from being difficult, a dog army has never been the plan here. The idea has always been for the dog to translate what he knows with his nose to an electronic device. And if scientists can reverse engineer what's going on in a dog's nose, they could use a robot nose. You know, maybe not for this pandemic, but for the next one. That would be helpful. But this is where the whole mystery of smell starts to be a big problem. And that's sort of what I want to focus on for the rest of the episode.

Because we don't really understand how smell works. Huh. We've got some of the basic mechanics down, but the fundamental issue is that we don't really know why one thing smells one way and why other things smell a different way. Yeah, you would need to know that. Yeah, and how can you harness the power of a dog's nose

if you don't know how smell works? But it's not like, we can't be totally in the dark here, right? You did say we know some basics of how odor works. Yeah, we know the basics. So what is, like, basically, what is smell? So there's stuff all over the place. Imagine, like, a candy bar or a tree or a candle, you know, anything. And molecules, these, like, small bits of stuff, are kind of flaking off and bouncing around. And smell is actually kind of like touch or taste. It's physical.

You're actually touching these substances with the inside of your nose. You are smelling materials that reach your nose and that physically, directly, materially interact with the receptors in your nose. That's Anne-Sophie Barwich. She's a professor at Indiana University. So when you go to a public toilet and you think, oh, somebody went there before me, it smells shitty. Well, you've got fecal molecules

indeed actually reaching your nose. So I'm ruining a lot of experiences. So we've got molecules hitting receptor proteins in the back of your nose, and then your brain makes sense of it all as a smell. That's the simple part. The tricky part is, even though we know how a smell gets into the nose,

we don't really know why things smell the way they do. For a long time, the best thing we could come up with was the shape of the molecules. So like, certain molecules, if they have a certain shape, they'd smell one way. And if they were configured differently, if they had a different shape, they would smell differently? Yeah, exactly. This is called the lock and key model. Basically, there are a lot of different receptors in your nose, lots of molecules out in the world,

and it's sort of like a kid's game. You know, put the triangle in the triangle hole, put the square into the square hole. And just to be clear, so the receptors here are the locks, and these molecules are the keys that would unlock them. Right, they get unlocked, and those send smell signals to your brain. But the more we've studied receptors, especially in the last few decades, the more we've realized that they can't really work like locks with keys.

First of all, the math just doesn't work. We have in humans about 400 receptors. We've got potentially one trillion odor molecules that humans can respond to. So clearly it can't be just a molecule-receptor interaction. I can smell one trillion different things. Yeah, one trillion different things. And there are definitely not a trillion different types of locks in your nose. There's only 400. So the idea that you might have certain shapes of certain molecules responsible for certain qualities was naive. Yeah. And then,

you know, the more we're learning about these receptors, they start to look even less like locks because scientists have discovered that receptors actually can like change their shape. These are not rigid entities because they're constantly changing conformation. They're squiggly and wiggly and they move a lot. And the lock and key looks so intuitive. It looks so plausible, but this is just not how it works. So is this idea like completely bunk?

Like smell has nothing to do with the shape of the molecules? Well, so I just think that shape is like one factor of many. Okay. Because like molecules that have similar shapes can smell different. And molecules that have different shapes can smell exactly the same. What you're saying here is that if I drew a picture of a molecule and took it to a scientist,

they wouldn't necessarily be able to tell me like what that molecule smells like. Right. There's a few molecules that might work for, but there's no systematic way to predict a smell just based on the shape. Essentially, the lock and key model might sound nice, but smell is just a lot more complicated and actually dynamic. You have like a tango almost going on. So sometimes you might have a dance partner and you've got this kind of erotic tension. Other times the feeling isn't there. And others you think, oh please, never, never again. Attraction is complicated. So in a similar way, receptors have different sensitivities. And each receptor is attracted to different kinds of things. So you

might have one receptor that goes, you know, I totally dig sulfur. I go for that. Another might be into a molecule with a kind of ring shape. How wiggly the molecule is, how flexible. It can be looking for more than one thing. And the next one might actually have 10 different features. The reason it's a dance though,

is because these two partners, they do more than just fit together. I like the analogy of tango because you've got two people dancing and interacting and you've got sometimes, of course, a more dominant person that leads the other. Sometimes the receptor can take over. So the receptor might also bend the molecule if it's a bit more flexible. Other times, a molecule can activate a receptor, turning it on. But in a different molecular context

and a different mixture, it decreases receptor activity. Which is more of a turnoff. And all of this is happening 400 times over for each receptor in the nose and then getting sent to the brain. How does the brain know what feature this receptor responds to? We don't really know.

That's the interesting question. So the nose is like really complex. I'm sold on that. It sounds to me like we just have no idea what this dance looks like for every conceivable odor. And that's the trouble here. Well, that's part of it. What goes on in your nose is

definitely super complicated. You know, this dance is what allows 400 receptors to create a trillion different smells. But there's actually a second level here. Different people can smell the exact same thing and experience different smells. Like, think about cilantro. I'm sure you know how some people think it smells and tastes like soap. I kind of think it tastes like soap and smells like soap. Oh no, that's sad. But I also don't hate it. Okay, that's weird. But I guess my nose is just doing a different dance than your nose. Yeah, that's basically it. And it's because you're a mutant, actually.

Oh. If cilantro smells like soap to you, it's because you have a mutation that causes your receptors to respond differently. Oh. But on top of all of this, there's a whole bunch of things to consider that has more to do with the brain. Things like language and culture, what your mom ate when you were in the womb, context, experience. All of these things impact how you smell. Okay. But here's where we get to the truly bonkers part.

Scientists are looking at all of this, the whole tango, and they are undeterred. Because the idea of harnessing this superpower in the dog's nose to smell disease, it's so enticing that some scientists are just full steam ahead on Operation Build a Robot Nose., do those people know how smell works? Not really, no. How can you build something like a robot nose when you just don't understand how smell works? They're just kind of ignoring the problem. Just ignoring it?

Yeah, it turns out you might actually be able to build a robot nose without really understanding how it works. Oh. Yeah, I'll tell you after a little break. Thank you. back in your pocket. Ramp says they give finance teams unprecedented control and insight into company spend. With Ramp, you're able to issue cards to every employee with limits and restrictions and automate expense reporting

so you can stop wasting time at the end of every month. And now you can get $250 when you join Ramp. You can go to ramp.com slash explained, ramp.com slash explained, R-A-M-P.com slash explained. Cards issued by Sutton Bank, member FDIC Terms and conditions apply. Unexplainable. Okay, Brian, we're back. Yep.

So three things to remember from the first half of the show. One, dogs can smell all kinds of diseases. Two, we don't really know how they do it because smell is this super complicated nose tango that we don't really understand. Yeah, got that one. And three, despite this enormous question mark, some people are still trying to make a robot nose. They're actually excited by how little we know. What attracted me to the whole game of smell was these dogs being able to do something that no analytical tool in my lab could do. This is Andreas Mershin. He's an excitable research scientist at MIT. And I have actually stored a hundred million dollars worth of

equipment. And it kind of pisses me off that a lowly dog can do better than a hundred million dollars worth of equipment. Something's off with that picture. I should be able to do this. And all this big, fancy equipment he has couldn't do this whole thing because of this complicated tango that's going on. Yeah, how the nose tango exactly works was still this big question mark. But Andreas thought he could solve it. I just decided that, look, the dogs have proven that they can do this. We can do it. We just need to figure out. And fortunately, Andreas got a great chance to figure it out. DARPA, the research and development agency for the Defense Department, they wanted to figure out this dog superpower too. Maybe use it to sniff out bombs, other things. So they held a contest to

see if someone could build a robot nose. The goal was pretty simple. Beating the dog. That was the challenge. It was an MIT scientist up against a dog. Andreas liked his chances. I thought I knew what I was doing. I was very confident. He got to work with a collaborator at MIT, Xu Wang Zhang, who had figured out how to grow real receptors in cells.

The same receptors that you have in your nose, we stabilized them using various technologies. And they sort of spread them out on top of a circuit board. It looked about the size of a desktop, maybe laying sideways. Wait a minute, wait a minute. You're telling, like, this is a cyborg. Yeah.

Like something out of science fiction. Yeah, this could be like a cyborg's nose or something. Yeah. He thought that it could sort of work like a souped up carbon monoxide detector, just one that would look for certain bomb-related chemicals. If it detected those, all of its kind of chemical alarms would go off. But there were so many chemicals it had to detect, the alarms just would have gone off way too often.

Essentially, he realized this model just wouldn't be able to decode the nose tango. So he's like right back at the beginning, square one now. Like he thought he could solve this whole tango sniffing thing and just failed? Yeah, but it actually gave him an idea. He thought he had to build something that decodes a smell. You know, that figures out exactly what it is based on its tiniest parts. But maybe that's not how smell works. Andreas kept thinking about Claire and her dog Daisy.

He was thinking, wait. Daisy, the dog of Claire, was trained on bladder cancer, and yet detected breast cancer. Which is really weird because the telltale molecules of those two types of cancer are different. There's nothing in common to them. And yet the dogs can generalize. You train them on one cancer, they realize, oh, this also smells cancer-y. And we have no idea how the dogs can do this.

We just know that they can. So is it kind of like they're learning what cancer smells like? Yeah, that's what he thought. You're teaching the dog how to smell cancer. Maybe when you smell something... You don't see a list of molecules, same as when you listen to music or you look at a painting. You don't see a list of pixels and color values. You see to music or you look at a painting. You don't see a list of pixels and color values. You see the whole painting at the same time.

So this idea is really key. I actually want to use one of Andreas' examples and try to think about a smell like a piece of music. Brian, you recognize this song, right? Obviously. And you know it because you recognize the pattern. You don't necessarily know it because you've memorized every last individual sound. Yeah, but I also know what the sounds are. Well, you can still recognize it when the sounds change.

Like if I play it with a synthesizer. Or a trash can. Or even when it's in a different key and the actual notes are different, you can recognize it. You still recognize it? Yeah, it's still somewhere over the rainbow. Right. Even though the parts are different, the relationship between the parts is the same. The pattern is the same. So smells are like these big holistic patterns we learn to recognize.

Yeah. It's like you just need that kind of whole impression. So here's where we get to Robot Nose 2.0. He called it the Nano Nose. The Nano Nose. Yeah. And this time he tried something different. Okay.

He realized he didn't need a chemical detector. What he needed was a brain. He looked at the example of a dog and he said, look, we don't know how the dogs are doing this. All we know is that they can do it and that they can be trained. So if Andreas could build something that recognizes a song, not the individual notes themselves, but the bigger picture, the pattern, he wouldn't need to figure out exactly how all those parts come together to form a smell. As long as he could train it like a

dog, it could work. So, like, how do you train a cyborg? Do you give it a treat? Honestly, kind of. I mean, the idea is so simple. It's either super brilliant or we're all very crazy and stupid. Because here's what we thought, okay? We thought, okay, we don't understand what the dog thinks about. We don't know how the dog's nose ultimately works. But if we follow the training that the dog is having. Meaning the dog gets a reward when it smells the right sock. The computer, it's slightly different instead of a reward.

It's a different button you push and it just informs all the receptors at that point, whatever you're sniffing now is the thing we're looking for. Find other ways to find that same thing. Oh, this is kind of like machine learning, like building up artificial intelligence. Yeah. Instead of solving this enormous, mysterious question mark

of exactly how smell works, you know, how every little part comes together to form a smell, Andreas just ignored the question mark. He built the nano-nose to recognize patterns, showed it what it needed to recognize for the test, and then he let the nano-nose teach itself how to recognize the smells. And when you've reached the same statistical results, one involving dogs, one involving machines. At that point, you can basically say, Okay, the two things are doing the same thing, to the limit of my understanding.

So Andreas ignored the question mark and still built a working nose. What's kind of spooky here is that what's going on in the nano-nose, it might not be the exact thing that's going on in a dog's head. Andreas doesn't really know. He just knows that it works. And it does work? Like, he built this? Well, sort of.

Andreas showed that it could smell certain molecules that DARPA was interested in. And it's pretty sensitive. But that was in an extremely controlled setting. Which might be why DARPA ultimately moved on from the program. The real world is just way harder. You know, there's all these smells bouncing around, but Andreas thinks the nano-nose can still get there. He's still pushing forward with the research, and right now he's working on making the brain part way smarter. Oh, so like this AI just needs to get better? Yeah, they're still trying to reverse engineer exactly what's going on in a dog's head

when it's smelling, like what exactly makes something smell cancer-y. Andreas and Claire actually just put out some really promising research on that front last month, but they haven't put it all together with the nano-nose in a real-world environment yet. Yeah, so how do you get it ready for the real world? I mean, it's not easy. He's going to have to expose it to tons and tons more smells to keep training the brain, make the AI smarter, and that could take a while. But as far as the actual nano-nose itself, he's gotten it way smaller than a desktop now.

At this point, it can actually fit inside your phone. Is that the goal here? Am I going to ask, like, hey, Siri, do I need a shower? Maybe. I don't know about Siri, but I can tell you that for Andreas, that is absolutely the next step. Think about it, right? Any single one of us can have a mole that becomes malignant and it has this period of six months, it's changing color, it's changing shape, and it's changing smell. If you wait six months, sometimes it becomes a death sentence. And this isn't some sort of far off sci-fi technology this could be happening soon

I think we're maybe five years away maybe a little bit less to get it from where it is now to fully inside of a phone and I'm talking to deploy into 100 million phones medical device phones are coming there's still something like just it's not solved here like we might get noses in our phones, but still we don't know how the nose works. Yeah, it's kind of weird.

But here's what might actually be my favorite part of the whole thing. It's true that he skipped over understanding, but what I love about this is that this move, this skip, might be the exact thing that ultimately gets him to the understanding. Richard Feynman, a famous physicist, he famously on his blackboard, the day he died, what was left on it was, if you cannot create something, you don't understand it.

It was kind of deep and powerful, but I didn't realize that what it really meant was, it's folly to think that you must understand something before you build it. That only works if you already know it. Maybe what Feynman was actually saying was more like this. Build it to understand it. That you can understand something better through the act of building it. The Wright brothers, who invented flight, they didn't know how flight worked.

They built the airplane in order to understand how it will fly. So this is what happened to us with the nose. The funny thing is here that scientists don't perfectly understand how planes fly. Yeah, that's probably for another episode. Yeah, we'll get to it. As far as the nose, Andreas still doesn't have the complete picture here, but he's closer and he's learning more the more work that he does. So build it to understand it, but that understand it part, not quite there yet.

Yeah, I kind of love this as an answer to an unexplainable question because, you know, it shows you don't need the full answer to actually make progress. Yeah, we have this incredible sounding new technology. Yeah, and I love this sort of confidence and hopefulness and optimism of Andreas' example because he's honest about the fact

that he doesn't really know how the nose works. But that big empty box, that question mark, is starting to feel less like an obstacle and more like an opportunity. This episode was reported and produced by Noam Hassenfeld, who wrote the music for it, too. Editing from Jillian Weinberger, Brian Resnick, and me, Bird Pinkerton. Liliana Michelena did the backtracking, Hannes Braun did the mixing and sound design, thanks Hannes,

and Meredith Hodnot is our senior producer. Also, Liz Kelly Nelson is the editorial director of Vox Podcasts. Special extra thanks this episode to Sarah Harrison and to Ella Fetter. Also, the mystery of smell has some pretty wild turns that we couldn't fit into this one episode. Like, there's this theory that the way our noses really work is by using quantum mechanics, which gets a little complicated. But if you want to dive really deep, please check out our show notes for a link to a piece by Anne Sophie Barwich. You heard from her in the first half of the show, and she wrote all about how the scientific community got really enamored with this quantum nose theory.

And then the theory wasn't really all that it was cracked up to be. And while you are down there in the notes, we've got a link to the Unexplainable website where you can find articles about our episodes, you can find show transcripts, you can find more. And make sure to jot down our email address so you can send us your thoughts. We to jot down our email address so you can

send us your thoughts. We are at unexplainable at vox.com. Please send us thoughts. We would love to hear thoughts. One more thing. If you want to learn more about the mystery of smell, I cannot recommend enough Anne's book, Smellosophy. It covers all kinds of fascinating history that we weren't able to get into in this episode. Unexplainable is part of the Vox Media Podcast Network, and we will be back in your feed next Wednesday.