Latest / Glovisor Podcast / Ep 15: He Built a Company That Uses Rats to Detect Bombs and Drugs
Transcript
- GLOVISOR: What if the best AI detector we've ever built is actually a living creature? My guest trains animals to detect bombs, drugs, allergens, and even disease. This is the future of biosecurity, and it's already here. This episode's guest is the founder and CEO of Canary Detections. They are building something that honestly sounds like science fiction, a hybrid system that combines biological sensing with artificial intelligence to detect threats that our best machines still miss. Think explosives, narcotics, allergens, and even medical conditions, identified with a speed and sensitivity that only nature has perfected over millions of years of evolution. This is one of the rare business ideas that make you think, why hasn't anyone else thought about this sooner? Glovisor welcomes Door Wolf. Dor Woolf: Hey, thank you, Joe. Thank you for having me here. It's my pleasure to be a guest here at your podcast. Thank you very much. GLOVISOR: Excellent. Looking forward to the conversation. Dor, let's start with the first question. What was the moment that you realized nature might be a better detection than any machine we could have ever built? Dor Woolf: So I don't think it was like in one day, know, like a light, strike you and you've realized that biology is better as you're working with machine. You finding that they have some problems ⁓ and you know, biology it's already developed for millions of years and it's quite amazing to see how biology adopted for specific and their specific needs. When you try to do it with artificial captability, ⁓ it's a bit hard to achieve the same result. So when I'm looking on system and I'm looking on biology, you know, this is why people are still using canine in order to do detection because they consider it to be the golden standard in detection. But also when you're working with biology, also have some you need to give some takeaways, I want to say. And looking at it as a system, was in Canary, we're trying to see how we can instrument it using the biology as a sensor and make it like the same abilities like in machines so we can enjoy both of the worlds combining together. GLOVISOR: Absolutely. But before Canary Detections, who was Dorr? And what path led you to where you are? Did you come from biology, AI, security? Dor Woolf: And so I'm coming from both worlds and actually I'm combining a few walls together. I'm combining also how you're taking technology and put it in all kinds of fields of security and defense. And also I'm a biologist, I'm a marine biologist and an animal behavior expert. And so I will not say that I'm the classic AI person or a technician person. I am more a person that knows how to combine those both worlds together. ⁓ In the end of the way, Canary is exactly what we are doing. We are taking the ability of the biology and taking the ability of the technology and combining it together. This is our mission. Even though some people it looks impossible doing it, we have a very good success in Canary doing so. GLOVISOR: Gotcha. All right, so that's a lot of experience, but you know, tell us actually what Canary does, but in plain English. Dor Woolf: And so Canary is taking the animal that is becoming the sensor. The animal is not our product. We are taking a sample from the field. We can take an air sample from any enclosed space or a swab or any sample that you want to take. We're collecting the particles on a small, let's say filter or device or whatever that we collecting. We put it inside the machine. And inside the machine, our animal is giving indication to the system if there is any positive or negative sampling that we're trying to collect. And our machine is doing the animal behavior analysis and displaying it on a software with a real-time performance of the animal. So basically, we're combining, we're using the animal as the sensor. But it's not only about the animal saying yes or no. It's more about our sophisticated system that doing the analysis of and projecting it on a software GLOVISOR: Gotcha. So is it any type of animal or is there a specific one? Dor Woolf: So as we were thinking about which kind of animal we can put inside our system, so an elephant will not work even though that he has a very long nose. You need a big machine. So we're looking for ⁓ an animal that can work and will be happy to work inside a small, let's say, ⁓ a machine. GLOVISOR: Yeah. Dor Woolf: So the best animal that can do it, have exactly the same abilities like the canine of the rats. I know that they don't have like a good reputation, but we're working with the nice one, the cute one, the white, even some people use as pets. And actually their natural behavior is to be inside enclosed spaces. If you will take the same animal and put it outside, let's say on the field, she will die from heart attack. So she prefer to be inside the enclosed space. So it's the perfect animal to work inside the space. and everything that we're doing inside it, she's really enjoying to be inside. She's waiting. It's like a dog that's waiting to go outside to a walk. She's waiting to do the job. GLOVISOR: Excellent. So now, based on your experience, did you know exactly that this would be the right animal for this process, or was there a little trial and error? Dor Woolf: ⁓ So as we were thinking about how to actually to give the solution to the detection, as you mentioned before, it's not that the machine are not working and it's not that the K9 is not working. We're not coming to replace not the K9 and not the machine. Each one have his own advantage, but each one of the system ⁓ is a bit falling down in some areas. As I mentioned in the beginning, K9 are falling down. on the way that it's more reliable on the handler. So it's made the human connection between the handler and the canine in order to do the job. And if you're working with a machine, so machine doesn't have the same sensitivity like the canine. So we're trying to close the gap. So the best animal that we were thinking to work with, it was the rats because they are very easy to handle, first of all. And like I said, they have a bad reputation, but they really like when you're petting them. We will change the, in Canary we will change how people are thinking about rats, but they very easy to handle and they're also very, very smart. A rat can memorize more than 60 different odors, ⁓ only one rat. So they are very smart animals. They are... ⁓ Very low maintenance. is another thing that we're thinking about because if you're coming to replace the cannon we want to give and not replace over it if you want to Give another benefit for the canine can and are very cost-effective ⁓ very cost Efficient because you need to take them out walk them bring food as you know All of us have dogs at home said they need detention. So rat is more Easy to maintain and you have like a box like a hamster box at home and you just need to give food and water you don't need to walk it and it's not and it's not barking it's very quiet GLOVISOR: So now, what are they exactly measuring? Dor Woolf: So our machine is basically collecting the signal from the animal behavior inside our machine. After a sample is being presented inside the machine, our machine is calculating the animal behavior and giving us a very clear result of a hit rate and a false alarm rate. On top of that, we are now working on an AI confidence. Another advantage is in our machine that everything is being captured by our software. So we can give another an AI ⁓ layer of detection because the machine is learning the animal. So we can give another layer of confidence beside the traditional hit rate and false alarm rate that you're getting for, let's say, working with a machine or with a canine by himself. GLOVISOR: So the animal could actually give you the hit and false alarm. Are there any other? Can it just give you a clear as well or? Dor Woolf: So yeah, so it's giving me a hit rate, or it's giving me the hit rate or the false alarm rate. I will explain if somebody doesn't understand, a hit is when the animal is saying yes, or the machine is saying yes to a positive ⁓ target order. Let's say that you want to detect assuming TNT, and if we're presenting a TNT to the machine, the machine need to detect if there is any target order, so a hit when the machine is saying yes. A false alarm. It's when the machine is saying yes when there is nothing there. On top of that, all the other sampling that's through our machine, if it's clean, it's coming out clean. GLOVISOR: Excellent. ⁓ You cover explosives, narcotics, allergens, and medical detection. ⁓ Are those all the same underlying system or does each domain require completely different training? Dor Woolf: So this is what is nice about our machine. We especially designed it for it. ⁓ Our machine doesn't need to change the workflow. Basically, the machine is staying the same, the animal inside the sensor is staying the same, all our, I wanna say SOPs is staying the same. The only thing it's changing, it's the target order. This is what helping our system to scale up compared to other system. If you want to take just for example, will compare it to the K9. Let's say a canine that know how to detect explosive need to work in environment that usually will find explosive. So you need to train a lot of by let's say behavior because they need to work in the same field. In our system, the animal is not going out there. We bring the sample to our system. So the system is staying the same. We just need to change the target order and that's it. It's the same thing. GLOVISOR: And how long is the training process? Dor Woolf: So the training process, I want to say that it's, if you're taking an animal, it's about two and a half months that you need to train the animal to understand the basic of the logic of the system, how to work inside the system. After that, this is what is very nice. This is why rats are so smart. If you want to ⁓ add another target order, it can be in less than a day because she already know how to work inside, just saying. This one is still positive one and that's it, you got it. So it's very fast training. Like I said, you don't need to train on all other behavior like how to sit, not to bark, how to behave in public. So the training is very fast and one person can train more than one rep. Also our system is working automatically. So we're trying to eliminate this human. as we can, not because human are not good, because there is a big problem, like I mentioned in the beginning, when you try to combine the animal and the human together. We try to take and make it as scientifically and lab as we can. GLOVISOR: How long does the training compare to a K9's training? Dor Woolf: So again, usually you're starting to train when it's becoming mature, first of all. So you need to train like the basic. It's take like one year after you've been getting the basic training, how to behave, how to work, how to be ready. And after that, you need to do the advanced training. This is depend on the target material that you want to do. If you want to be explosive or narcotics or in the custom. So each one need to have additional training. So it's much faster comparing to the K9. Yeah. GLOVISOR: much faster. I can see that. Now you designed the hardware yourself. What was the hardest engineering problem that you didn't expect at the beginning? Dor Woolf: Yes. And so the hardest thing, you know, when you're trying to take animal behavior that it's not a constant and you want to make it ⁓ more like a system. So you want to create a system that it will work exactly the same and you have repetition exactly the same time. It doesn't matter which, which animal you're working or if she's waking up on a bad day, like happen also to humans, you know, some of us waking up and we're not feeling good. So our system, the hardest thing, was, I will try to explain. If you're building a new device that have a hardware and a software, so if something is not working on the hardware side, the people from the software say it's a software problem, and likewise. So, you know, they're fighting between them. Now suddenly you have another wheel that it's the biology behavior. So all three wheels need to be working combining. If one component inside the machine is not working, then you have a problem. So the hardest thing for Canary, is to find exactly the way how those three wheels are working together. So it's a required design because it's a small animal and you need sensors inside it. And on top of it, you need sensors that know how to read and software that know how to calculate all those data and project it in a software that can see very clean and consistent results. GLOVISOR: process. Yeah, that makes sense. ⁓ So now what is one benefit that you probably have identified since the beginning that you didn't expect? Like one of those happy moments, know, something that just came through and you're like, wow, I didn't expect this to be the case, but it's much better than I expected. Dor Woolf: repeatable process, ⁓ I will need to explain the question. GLOVISOR: ⁓ What is one thing that you didn't expect when you first started canary detections that has been a very happy moment that you've thought this is a great plus? ⁓ Maybe the rats are doing better than I had expected. I was able to train them faster than I thought. Dor Woolf: Okay. So actually there is a lot as you go, as you're working and you're finding your future customers. So as you try to imagine, it's coming from the security area, the Sheriff K9 unit. Even now we found a very big, I would say pain for the... fire investigators. They're relying only on the canine and it's a very limited resource. So as you're showing them what we can do at the beginning, even if you want to compare it for the canine units, at the beginning I was afraid that they will think, you know what, this is the future. It's going to take out all the canine and we will be very, you know, not brought this system because we are dog trainers. But it was surprisingly the opposite when I showed, ⁓ let's say the sheriff from the K9 unit, our system, he literally sat back in his chair and said, ⁓ my God, it's amazing because we're resolving all the problems that he knows about it. Like I even didn't mention all the problems, but it's much more problem when you're working with a live animal and you need to take care of it and you need the connection. So you're getting a very strong feedback from those people that seeing the potential of our system. Like I said, it's not going to replace the canine. Just for you to understand and for the audience, I cannot send the rat running on the street looking for explosive. It will not work. It's not the type of system that we are doing. So the canine are still necessary to do and go find exactly if you want to find a target order. ⁓ Our system is more for the high volume. Imagine that you want to check ⁓ UPS truck, shipping containers. Instead of unloading it from the ship, we're just taking the sample from inside and in less than 20 seconds, we can give indication if there is any positive target order that we want to find inside. And also it's non-invasive, so it can work in theme parks. Imagine that if you're going now to Disney, you can take an air sample from inside the car when you're buying the tickets and I'm sending you by, everything is nice. And then if there is any positive, I know where you park, then I'm coming to do the search. So I'm searching everybody that's sent inside the car, non-invasive way. Some people are afraid from canine. You don't see any something that you need to be aware of. And I'm getting very good feedback. So it's all the time surprising. how our system is benefit and also like I mentioned, our system can detect anything that have an odor. So when I'm presenting our system, it's all the time coming into mind new fields that need to be looking for a good solution at the moment. So it's great, it's all the time adopting but as a company we're trying to stick for the first customer and after that. our system, like I said, it can scale up very easily to other fields. GLOVISOR: Sure. So now, when you first started, I guess you had the security in mind with this process. Who are your target customers now? Dor Woolf: So yeah, as you're thinking about it, the security is the main, I want to say the sexiest business that everybody is trying to reach. And it's true because it's a very big industry as we know, and it's all the time, it's not going to disappear. Unfortunately, as a society, we're all the time having threats coming from all kinds of directions. And Canary, it's a very good fit for it. Try to imagine that you're sitting on the border ⁓ of Mexico and you want to check cars. You can check everybody very fast. Nobody can get through the system because we're checking. We are designed for high volume. But for us, think for ⁓ starting our technology, it's first of all, it's not about if it's going to work and not trying to convince people that it's working because people from the industry when they're coming and seeing the technology in action, they immediately understand it. Our next is to have a validation because when you have a new technology, it needs to be validated by a third party. So I think that the fire investigator will be our nearest customer or nearest, I want to say, pilot. test because they have a very significant pain and think that canary can be a very good solution for that even though that when we went out there like you said I even didn't think about it. GLOVISOR: It's so interesting. ⁓ So what's the hardest thing about selling biosensing products to a security or let's say a medical buyer? Is it the regulatory path? Is that challenging? Is it trust or is it something else? Dor Woolf: Yeah. So I want to say it's all above. So first of all, yeah, you have the regulatory, I want to say, I think that you need to go over it. And each, I want to say target material have his own regulatory pass that you need to go because medical is not the same as explosive and explosive is not the same as narcotics and in the custom is not the same as the others. GLOVISOR: Okay. Dor Woolf: So each one have his own world of regulation. And the hardest thing for a system is first of all to be a knowledge that it's working, it's consistent and you can rely on it. You know, people are working with K9 for hundreds of years. So everybody knows that K9 know how to do it. Now our challenge in Canary is to show the world that we can produce the same result and even better. GLOVISOR: That's a total reframe right there. So five years from now, ⁓ what does the world with canary detections deployed at full scale look like? Dor Woolf: So in five years, canary detection will be out there on the field. In the security area, I believe that when you're going to the airport, probably you will not see the rats because still people doesn't want to see rats in life, but you will have machines that are checking the luggages, checking all the FedEx, UPS, all the commercial cargo that being fly in aviation. Or like I said, I think that in the borders, it can really secure our borders and make sure that no drugs, no explosive, even like I said at the beginning, every time that we are developing to a new field, there is a human smuggling from the borders. So our machine can give very good result. And even we have talks now on the medical field. It's still at the beginning, but there are big interest of detection cancer in the early stage. So in the future, I believe that our system will be working as a prevention for early stage cancer. GLOVISOR: there seems to be a lot of potential here. ⁓ But on the flip side, let's think about this. Are there any limits that biology can detect ⁓ a ceiling, maybe of sensing sensitivity, or is the potential always open-ended? Dor Woolf: There is a, I agree. So like every single life, I think there is limitation. There is nothing that doesn't have limitation. I don't believe in that. If somebody is saying in the system that it doesn't have limitation, you need to be a bit suspicious about it. ⁓ But yes, so biology have limitation and we cannot, let's say, detect stuff that doesn't have an order. If it doesn't have an order, it's in a different ⁓ area that there is a good solution for that. ⁓ But we are very far. Like I said, the technology, even though that we have an AI and we have all kinds of crazy technology developing every day, we're still very far to be even close to understand how the animal is doing the detection. So the limitation is out there, but we are not, very, we have a lot of years until we can catch up with what the evolution did. GLOVISOR: Now for someone who's listening, who works in security, healthcare, biotech, what will you want them to take away from this conversation? Dor Woolf: So I want to understand when sometime when you're saying to people that we're using biology as the same source, I want them to understand it's not something that is primitive on the other hand. Now we're all the time moving to the technology. So I want them to understand that biology it's still in some areas is still domaining and there is no other solution than the biologic still. giving us and we need to understand how we can take those abilities that we're having from the biology and making them to being used for our needs and to make the world much safer. GLOVISOR: Absolutely. Now who should be reaching out to you right now? Dor Woolf: And so at the moment in our stage, it's a first of all, anybody that is working in technology area and is fascinating is more than welcome to connect. And we're looking for anybody from the security industry, border patrol, custom, the police, anybody that is coming from the security area. and is more than welcome to connect. And I believe that when you're a collaboration together, your result is much more efficient. GLOVISOR: It sounds like by the day you're making ⁓ going leaps and bounds, going through the learning curve. Fantastic. ⁓ So Dor, where can people connect with you? Dor Woolf: So first of all, anybody is welcome to go to our website, canarydetection.com. In the website, you can see all the information on the fields, all the updates that we're having. Like I said, every day, we're finding out a new area that need to be, ⁓ to have a solution for. Also, of course, on my LinkedIn, of DoorWolf. can send a direct email also to canarydetection.gmail.com anytime, more than welcome. GLOVISOR: Great, and we'll have that information in the description of the video and podcast episodes. ⁓ Dor, thank you for the powerful conversation and for breaking down the work that you do with canary detections. It's been a pleasure having you here today. Dor Woolf: Thank you so much. GLOVISOR: Till the next time. Dor Woolf: See you next time!