Donate
For our final episode of our odor discrimination series, Kayla talks with Paul Bunker from Chiron K9 about generalization and discrimination.
Science Highlight: None this week
Links Mentioned in the Episode:
None
Where you can find Paul: Website | Instagram
You can support the K9 Conservationists Podcast by joining our Patreon at patreon.com/k9conservationists.
K9 Conservationists Website | Merch | Support Our Work | Facebook | Instagram | TikTok
Summary
By Maddie Lamb, with the help of ChatGPT
1. Paul’s Balkan TNT Detection Project:
- About 23 years ago, Paul’s involvement led dogs to detect a specific Yugoslavian TNT explosive in the Balkans, achieving a notable 100% detection rate.
- Surprisingly, these same dogs struggled to recognize a different TNT type in the US.
- Takeaway: For effective detection, dogs need training on multiple scent variations.
2. Grasping Generalization & Discrimination:
- Generalization means a dog recognizes various versions of a scent. Discrimination ensures a dog can tell a target scent from similar ones.
- Factors like environment, an animal’s diet, and medications can alter scent profiles.
- Recommendation: Train dogs on at least three variations of the target scent.
3. Training: Planning and Implementation:
- Establish the goal upfront: generalization or discrimination.
- Using Paul’s “three-legged stool” principle, trainers should expose dogs to either scent variations or potential distractors, ensuring they can differentiate effectively.
4. Addressing Oil Spill Detection Challenges:
- Dogs trained under this project showcased an impressive ability to distinguish between fresh oil and naturally weathered variants despite the complexity of the hydrocarbon chain.
- Future plans involve real-world tests during actual oil spills to validate these findings further.
5. Emphasizing the Age of Target Odors:
- Paul underlines a dog’s potential ability to discern based on an odor’s age. This capability has critical implications, especially in conservation tasks where the age of a target could affect results.
6. The Saola Project’s Unique Approach:
- Aim: To locate the saola, a critically endangered mammal, in Laos.
- Major hurdle: The absence of saola scat for training. Paul’s solution involves generalization, with dogs trained on scat samples from species similar to the saola.
7. Individual Dog Preferences and Tailored Training:
- Not all dogs generalize in the same way. Training plans must accommodate a dog’s unique tendencies.
- Crucially, the dog’s hunt or search behavior is a key focus. A handler’s expertise lies in reading subtle behavioral shifts signaling scent detection.
8. The Value of a Distinct Alert:
- For less experienced handlers, having a dog with a clear, distinct alert is invaluable. It ensures more accurate detection and streamlines the process, reducing potential errors.
Transcript (AI-Generated)
Kayla Fratt 00:09
Hello and welcome to the K9Conservationists podcast, where we’re positively obsessed with conservation detection dogs. Join us every week to discuss detection, training, canine welfare, and conservation, biology and everything in between. I’m Kayla Fratt, one of the cofounders of K9Conservationists, where we train dogs to detect data for land managers, researchers, agencies and NGOs.
Today, I’m really excited to be rounding out our Discrimination and Generalization discussion with Paul Bunker of ChironK9. Paul has been on the podcast before so many of y’all are already familiar with him. And Paul has a couple different projects he’s going to be highlighting for us as far as generalization and discrimination and how those things play into getting successful projects off the ground. So, Paul, take it away.
Paul Bunker 00:54
Hi, Kayla. Thanks for the invitation to come back on again, very much appreciated. And I’m happy to talk today about discrimination and generalization. First of all, a little background to my learning experience with generalization. So 20 whole three years ago, I had a project. It was looking for a particular type of TNT explosive in the Balkans, and we’d received Yugoslavian TNT, and imprinted the dogs on those and we were training the dogs to very trace amounts of the explosive because they had to find TNT, which had naturalized buried within the environment.
So the dogs were really good achieving 100% detection down to 0.25 of a gram of TNT. Now, some of the dogs after a couple years of work, they were brought to the United States to support a US military based project. And when we arrived, we received some TNT from the US ammunition supply. And they look different. The TNT we’ve been working on was a yellow color, and the one we received is more kind of a green color. You know, in my mind, TNT is TNT, and the dogs were really good at finding TNT. So I didn’t see any issues. When we pressed the explosive out the TNT and work the dogs, or none of them would respond on the odor. And we had six dogs. And that was very confusing to me, we spent, I don’t know, less than half a day imprinting quickly pay on sniff typical processes. And by the end of the day, those dogs were responding 100% on even trace amounts of that TNT.
So this confused me as to why I had six dogs that were really good at detecting TNT. And then I presented another example of what I thought was exactly the same substance, and the dogs couldn’t generalize. And it was Dr. Goldblum from, I think he’s Tel Aviv University that had written a paper and talked about an expert nose and saying, you know, if you train a dog on one particular odor, regardless, if there’s other examples of that odor that exist, the dog will become so specialized that only respond on that particular odor. And once I learned this, then we started to generalize the dogs across all exam examples of TNT and never encountered that problem again.
So that was kind of a big learning curve. For me that generalization is very important. And I think there are a lot of organizations, whether you’re law enforcement, military, conservation, etc. The limitation is that your targets are dependent on you sourcing them. And sometimes the source might be from a zoo, for instance. And therefore the potential exists that you’re only training a dog to find a captive bred example of your species. And that scat, if you’re using scat, for instance, the odor of the scat is dependent on the environment that the reptile is living in or species whatever it is, the food is being given a new medications it’s receiving, etc. And that older is potentially very different from wild species. And therefore, you could have a generalization problem.
Now, the amount of odors required for generalization is, as far as I’m aware, not yet known. And I know Dr. Nathan Hall gave a presentation at K9SciCon, and was talking about generalization. And there is research on gearing, but I don’t think anyone no one has come up with an exact number and I don’t think you have a word because each dog is an individual and some dogs being pedantic and some dogs being liberal, they will make choices based on their learning experiences and the older presented to him. I genuinely say you want at least three targets. And I say the least because you want as many as possible. I know others have said for five examples of different targets.
So I think you just want to get as many different examples of your source target to allow for generalization. But that also underlines the fact that discrimination can exist, whether it’s on purpose, because you particularly want a dog to only find one specific odor. It could be that inadvertently you train the dog to absolute discrimination because you’re only presenting one example of a target. One of the biggest challenges, potentially you face is sourcing your targets. And this is where and developing your training plan is extremely important.
First of all, you need to determine do I need generalization? Or do I need a discrimination? Once you’ve determined what you actually need, then you need to look at the sources and determine what how many Or where can I source different types of the same target odor. So I can get generalization. Or, if I want discrimination, I need to really ensure I have the exact target I’m looking for, so that I make the dog the expert at discriminating that target from others. Once you’ve established the target, you need then genuinely I in print or not generally I do in print in a clean environment, I get as clean a source of the target as possible. And I imprint in a clean environment, I want the dog to 100% Understand the headspace of the target.
Those of you that have either read my workbook or have heard me get presentations before I talk about principles of the three legged stool, which is the principles I use in the imprint process. The first one is that I use trace amounts of a target, but the other to talk about generalization and discrimination. And then with the generalization, we talk about the act or process whereby the land response is made to a stimuli similar to but not identical with the conditioned stimuli. And if I want generalization across targets, at the end of the initial entry period, I start my generalization process within a lab or room or whatever you want to call it type controlled environment, before I move outside with discrimination is the ability to detect the target despite it being mixed with other materials.
And this is slightly different from maybe the discrimination that you’ve heard before. It’s just a term I use in that one of the steps I do take during Imprinting is to take the target odor, and I mix it with examples of the environment, or distractors. What I want the dog to understand is how to pick out its odor when it’s been mixed in with other odors. One of the reasons for this is if you’re continually presenting clean clinical examples of your target, that is potentially not what you’re going to find in real life. Because if you’re going into the field, and obviously you’re working in woods, and there’s animals and insects and all the leaves from the vegetation and grasses, etc. They all intermix with your target odor. And they can change it, particularly if it’s buried, because the target odor moving through sand or soil or whatever the substrate is gets filtered, and the headspace presentation is different. And we also want the dog to be able to actually pick out its target from within other odors.
So when I talk about discrimination within the three legged stool, what I’m talking about is that act of mixing my target with other odors to teach the dog that there’s a range of headspace profiles that their target could potentially present itself in and is still target. Now we obviously have to add to that discrimination in that it might be a discrimination between different species. It might be I’m training a dog to find lizard scan, and I’ll use snake or other reptile scats as a discriminator, which I expect the dog to ignore. That is also discrimination obviously. So you can class that discrimination the same as I would with MCs I do the same as I would when I’m mixing my target within distractors. So that’s kind of the background to my approach to generalization discrimination is planning my project, it’s working out how this fits into the process. And I start very early on in the imprint process.
Once the dog understands odor, I start to then either generalize or discriminate. And even if I’m generalizing the odor is mixed with other odors to get that discrimination from the environment.
Kayla Fratt 10:33
Maybe we’ll get into this with your examples, but when you’re talking about mixing the odor with odors from the environment, are you doing that by going out and collecting odors from the environment? Or are you doing that by placing the odor in kind of naturalistic environments and letting the environment kind of do the training?
Paul Bunker 10:52
The first steps is actually collecting examples of the environment and bringing it into the lab. And then I mix it in mason jars. And I’m using whether it’s lineup or carousel or whatever I’m doing. And that’s where the dogs receive its first presentations of that change of target in a controlled environment. So that in some way, I can control what the dog has been exposed to. So it might be that I’m going to start with a very neutral odor, maybe some steaks that’s mixed in with the target, and then I’ll move to leaves and then I’ll bury under some native soil. And I’ll start to increase the complexity of the odor that’s actually mixed with that distract off the distractor, this mixed with the odor. So again, we’re getting different changes, or presentations of this head, same headspace and the dog has to discriminate, which which distractor contains its target. In the other containers, the other parts, there will be examples of the environmental odors as well.
So today, for instance, we collected some bamboo, some leaves some sand for the project I’ll talk about in a moment. The first steps will be the highlight of clean odor out. And then as distractors in separate parts are lamp leaves and bamboo and steaks, and then and sound. And then the next day, potentially, I will put the target odor in with some steaks. And then the next day, it would be with leaves, and the next day with sand buried. So I’m starting to increase the complexity. But also at the same time, there is the same distractors in the pots as distractors. So the dog has to ignore examples of the leaves, but only respond on the example of the leaves that contains his target. Hopefully, that’s a bit clearer.
Kayla Fratt 12:49
Yes, that makes perfect sense. And I love that setup. And I love how clean it is. And I also what I’m really liking here is how similar this seems approach wise to some of the approaches that were described in early episodes of this discrimination series that we’re doing. Because it’s nice to hear that a lot of the folks who are thinking really hard and cleanly about doing this cleanly are doing it in similar ways that to me suggests that there there is actually more of an understanding of how to do this. Well, then maybe I initially thought which is exactly what I was hoping to find. So why don’t we get into if you’re ready, some of the case study examples that you’ve got ready for us?
Paul Bunker 13:32
Yeah, sure. So the first one I’ll quickly just mentioned, because it is a published paper, and it’s open source. So people are very welcome, obviously, to go online. Just you can download it. It was Dr. Mallory DeChant, myself, and Dr. Nathan Hall of Texas Tech University. And what had happened here was that we deployed up to Canada with some oil spill response stocks after a major incident up there. And the banks had achieved over 10,000 confirmed alerts, responses on oil. But some of those responses on oil were very small amounts that in a typical spill response would not need cleaned up. Because biodegredation and bacteria would actually work on those small examples of oil, and just, they will disappear into the environment, and they’re not worth being actioned on. However, if a dog responds on example of oil, it has to be recorded data recorded GPS, photographs, measurements, all those sort of things and reported at the end of the day and collected and disposed of even though it didn’t need to be if it was just a human survey.
So one of the questions from that was, well, can you not calibrate a dog to actually only respond on a particular level of a target? I had the belief that we could. We received some from funding from American Petroleum Institute to investigate, and Dr. Nathan Hall and his team very graciously said they would take on that research. The paper is called “Stimuli or stimulus control of odor concentration: pilot study of generalization and discrimination of odor concentrations in canines.” And it was published in on the 18th of January 2021. And basically, in this study, what they did was they trained dogs to a particular level of an odorant. They then saw or settled at how far below that level, that threshold they’ve been trained to, the dogs will naturally generalize. And what they found was the dogs will naturally generalize to 10 fold below, and then start to ignore the target. So one of the steps they then took was to actually reinforce the discrimination of the target based on amount. So the dogs were taught a particular threshold, when they got when the odor was presented at 10, far below or lower, and the dog ignored it, it was reinforced with a reward or treat. responded on the threshold that had been taught to actual respond on it was reinforced with a reward or a treat, in this way, with the dogs were able to be calibrated to discriminate based on amount. And the dogs did really well.
You know, it was a pilot study, there was only 30 dogs in the actual study. But it demonstrated that the potential exists, that you can actually train a dog to discriminate based on quantity based on the odor profile amount that’s been presented. And I kind of knew that that would happen because we had a limitation in one of the canine schools, I was an instructor for six years. So a lot of experienced, hundreds and hundreds of dogs went through our program, and that we were not allowed to alter the size of the target that was issued to us from the ammunition supply in like Seafarer explosive, for instance, comes in a one and a quarter pound block. That’s exactly how we had to train the dogs and nothing else one and a quarter pounds.
And what I saw was when we went to the field phase in Arizona, but the dogs would struggle on quantities higher or lower than that. And my thought was because we always train on one and a quarter the dogs only find one and a quarter. So when they asked me, “Can we not calibrate the dogs to only respond on certain amounts,” I kind of thought this potentially does exist based on my experience of a training limitation that we recognize. And we would actually then train the dogs on various amounts when we got out to Arizona because we have a lot more flexibility on the amount of targets we were able to place out.
And this study actually demonstrates that that potential exists, so quantity is important. I also, if I recall correctly, and I was catching up with your podcasts yesterday on a long drive I was on that there was a science highlight about potentially dogs can discriminate based on heat. And that’s not necessarily something we think about. But I know there has been research that dogs can detect variants in here and discriminate the one level of heat from another or potentially the exit that exists because the nose is a thermo sensor as well. So that was just a quick study I wanted to talk about. And now I’ll go into a one of our projects, which involves discrimination. This project, it’s been a few years in development. And the background is that down on the Gulf. This is an oil spill response project down on the Gulf Coast and I’m in Texas, about two hours from Corpus Christi, they have a lot of tar balls washed up. Now tar balls are weathering that from natural seeps out at sea and oily seeping into their sea all the time on the Gulf Coast and over in California, particularly there it could be from ships dumping some animal out at sea, or it could be when they’re being refueled and some oil is spilled but there exists within the environment. Energy is out at sea it gets washed up. And these tar balls can be anything from the size of your little fingernail or smaller all the way up to several meters in size and I broke down every month doing surveys for Texas General Land Office on the beaches, and we’ll find anything from the lowest number I’ve ever found is eight with my canine and that includes Barry topples and the mostly families over 1600 In one survey.
So, you can understand there’s a lot of tar that’s been washed upon the beaches and I’m sure you know listeners on know if they’ve been on a beach, particularly in in the Gulf Coast or California areas that at the end have been on the beach, there might be little black spots on their feet on the terrace, I mean, and genuinely that would be tar. All over in this rush job, and it’s within the environment. And it’s absolutely natural.
The problem is, if there’s been a spill within the Gulf Coast, and I take my canine down, she doesn’t know which is spilled oil, the fresh oil that’s been washed up from an oil spill off offshore or river delta that naturally exists for the environment. So she’s going to be responding on all of it. The issue there is obviously when you have 1600 tar balls in a pretty small area. And it’s not a huge area that we survey every month. That is a lot of time, it’s a lot of effort, because you have to record it data collect etc, like I said, but also that dog has to be rewarded or reinforced infinite, that’s praise or something. Every time it has an example of a find there has to be some sort of positive interaction. And I have a daily that is extremely tiring for the dog.
On top of that, the responsible party, the people that had the oil spill, have to pay for the responders to actually conduct the response. They’re not going to be happy if they find now I spend 80% of my day, actually finding other people’s oil, whether it’s naturally washed up or ships that have dumped it or whatever, you know, they want to pay me to clean up the oil they’re responsible for. So that was a real issue. And that was raised by Texas General Land Office. You know, if you come down here, we’re going to spend all your time just finding stuff that exists within the environment and not the stuff we’re looking for.
So, again, a discussion with Dr. Nathan Hall. We theorize that potentially you could train a dog to discriminate oil based on age and Texas rental General Land Office actually funded the research because they really would like canines to support oil spill response, but they couldn’t support canines that are responding to all this naturally existing oil. So the first phase was that Dr. Hall and his team and it was Dr. Mallory DeChant actually conducted research at Texas Tech. They had some of my dogs and one of their own dogs that were green at the time and very imprinted the dogs to to fresher examples of oil, crude oil. And then they introduced weathered crude oils to the dock and they used an olfactometer, which is an automated computer driven device that presents odors in a very particular way, but it’s double blind. So the computer decides when and how its presenting the odor. And the only way we know if the dog is correct when the computer actually gives a beep and you reinforce it.
So they introduced weathered oil as the discriminators. And the dogs did phenomenal. They very quickly could pick out the difference between fresher crude oil, and aged or weathered crude oil, which to me was super interesting and actually surprising because crude oil is a complex odor. And it has a long hydrocarbon chain, particularly the type I use for imprinting early on is a West Texas Intermediate crude, and that has a very long hydrocarbon chain. And I was thinking and that’s one of the areas we train the dogs on, or Dr. Hall’s trains the dogs on, because they had the longest chain so it would capture a lot of the other crude oils. And I was surprised how easy it was for the dogs to actually learn, “Well, I only respond on a fresher example of crude oil. And I completely ignore the weathered examples.”
So they’ve completed the research and said yeah, you know the results there was one of the dogs I want to say it was 100% The others were well over 80 to 90% detection rates within the air fight Tommy to differentiating between fresher oils and crude oils. So it was a really clear example that the dogs could discriminate.
I then took one of those dogs that have been on the study and I took a green dog that I had and train them in field survey so taught them why they research quarter in question, whatever you want to call it off leash detection. I took the dog that Texas Tech already trained with the olfactometer and I trained them on the carousel and older stands, and then out in the field and tags using the training aid delivery devices, because that’s what we’d have to use on the beach.
And then I also took the green dog. And I developed a training program based on the lessons learned from Texas dog, and actually trained her through the processes of a dog trainer, if you like. So, here’s a green dog training to be this specific oil detection dogs. And that’s what we call them in the end. Talking about them to Corpus Corpus Christi, on two separate occasions, the first one went down for a couple of weeks, worked him along the beach, and not once did he give a response on any naturally occurring oils. No toggles no tar powders, there was a tar Patty, which is a large table, it was a good three feet by one foot in size massive. And it was still you know, it wasn’t drying, it was a little wet inside, but it had tons of odor on it. The dog, when I got it on video with a dog went over sniffed it and carried on searching. Whereas when I broke my dog behind the Chi obviously responded on this huge amount of oil, but also all the other examples of tar.
So the discrimination of this canine was really interesting, it kind of shocked me a little it was that good. And since then we’ve conducted monthly surveys, the specific hole detection dogs go first, they’ve never once responded on any example of weathered oil. And they’ve always responded on the tags containing fresh warnings, and then conduct a survey afterwards, we might generalist oil detection dogs, and she’s like, say cheese found anything from eight one month to one and a half over one and a half 1000. Another month, typically, we’re looking at 100 to 300 tarballs per survey that she’ll find. So that level of discrimination just demonstrates the capability that the canines have, when we’ve taken something with a complex array of embodied odors, such as a long hydrocarbon chain, and just remove some of those hydrocarbons through weathering that the dog can tell the difference very easily. And be able to demonstrate in the field more importantly, that it is able to conduct this discrimination outside of the lab environment. And I think that, again, is another important part of this research is there, yes, potentially within a lab, you can have this absolute discrimination using an Olfactometer. But when the dog goes onto a beach, and it’s real life, and the presentations are being made, and the oil has been heated by the sun, and it’s melting, and it’s given off all these odors of hydrocarbon, etc, the bog standard was able to discriminate. So the actual applied phase really does discriminate just how incredible they are at being able to tell the difference based on age in this place.
Kayla Fratt 28:14
Yeah, that’s absolutely fascinating, and really, really encouraging, I think, to hear in a lot of ways. I mean, I love how we were able to grow to go from. Okay, so we know that this is possible, because it’s a problem for us. So that example of C4 that you gave earlier, to kind of confirming that and research to then saying, oh, so we can we use this to our advantage to meet the needs of a specific client. That’s, you know, it’s just a nice, nice, clean progression for us. Yeah, what else? What else do you have to say about that particular project?
Paul Bunker 28:50
Yeah, so that’s when we actually completed the first year of research and it was 100% success, the two bands will say available each they will demonstrate the will not respond on naturally weathered oil, and then only respond on presentations of fresh oil. We do have a potential follow-on period of training, just to gather some more data, but at the minute that’s completed, and it’s been a complete success, you know, there’s no issue that, that was 100% success. Now, we just have to wait for a real spill to demonstrate in real life.
But, you know, I think we’ve got to a point where we can’t prove any more than those dogs can discriminate based on age of target. You know, and that does exist clearly when we think about it. In the conservation world. If you always train on old scat, then potentially your dogs not going to find fresh scat, or if you always train on fresh your dog may not locate aged. So you do need to provide those across-the-baord generalization examples if you need generalization based on age, but also it demonstrates, if you want a specific age range of a target that the potential exists, that you could actually train the dog to discriminate the same target.
But based on age, and, again, that’s something that I think is very powerful, both as a learning tool that we have to understand that could be a limitation in training if we don’t plan for that. But also, it could offer you an opportunity to conduct a specific detection project. If someone was saying, Well, I only need to find a certain thing that’s of this age range. I don’t need to find particularly old examples of this scab because then it’s not worth me assessing for DNA or something.
Kayla Fratt 30:49
Right? Well, I know when we were in Kenya, we had some conversations about there was interest in a project where they wanted to find this scat of lactating female bongos, because there are problems in zoos with keeping baby bongos alive. So they were curious to see if they could find the scats of the lactating wild females and then look at any differences between the female, the wild females and the captive females. And, you know, anyway, and it was one of those things that the project kind of died in the discussion phase because of some behavioral characteristics of bongos that make it potentially really dangerous to have detection dogs out in their territories.
But it is a great example of, you know, where we would want the dogs to be much more specific than kind of a “normal” versus then something I’m actually dealing with right now, where Barley has worked on puma scat in Guatemala, and we are now about to head to the California coast to also be finding puma scat. And, you know, we’re trying to figure out how to get all of the samples that we need in order to give him a couple examples of California Puma to make sure that he knows that hey, we don’t just want Puma that eat Scarlet macaws and howler monkeys, we’re also interested in Pumas that, you know, eat cottontails. And, you know, whatever else the California Pumas are eating. So it’s just there’s so many examples of where this comes into play. And I think it’s something that is to be ignored at a practitioners peril.
Paul Bunker 32:25
Yeah, do you ever goes back to that you know, how important your training plan is, and having this written and articulated in your training plan to ensure that you understand specifically what you’re trying to, to achieve, but then you are putting protocols in place to actually achieve that goal. And of course, there are limitations, you know that there’s a case where you may not be able to get specific examples of your wild target. But at least if you start and do some generalization, which I’ll talk about in a moment, then you can do the best you can based on what’s presented in front of you. But you have to be doing the best you can, as we can see in the first place. If you don’t, then obviously, you run the risk of selling yourself short, but more importantly, selling the canine short, because you’re not given that level of communication it needs.
Kayla Fratt 33:18
Yes, yes, exactly. And that’s, I mean, that’s actually exactly what we’re running into with our upcoming project is we just got our, our DNA results back from our training samples, and none of our training samples came back as amplifier with Puma DNA. So now, you know, the question is, were they too old and degraded to come back as Puma and they still may or may not be useful as training samples or, or not. So right now we’re erring on the side of not training and hoping that Barley will be able to generalize. You know, we’ve and we’ve got some plans in the works to get some California Puma scats to Barley before fieldwork starts.
But yeah, it’s I mean, I think that’s always important to acknowledge that this we’ve got our perfect examples and our textbook examples and then we do you know, the best we can within that so.
K9Conservationists is thrilled to offer a self study on class for those interested in joining the field of conservation dog professionals. This course includes 18 modules of video lecture assigned reading homework and quizzes. We have lectures from 10 amazing guest instructors, including Fostering Motivation and Joy Through Hide Placement Training with Laura Holder of Conservation Dogs Collective, Safety Training and Assessments of Dog Teams with guests Fiona Jackson and Tracy Litton of Skylos Ecology, Special Considerations for Insect and Plant Training Samples with Arden Blumenthal of the New York New Jersey Trail Conference and Building Networks, Acquiring Clients with Paul Bunker. Our alumni group is active and supportive and we welcome students of all levels and backgrounds. The course is priced at $750 with generous financial aid and discounts available for Patreon members, learn more and sign up at k9conservationists.org/class.
Paul Bunker 35:04
Yeah, so then, kind of transitioning into generalization then you talked about a Bongo. Bongo is one of the targets we’re actually training some dogs on right now. So the background is I’m supporting Working Dogs for Conservation on a project, which will be going into Laos. In Asia, and we’re supporting the Saola Foundation. The saola is a actually is a bovid. It looks to me, and I know explain these things kind of deer-like, but it is a bovid. It lives in the border region of Laos and Vietnam, in the jungle environment. It was only discovered in the 90s, late 90s, as even existed, it was so remote and so rare. And it’s estimated there may only be 50 examples of this species in existence. It’s the most critically endangered mammal in the world.
And Working Dogs for Conservation, were tasked by the Saola Foundation to support a two year project in trying to locate sale within its natural environment. And part of that project is canines. So we’re currently training two canines, which will be going over to Laos in August with myself, will be trained in some local teams with those dogs and then there’s a two year project within the jungle allow on the border to try and locate Saola. So the issue is there is no example of captive Saola in existence, and we’re looking for scat. There’s no scat, there is no target to train the dogs on. So what can we do then if we go into a jungle environment looking for species but we have no target on which to train them on? Well, the decision was made to generalize teach the dogs generalization of scat and go into the environment and hope that when they encounter Saola that they will have enough interest in generalization to say, “I found some scat, here we go.”
For that reason, then we are training the dogs here in Texas on a number of different similar odors from captive bred species. And there has been a number of zoos and foundations within the US that have supported us. They’ve provided us different examples of odor like Bongo, to stick to tunda GWA, we’ve got bear; bear do exist in the jungle area we’ll be going to, although extremely rare, but the dogs could encounter bear. And it would be nice for the ecologist to actually know where and if bears exist with that environment.
So we’ve got we’ve got some scenario stats, we’ve got these different examples of scat. And the approach that I took was we train the dogs on three samples first, so set it under a sitter. Tonga was the first one, we then train them on Bongo and Kudo. So the dogs have been working on those three target odors, and then we’re going to start placing out examples of ever target odors that we’ve been provided, and our guar, bear, etc.
But I’m not going to imprint the dogs specifically onto those targets. What we’re going to do is have them in a field environment. In this case, when the dog encounters it, depending on its approach, what we’re hoping is we get some sort of response, which will be reinforced. If not, then it will be paid on sniff to reinforce.
But basically, we wanted dogs to start to offer generalization themselves, instead of us just imprinting the targets we have in the inventory, and then they end up with this array of targets that they’ve been imprinted on. And they understand as we enforceable. So that’s that is a little different than potentially you would normally do, because I’m trying to assess generalization of the dogs purely because when we go to Laos, we’re not going to be able to imprint them on target. And therefore I need that self generalization of the dog certainly enough that the handler can see something is going on and say let’s check this area out if not a full response in generalization.
Now when we get to Laos, we are going to have other examples of those that live with it allow environment and different animals, mammals that live there and bear scat. So we will be able to continue the generalization to some of the targets one that they’re going to be found in in the jungle. Like there is wild pig there’s manager The there are bad and they’re very rare etc. So we will be giving them examples of those types of odors as well. So the generalization process will actually go into theater, and it’s like a four to six week training period. And part of that is in the jungle itself, just to see how they transition into the jungle environment in generalize.
So in this case, obviously, we had a very difficult challenge to train on an odor, which doesn’t exist, certainly within a captive bred or presentation to a canine capability. But potentially, we’re going to use generalization and the dog’s ability to generalize to locate the target, and actually reinforce that, obviously, once Saola scat has been located, then we can use it for imprinting, and then we will just concentrate on Saola, the scat will completely wipe away the other odors, they will go from the inventory, and it will be 100% Saola, and then and then if the dog generalizes to whether it’s scat within the environment, fine. You know, it’s not the end of the world, it doesn’t exist in any great amount is certainly enough to maintain motivation of the dog, but also maintain generalization. But then they would be concentrated on Saola scat once we have some samples.
Kayla Fratt 41:18
Oh, that’s just so interesting. And that was kind of what, you know, I know, I was having a discussion with someone actually one of my potential grad school advisors a couple months ago about this solo project. And they asked me, they were like, how do you think that they’re doing this, there aren’t solids to train. And, you know, just kind of spitballing this, was more or less what I came up with, although it sounds like y’all have come up with a much more elegant solution than you know what I came up with over coffee.
So have you done kind of testing on the dogs preferences and propensities to generalize to help kind of narrow down which dogs you feel are a good fit for this project? Or kind of knowing? You know, I just, you know, we’ve talked about this before on the show with signal detection theory with Dr. Simon Gadbois. Also, you know, I know I’ve talked about this previously, on the show with my two dogs, one of my dogs barley is a very easy generalizer And my other dog niffler tends to be much more specific and literal. So is that something you’re looking at with the dogs that you’re sending to allow?
Paul Bunker 42:23
So one of them is liberal more than the other? So you know, that is a training consideration. And, you know, I listening to that same podcast, you know, about the conservative dog and the liberal dog. Taking that into account means you adjust the training plan to the individual dog, depending on what you want. You know, if you want a dog to generalize, then that level dog is an ideal dog because, chances are I’m always poor, each example of scale finds a way that you can reward me, are you going to reward me? Whereas the the conservative dog is so pedantic that potentially might say, No, that’s not exactly what I’ve been taught. And in that case, you really need to constrain a lot of generalization knowing if you have a conservative dog, less than than a liberal dog if you do want generalization.
So within the training plan, you know, that’s taking into account now we’re only just starting the assessment, a generalization right now we’ve been working on the odors, we also have to train them to find snare, because the local people go into the jungle and they place rows of snares out.
So we want the dogs to be able to locate the snare but fringe set. So that’s again, something we’ve been working on at the minute. We want that in place. And then pretty soon we’re going to start the assessment, a generalization. The two dogs I picked they both Spaniels, one a Cocker one a springer. They both demonstrate that they are liberal dogs. And it wasn’t necessarily specific that I chose that, because I was prepared to actually train generalization as needed. But they’re both kind of offering that that liberalization approach to our training. And we’ve started now mixing the target with distractors leaves steaks, as I described earlier, to get that to get that discrimination phase into him before we actually start the generalization outside.
Kayla Fratt 44:34
Okay, yeah, I think that makes sense to me. And hopefully it makes sense to everyone at home. Gosh, I’m so excited to hear more about how this project progresses. And hopefully you’re able to get some evidence of the Saola. So what else? What else would you like to say or circle back to or expand on as far as kind of discrimination and generalization for for dog teams?
Paul Bunker 44:58
Yeah, I think you know, I’ve stressed a couple of times planning in your training plan is critical. And if you don’t plan for success, then you’re planning for failure. And, you know, it’s down to the trainer, the handler to actually understand the complexities of what they’re presenting to the door. If you’re just using the same time with day in day out, than expect that you are training the dog to find the same target day in day out. If you have opportunities to conduct generalization, you need generalization, then that’s definitely what you need to do. If you need generalization, vary your targets as much as possible. Get them from different sources as much as possible, different ages as much as possible, you know, you need that broad spectrum of presentations for generalization. If you need discrimination, that really were hard and ensuring that dog understands that older profile in all these contexts, but it’s only that one odor profile. And I think all this doesn’t line how important your plan is in ensuring your dog understands what you’re communicating to it, because it can only react to what you’re presenting to it, it can only react to what it understands is the target. And if you don’t present in the target correctly, it’s not the dog’s fault, you know, so plan and ensure your plan is rock solid. And that way you can move forward with the dog, ensuring that it understands exactly what it’s supposed to be doing.
Kayla Fratt 46:31
Yeah, that makes sense. And, yeah, so I’ve got one kind of off the wall question to throw at you because it’s one that I’ve kind of thought about a little bit. And I don’t know if I’ve heard of an example, being done for this. So, for example, when when we were in Guatemala, we were hoping to get the dogs to generalize to anything kind of in the carnivore umbrella. So we had trained the dogs on Jaguar, Ocelot, Guma, Gray Fox, and I think we had a couple of Tate raw samples. And you know, what we saw was Barley was able to generalize to several samples that we thought were jaguarundi, margay, and then some smaller mustelids as well, which was exactly what we were hoping to see.
And just as we were kind of out and talking in the field, we were spitballing over what it would look like had we potentially one of the dogs to find, say all of those carnivores and minus gray fox or something like that, which maybe is a bad example, because that’s the only canid in that list. But you know, that’s where I feel like I start getting a little bit stuck now on generalization and discrimination is trying to think of times where you want the dog to be finding a relatively broad umbrella, but minus one chunk. Is that something you’ve done or something you’ve heard of people doing in the past?
Paul Bunker 47:53
I’m not immediately off the top of my head, but I would not worry. If someone asked me to do that, I think because you can train them discrimination of you know, you can generalize these targets except this one. So ignore this one example. And like presenting that one example, as a normally enforceable target. I think you could train the dogs to discriminate that actually, that isn’t one of the targets that I generalize across. I don’t see that as being an issue at all, unless you know, the headspace is so close to the other species, which is something to consider. And I think you and I have spoke about that. When he was over in Africa that was metric, we want discrimination.
The problem is if the target molecules the dog is using to detect its target exists within a similar species, then that’s gonna be very difficult for the dog to discriminate. And, you know, in some ways, that’s unfair for us to expect the dog to be able to discriminate, because unless there’s something that he’s able to pick out and say, No, that is it, or Yes, or, you know, those No, that is not it, then it’s, it’s very difficult.
Now, I think that if you have a target, and let’s say it’s formed of molecules, any B, C and D, you present that target, and then you present the same target, but the letter D is removed. So it’s only ABC, I think that you could actually train the dog to only respond on a target when it includes this additional odor. Obviously, that I don’t know if there’s anything that exists that’s proven that but I think based on my experience with the hydrocarbon detection dog, the specific oil detection dog and how they can actually discriminate the age of a hydrocarbon chain. I really think we’re not tapping into that discrimination capability that potentially a dog could do. discriminate a target odor, which is missing one of those odors, which is what the specific kernel dog does it, you know, it finds a target, which contains only one extra odor component, potentially, you know, one section of odor compared to the target that doesn’t, which is exactly the same, it just doesn’t contain a couple of extra odors. So I think the principle exists without trying. It’s a difficult one to know. And I’m sure, you know, maybe it has been done. Maybe someone has researched that, but I can’t think off the top of my head, you know, immediately have an example of it.
Kayla Fratt 50:39
Yeah, I can’t either. And I mean, I think your approach makes sense to me and seems about what I would expect. And then, you know, again, we’re always getting into those individual dog questions, I think, you know, I would be a little bit hesitant to put barley on that project as my first dog because he is someone who is such a, he’s so liberal. He’s such a gambler, he loves to generalize. But yeah, it seems like with good training, that shouldn’t necessarily be an issue. And I haven’t, I kind of struggled to think of an example where that would be necessary, where you would really want, you know, all carnivores minus x to a degree where it would even necessarily be worth putting in that training time, although I’m sure there’s, I’m sure there’s an example that I just haven’t thought of for that.
Paul Bunker 51:32
Yeah. Again, the dogs and information tool, and if it gets a response, and you walk over and look and say, That’s not what we’re looking for, carry on, you know, you’ve lost a few seconds of time compared to the amount of training you would have to do to make the dog ignore that one particular target. So you have to balance that as well. It’s a team effort. And if the dad gives me information, and I look at it and say, well, actually, that’s not what we want, but Okay, let’s move on. Does that outweigh the amount of training time for the dog to never respond on that one target?
Kayla Fratt 52:07
Yeah, exactly. That again, makes sense to me. And I think it’s always intelligent kind of think through that cost benefit. And you know, really question whether or not it is an issue. And, you know, so Paul and I are recording on Monday and, we’re having a cheetah discrimination episode that’s coming out tomorrow, Tuesday. So you all will be hearing this before it comes out. But I’ve now got a whole episode, Paul coming out that talks about kind of the, the problem that we ran into in Kenya with those dogs.
You know, the advice, I saw it from you and Simon, and then the approach we ended up taking, and how that ended up going for us. And we ended up basically, to, I guess, now reverse spoil it, because people will be hearing this after, after you and I are talking about this now was that I my main suspicion became that actually the alert was being overemphasized for these dogs. And we went on kind of a path of extinction, getting them to learn that alerting didn’t lead to them getting any information or reward or anything, which I think was more or less one of the approaches that you had suggested to me, it was just to wait out those false alerts.
Paul Bunker 53:19
Yeah, yeah. Again, you know, that’s generally the approach I will use in certain circumstances. But I think I like them to solve the problem, not for me to tell them what the solution is. And I think, you know, the dogs understand a lot more of the background as to why they’re ignoring something, if they’ve worked out, it’s not worth responding to.
Kayla Fratt 53:46
Right. Right, well, and what we really what I really kind of started getting watching the video of these dogs was that it seemed to me that what was happening is they were going out into the scent room and there was always cheetah and there was always you know, it was either caracol or leopard that they were mostly struggling with. And basically, if the dogs alerted to caracol or leopard, they were immediately being told whether or not they were correct. So it basically became an easier and faster and more kind of expedient exercise for them to alert to whatever it was they found first, and then they would either get the reward if they were right, or they would be told where to go next. And it wasn’t necessarily a difficult detection task. So they didn’t really lose anything by making a false alert. And there’s, there’s a really good video that I’ll be sure to link in the show notes where you can actually see one of the dogs as we’re waiting out one of her false alerts like kind of try to alert harder, tried to alert more perfectly, kind of look around a little bit and then re-examine, re-sniff that scat and then get up and just immediately go and find the correct scat. And it was just a really nice example. You can almost see the thought process going across her head of like, “Oh, maybe this? Oh no, I guess not,” you know, now obviously we can’t say for sure that’s what it was. But it that’s sure what it looks like in the video.
Paul Bunker 55:08
I have seen that more than ones that you know, the feedback from the trainer handler is too quick. So the dog learns well, I’ll follow suit here, and you’ll tell me if I’m right or wrong. And I’ll move to the next one, you’ll tell me your move to the next one. You don’t tell me I’m right, okay, I’m getting my reward.
And I had her on the Olfactometer that we had a Labrador, she worked out very quickly, the dogs have to hold their nose in a pole for four seconds. And there’s three ports. And she learned very quickly, there’s no consequence to holding your nose in a pool for four seconds during training. So she would just walk up to the first one hold the nose for seconds, there’s no bleep Correct, she would move to the next one. And she would move down the line until she got a quick bleep and she would get a reward. Because she learned, you know, just respond until you tell me I’m right or wrong.
Kayla Fratt 55:57
Yeah, seconds. Long, you know, it’s, it’s not a huge loss, that makes sense, I would probably do something if I was a dog.
Paul Bunker 56:05
As soon as we added the consequence of a tone, which said you’re incorrect, and the game then finished and the dog received no reward, she stopped my behavior because she knows when I play if I honest response, but if I give a response, which isn’t correct, you’re gonna end the game and the cheese disappears, I’m not willing to risk that. And that can get that behavior very quick. But yeah, exactly. As you said, when you give the feedback too quick, you are actually training the dog to actually just respond on everything and one for you to guide it until it gets to the correct target.
Kayla Fratt 56:39
Absolutely, yeah, and one of the other things that we saw with these dogs is that they had a lot of the handlers had gotten a lot of really excellent mentorship from explosive dogs, folks. And they were really focused on these very snappy, very kind of social media ready alerts that make a lot of sense as a necessity for explosives, dogs. And one of the things that we saw as well was that in these these training scenarios, that dogs might search for 20 minutes source, a pretty challenging puzzle, and then get there. And if their alert wasn’t kind of perfectly square to the target, and they weren’t perfectly focused, staring at the target, their reward would be withheld until that alert was fixed. So the dogs were really learning this lesson, over and over and over that the alert was the most important part of the search.
And I think all of those things kind of came together. You know, and I don’t think there’s anything wrong with wanting a really good alert on your dogs. But it seemed like the there was a little bit of a back a balance issue in some of their training programs that in my mind was part of the part of the problem there.
Paul Bunker 57:47
Yeah, again, I’ve seen it you know, when I want to say it was Simon Prinze, and I had a discussion on a Facebook group or something once where I think he’d asked what’s more important, the response or the hunt, the search, you know, and my take on it was actually the hunt. Because as soon as I see a change in behavior of anytime I’m going to call it, the Bob doesn’t have to give a beautiful say. And in fact, in Canada, we are in one day out 183 finds, that is 183 responses. And by lunchtime, my dog is not giving me a classic set, you know, he’s changing behavior, which is something that I can say there’s something here, and that’s enough for me, I don’t expect that full, nice response out in the field on a long day and never covering 20 miles a day. But I do need that depth to hunt all day. Because if it’s not going to hunt, I’m not going to see a change in behavior, because it hasn’t, you know, it doesn’t do anything when it hits the odor.
So to me, Hunt is important. As long as there is a change in behavior, I can read that I did that I can identify and say there’s something here. If that’s not a felon, suit response, Soviet, you know, that’s my job as the handler to read my door when the information is given me. And that’s the approach I take.
Kayla Fratt 59:10
Yeah, yeah, I mean, I certainly don’t disagree. And it’s been something that has been a really good reminder for myself, as I’m working through with a younger, greener dog, as well. And as I’m moving more and more into mentorship, and then there’s always that balance of, I do still want the dog to have some sort of alert behavior, particularly if that dog is going to be paired with a less experienced handler.
We have a lot of students in our mentoring group who have a little bit more of an informal alert or kind of a look back sort of alert. And I’ve had a lot of kind of going back with my less experienced students on whether or not that is actually the wisest choice of an alert for them and their dog because if they’re not really really skilled at the dog, and then the dogs alert is something that to me, look back alert is is very ambiguous in a lot of cases, you know, is that really what we want to be pushing the dog towards? Or, you know, in training? Should we set up something that is clearer? And that way if that if and when that degrades out in the real world, we still have something that is a little bit better to read again, especially for working with kind of a less experienced handler.
Paul Bunker 1:00:23
Yeah, no, that makes perfect sense.
Kayla Fratt 1:00:25
Yeah. Well, Paul, I don’t want to take up too much more of your time. I mean, I would love to, but we should both get back to our days. Is there anything else you wanted to circle back to or explore or expand upon or explore within kind of this broader topic?
Paul Bunker 1:00:39
No, I think, you know, we’ve covered a lot of information. And I think we’ve stressed the importance of people take into consideration either discrimination and or generalization and how important it is to plan and building that plan and execute your plan. So hopefully, the listeners got something from it. And as always, I’m open to questions if people need to discuss it further. But I know we covered a broad spectrum on discrimination and generalization.
Kayla Fratt 1:01:10
Yes, we definitely did. And I know I’m going to be trying to compile all of my notes in between this recording and the end of this discrimination mini-series to put out a bit of a q&a summary episode. But I really hope people found this useful. And if anyone has any questions, clarifying thoughts, or anything like that, I hope they feel really welcome to reach out to me or any of our guests to explore this topic, more and more deeply. So Paul, where can people find you if they’re interested in staying abreast of all of your really exciting projects.
Paul Bunker 1:01:43
Um, so now I post on Instagram, ChironK9 on Instagram. I do copy some of those posts across to the show. I’m keen on Facebook, but not so much, Instagram now is my main source. I have a website, www.chiron-k9.com. And the study paper the research papers that I spoke about, they’re posted on there as well as well as others that I’ve been involved with. And then email [email protected].
Kayla Fratt 1:02:23
Excellent. And for everyone at home. I hope that by now you all know you can find us at K9Conservationists.org, where you can find all of these episodes, AI-generated transcripts, show notes, you know, all that sort of stuff, join our courses and find our webinars again at k9conservationists.org. So we’ll be back in your ears next week. Bye!