Brown Marmorated Stink Bug: Assistance to eliminate with AI

Steve Peios:

Hello again everybody, and welcome to Detect and Protect, the Australian biosecurity podcast. This podcast series is all about sharing information on biosecurity and the difference that this makes to our everyday lives. Today, we're going to be talking about a very annoying little bug. It's called the brown marmorated stink bug. It is a pest that is known to hitch-hike on personal goods, transport, shipping containers, and imported items coming into Australia. If established in Australia, it is a very huge potential biosecurity risk, especially to crop plant industries, and also to over 300 ornamental plant species. Our plant industries would be impacted greatly if this was to come in and set itself up in our country.

Steve Peios:

We currently have strong measures in place to keep our plants BMSB free. And one of these identifying methods is an app that is being developed that can hopefully identify the BMSB and helping to stop a potential spread before it even starts.

Steve Peios:

So it is an absolutely fascinating topic today, and I'm very much looking forward to speaking to my very special guest that is joining me on the podcast. It is Dr. Alexander Schmidt-Lebuhn. He's a fantastic scientist from the CSIRO. And Dr. Schmidt-Lebuhn is at the forefront of the electronic app that is being developed to identify the brown marmorated stink bug.

Steve Peios:

Today we'll be learning more about how the app came to be and the process that it's going through in its development at the moment, and also talking about the risk that BMSB poses to Australia. So without any further ado, let's welcome our guest to the podcast. Thank you very much for joining us today, Dr. Alexander Schmidt-Lebuhn.

Alexander Schmidt-Lebuhn:

Thank you. Glad to be here. Thanks for having me here.

Steve Peios:

Not a problem, Alexander. It's absolutely great. I'm very much looking forward to this chat today about this app that is being developed and also our little friend, the brown marmorated stink bug. First of all, can you please tell us a little bit about your role at the CSIRO and also your involvement with biosecurity?

Alexander Schmidt-Lebuhn:

Yeah, thanks. Although we're talking about a pest insect today, funnily enough, my background is actually that of a plant systematist, taxonomist, and phylogeneticist. And the way I came at this is because in my role as a taxonomist, there's two sides to my work. The first is to name plants and understand how they're related and should be classified. And then the flip side of that is producing identification tools that allow any kind of end user who has a specimen they need to put a name on it, to identify that specimen, figure out where it belongs in the classification, and what name the specimen should have.

Alexander Schmidt-Lebuhn:

And one of the areas where identification is absolutely critical is biosecurity, because we need to know if this is a pest or a weed of concern, or whether it's a native species that we don't need to destroy, or don't want to destroy, et cetera, et cetera.

Alexander Schmidt-Lebuhn:

And so a few years ago, 2016/17, coming from my perspective as a botanist, I developed a wheat seed identification key for the Department of Agriculture at the time. And after that, we then started looking into to what degree artificial intelligence image classification can already be useful to make more user-friendly identification tools where the end users don't need to know technical terminology. It's like basically having a taxonomist looking over your shoulder and recognising something for you.

Alexander Schmidt-Lebuhn:

Again, we started doing that with wheat seeds, but talking to the department, we then soon learned that a much greater imminent problem was the brown marmorated stink bug. And so then I teamed up with my colleagues at the Australian National Insect Collection and CSIRO to move into that area. And that is where we arrive at today.

Steve Peios:

I love that. I must say that's very interesting to me because being able to have, like you said, that presence of somebody over your shoulder, as such as we're saying, it must make that identification process really... Well, it's helpful. I wouldn't say easy, but it makes it easier for the end user.

Steve Peios:

Just quickly on that. How extensive is that process when it comes to actually identifying, be that a seed or a bug or something like that, because that's quite fascinating to me. I've always wondered who puts all the fantastic names against a plant's name or a specie. How intensive is that process, Alexander?

Alexander Schmidt-Lebuhn:

Yeah. Well, with traditional methods, again, it does take a lot of training and a lot of work because not only do you have to be able to understand the terminology to describe to your identification tool what you're seeing there, basically, so again, I'm coming from a botanical perspective, you have to know what a barbellate pappus is or a carpopodium or something like that to describe my daisy seeds. And there's-

Steve Peios:

Oh, that's easy. Yeah, of course.

Alexander Schmidt-Lebuhn:

According terminology in the bugs or in the beetles or in moths or whatever groups you are concerned with. And only then can you actually answer the questions in that identification tool. Then the really old ones, the traditional ones that taxonomists have produced for decades, they are dichotomous. So another complicating factor is that you have to answer all the questions in a certain order. And if the first question is about something that has broken off in your specimen, for example, then you need to go both ways. And it soon becomes very tedious and complicated.

Alexander Schmidt-Lebuhn:

So having a more intuitive image recognition system that is like a taxonomist that really knows their group and just looks at something, goes, "Oh, I think I know what this is. Let's check that against an example image, for example," that is kind of a much easier approach. With a caveat, of course, that it can only work in a group where you actually have visual characteristics that allow you to distinguish. Like if it is a group of beetles where you have to dissect the genitalia to be able to identify them, well, just pointing an app at it won't do the trick, of course.

Steve Peios:

Absolutely. No, that's fascinating. And I think what I'm gathering from that as well, Alexander, is that there's a process now of better record keeping as well, when you're talking about photos, when you're talking about the pathways that you're going down, the better that we're keeping those records of what we've gone down for certain species. We can easily reference that as we move forward. Again, perhaps easy is not the word, but it makes that tedious process a little bit more simple as we move on. With regards to our little friend, the BMSB, what can you tell us about it and its risk to Australian plant life please?

Alexander Schmidt-Lebuhn:

So the brown marmorated stink bug itself is part of a group of thousands of stink bugs, of Pentatomids worldwide. They are sap-sucking herbivores, and the brown marmorated stink bug itself, it has got two key problems. First of all, it is a very generalist herbivore. So it can damage lots of different plant species, as opposed to being specialised to very few close-related plant species. There's a lot of different agriculture and horticultural sectors, therefore, that might be affected, in particular fruit and nut trees. So it could be anything from apples and pears across stone fruit, hazelnuts to grapes, for example.

Alexander Schmidt-Lebuhn:

The second problem is its ecology makes it a particularly successful invader, because a lot of species they might be carried in with a cargo shipment or in the luggage of a tourist, for example, but then there's a single individual that has arrived in a new country. And generally they don't establish because they don't find a mate. They just die out unless they're already pregnant with eggs or something like that.

Alexander Schmidt-Lebuhn:

Whereas the brown marmorated stink bug, they use pheromones to attract each other in larger groups. And then together they seek a place where they can hibernate, where they are somewhat insulated from the cold. That might be, for example, the hollow of some kind of shipment, for example of pottery, or a crate or something like that, that they identify as a somewhat protected area. Then they get moved to a new country and out comes an entire group of stink bugs of the same species. So immediately, they have got a little population. They can mate, they can produce offspring.

Steve Peios:

Wow. So that's fascinating to me Alexander, in that, speaking to everybody listening to this podcast, what we're saying here is that it's... Would it be fair to say it's smart enough that it knows to sort of bring a troop with it when it goes places? It's in its ecology to go, right, let's get a group of us together before we actually sort of... Not that they're prepared to start a migration, but when you're talking about that pheromone attraction and that sort of thing, it's almost like this is the way we're going to come together because that would assist it in terms of getting it established somewhere, is needing that mate. Is that right? Effectively, it needs that mate and you need that ability to procreate to actually start a colony or whatever it is, or we call it an outbreak, but for them to start a colony or whatever the case may be.

Alexander Schmidt-Lebuhn:

Yeah. Well, smart is, of course, a bit of a loaded word. The behaviour is presumably-

Steve Peios:

Absolutely.

Alexander Schmidt-Lebuhn:

... all just instinctual. But it kind of touches upon a very interesting question in invasion biology. And note that I'm not really an invasion biologist, so you would have to talk to a specialist in that field really. But it's been a long-going controversy and question. Are there actually characteristics that predispose a plant or an animal for being a good invader, and would that allow us to find out some that are a risk factor before they even come into the country or before they even leave their indigenous areas?

Alexander Schmidt-Lebuhn:

So that's a very interesting question. And certainly things like being able to reproduce if there's only one of you in a new area is, for example, an immediate advantage, even for a plant. If you just get a single seed in an area and it can reproduce without finding mates. The same for insects. I can only say that seems like an obvious factor, but it's not really my core area of expertise to speculate on others.

Steve Peios:

Well, I must say, your speculation in other areas is fascinating me. You're a very knowledgeable man. When we talk about the BMSB, what can you tell us about the differences when we talk about the Australian native stink bugs? I know I've seen native stink bugs here in Australia before, but the actual brown marmorated stink bug. You've talked about it a little bit there, but then there's also concepts that have to do with their enemies as well and that sort of thing. What can you tell us about that, please?

Alexander Schmidt-Lebuhn:

Yeah. So I've already touched upon one of the main problems with it. That is that it's very generalist, that it will affect a lot of plants, whereas many native species are specialised. There is the morphological aspect, but precisely because it is not so easy to immediately recognise them for the non-specialist, we need identification tools.

Alexander Schmidt-Lebuhn:

And then what you are referring to is the enemy release hypothesis. One of the key problems, and that's not just for the brown marmorated stink bug, that's for a lot of invasive species.

Steve Peios:

For many, yeah.

Alexander Schmidt-Lebuhn:

Yeah. Is that if they are brought to a new naive area, they don't have any natural enemies. So there will not be any parasites or diseases or something like that in Australia that keeps the population in check. That is the hypothesis why invasive species, be they insect or weeds, are often so overwhelming, because nothing controls them.

Steve Peios:

Yeah. Understood. So with that then, and I know you've answered this somewhat previously before, but expanding on the way that they could arrive, it's a bug that's originally home in Asia, is that correct?

Alexander Schmidt-Lebuhn:

Yeah.

Steve Peios:

It's made itself endemic to other areas as well now, when we talk about north America, Europe, and that sort of thing, but that's where it originated. So we know the way that it comes together, it can move on cargo and imported goods and that side of things. I guess my question is more so, how difficult or easy would it be for it to become apparent in Australia, if we got in a position where there was enough of them that managed to hitch-hike their way and make their way here? That establishment process, how would that work? And then in terms of its actual ability to come out of that hibernation and start that process, if we didn't get onto it early enough to discover and destroy it, how would that work for us?

Alexander Schmidt-Lebuhn:

Yeah, well, I should hasten to add that the key use for this app that we're talking about today would be actually at the border. So that would be at the frontline where you try to even discover it before it ever establishes. So you're just trying to keep it out.

Alexander Schmidt-Lebuhn:

The second level of defence will then be, if it starts establishing itself, somewhere behind the border, you would want to eradicate it. And if you have a pest that is so established that you have no chance of eradication anymore, then you get to the containment phase where you're just trying to keep it contained in a certain area and not move to another state, for example.

Alexander Schmidt-Lebuhn:

And then finally, if it has actually become endemic in a country, then you would have to manage it, for example, with biological control, by trying to bring any parasites into the country under very controlled conditions that would manage it.

Alexander Schmidt-Lebuhn:

So in the last few years, we actually had a case where the brown marmorated stink bug was discovered trying to establish itself in the Sydney area, in horticultural companies and so on, and the Department of Agriculture made an enormous effort and was apparently able to eradicate the species again. So that has happened. And that is also, of course, what brings a sense of urgency to trying to get the species at the border already. Because it's always easier, cheaper and safer the earlier in this progression, from keeping it out, eradicating, containing and managing it. The earlier you get rid of it, the better, the more efficient.

Steve Peios:

And I know we talk about that in our department as a whole, when it comes to the work that DAWE does, and working with agencies like yourself, is that primary prevention is always the key. The fact if we can get on top of things early, prevent them before they come to our borders, do things offshore, treatments offshore, measures. I know that with the BMSB measures that we've had in place over the last few years, we've had different phased measures about the way that we've actually implemented different things to make sure that industry are on board as well, and everybody that's involved in this huge business that we call primary industries of import and export, that's such a massive part of that.

Steve Peios:

Just a quick question before I move on. I wanted to ask a quick one about, if it was to get established, and of course, what you can tell us about this, but if it was to get established in a set area, in a set state, let's say, New South Wales, how easy or difficult is it for it to then get to another state? Is it just a process then of it sort of needing to hitch-hike its way there again, or if it actually sets itself up, can it actually move? I know I've seen pictures of BMSB before where it's on the side of a house in a country where it's endemic and there'd have to be hundreds of thousands of them in one certain area. Is it just as simple as they can just get up and fly away and move areas, or is it a little bit more complicated than that?

Alexander Schmidt-Lebuhn:

Yeah, well, ideal you talk to an invasion biologist who study the species more, how it has spread around America, for example. But certainly it is a flight capable insect. So certainly it would move certain distances out of its own accord, and maybe with a bit of help of strong winds being blown even further than it would by itself.

Alexander Schmidt-Lebuhn:

But I think again, another key risk would be that it just hitch-hikes on cargo that's being carried around the country. Once you've got it in Australia, I think it will be much more difficult because you have less controls between states than you would have at the border to Australia. So again, it comes back to, let's try to keep it out of the country in the first place and then it will all be easier. Yeah.

Steve Peios:

That's always the key. I'm going to mark that down with my producers. An invasion biologist is what we're going to look at, at some point. That all sounds fantastic. And thanks for the idea, doctor. That's much appreciated.

Steve Peios:

You talked before about the background behind creating and identifying the app for a BMSB. You talked about the process of seed identification and having that toolkit behind you of that assistance over your shoulder. Can you expand on that in any way, and can you tell us about how it actually came to be with the department? I know that our department has a big focus on innovation moving forward, and that's been sped up with the pandemic over the last two years. So how did it come to be in that respect when it comes to funding and partnering with our department?

Alexander Schmidt-Lebuhn:

Yeah. I should start then with the genesis of the app. Again, that is slightly related to what I said earlier. But the problem that I saw with the identification key I made for the weeds, it was, again, the end user needs to understand the terminology, at least to a degree, even if I try to make it as clear as possible. And the more species you put in there, the more overwhelming the possibilities and the amount of questions become. So then we thought, well, can we use it? Can we use image classification via artificial intelligence to help us there? And because I'd just come out of weed seeds, we tried a weed seed model. We teamed up, that was in 2018, with Microsoft and co-invested in making the app prototype, based on our understanding at the moment of what the needs of the department were.

Alexander Schmidt-Lebuhn:

Obviously this hadn't been adopted by the department, anything yet, but just the conversations that I had, I kind of understood were, this is the kind of functionality that might be really useful to people who I've just worked with on this identification key. So Microsoft put in money to get the app coded by a company called Altus. And we photographed seeds, we trained the AI model and so on and so on. We put that together, and this is still, by the way, the most convenient way for me to demonstrate this potential solution to people, because those weed seeds, you can just carry them around in a little vial. They're very stable. You wouldn't do that with a-

Steve Peios:

In your pocket.

Alexander Schmidt-Lebuhn:

Yeah. You wouldn't do that with a stink bug model. You wouldn't fly to Victoria, for example, from here, and then half the legs have broken off potentially, or something like that. So it's really convenient just to pull the seeds out and demonstrate that.

Alexander Schmidt-Lebuhn:

So then we had conversations with the Department of Agriculture, Water and the Environment, in particular with the Australian Chief Plant Protection Office. And we partnered up and got funding for a feasibility study on the stink bug identification, using the exact same approach, the same image classification approach, and the same app format, through the Biosecurity Innovation Program in 2019/20. So we're now on the second round currently, where we have got another round from the Biosecurity Innovation Program to expand the stink bug model to around 40 species, based on a priority list from DAWE, so that it comes to the point where it'll actually really be useful beyond a feasibility study and can be deployed.

Alexander Schmidt-Lebuhn:

At the same time, we are exploring Hymenopterans, so non-European honey bees and the bumblebee that has been established in Tasmania, for example, to see how that works. Not that I've got any doubts about the bumblebee, but for example, about distinguishing different honeybee species. That's a bit trickier. Again at the same time, DAWE is currently doing end-user testing. So I think nearly 50 people have got the app prototype with the stink bug model on their devices now, and they are providing feedback about what they like about the functionality, how it handles, what they would like to see changed or added, in the case that the department decides to either adopt this and build on the prototype, or re-implement the functionality in a new code base.

Steve Peios:

Is it fair for me to assume as well, that there's a quite a diverse range of people that are using it in our department? When you're talking 40 to 50 people, there'd be different people that are on the ground in terms of our inspectors, our assessment staff, in terms of usability. So it'd be a pretty diverse group of people that would be testing it. Is that fair to say?

Alexander Schmidt-Lebuhn:

Yeah, that's my understanding. I obviously haven't drawn up the list. That was Teams and DAWE, but my understanding is that it is both DAWE inspectors and people from industry partners under approved arrangements who are considered here, and then also entomologists endorsed. So the kind of people who would've been sent specimens for identification.

Steve Peios:

Fantastic. Fantastic. All right, Alexander, the big, serious questions now for you. How will the app work? We've talked about people that would potentially be using it, and of course that comes down to the app being operationalized and that side of things and getting its final approval stages. But how will the apps specifically work? You mentioned a little bit of the process before with the seeds, but in terms of this, for the stink bug, how are we looking to use this app?

Alexander Schmidt-Lebuhn:

So yeah, again, the department may obviously find additional functionality that it wants added before it gets deployed, but I can describe what the prototype is doing at the moment. I can also say that it's something that I wish I had in some circumstances and other circumstances for myself.

Alexander Schmidt-Lebuhn:

So what the app prototype does at the moment is that the main interface is the identification interface. So you point the camera of your smartphone at your specimen, let's say a stink bug, and it uses the live video feed to constantly give you an updated identification estimate. So it's not a static image that you once ping it and you get one answer. It's an interactive experience where you can go with your camera around the bug. You can zoom in, you can zoom out, you can make sure that you've got the focus right. You can try the other side of the insect, and constantly the identification estimate at the bottom changes.

Alexander Schmidt-Lebuhn:

You've got a tiny little thumbnail next to it, so you can already see, this one looks like it might be a good one. And you can click on the name that's in the list of suggestions down there, and a species profile with a larger example image pops up and allows you to compare. And that's actually really important to me, that the end user doesn't blindly trust the model, but that they can second guess what it's being shown. So they can look at that example image then and say, "Oh, actually I don't think this is right. So maybe I'll try the other side of the insect. Or maybe I need to search for better light conditions." And when they found the one that matches, they can actually gain the confidence. "Yeah, that looks exactly like the one that I'm seeing here."

Steve Peios:

I love this. This is great.

Alexander Schmidt-Lebuhn:

Then you can go back to the identification interface and there's a button there, it's labelled save at the moment. You just press that, and an observation record is created that has whatever the camera sees at the second where you press the button. So you get a photo of the bug. It gets the coordinates where you've taken the photo. Potentially, it will also even give you the address, depending on whether it can figure that out. The moment, the date and time where you took the photo becomes part of the file name.

Alexander Schmidt-Lebuhn:

Then finally, obviously it saves the identification estimate from the model with the name and the confidence that it had, all into that record. Now you've got all these records then on the phone and the one functionality that isn't implemented yet, of course, there's a little button then on the record display that would theoretically allow you to upload that to some kind of DAWE database or web service.

Steve Peios:

Some sort of platform that you can actually go in for the recognition.

Alexander Schmidt-Lebuhn:

Exactly. This is the thing. You could then forward certain ones that you think are critical, you could forward them for reporting for second examination by an entomologist by the department. Or you could use this for example, and this is one thing that was very important in the discussions I had around the weed seed identification key, to collect data for the department to analyse, to understand, to do pathway analysis, for example. So even with things that the model cancelled out, you could then say, "Oh, we're suddenly starting to get this particular thing that looks like this. We don't know what it is. We're starting to get this coming in from that continent. What's going on here?"

Steve Peios:

That's trends. That's a big part of scientific research. Is that fair to say? Trends and looking at certain things that are coming here? It may not be in the specific field you've just looked at or what we are talking about here, but when you're seeing those trends, that must be invaluable for yourself.

Alexander Schmidt-Lebuhn:

Well, but at the moment, I'm primarily thinking, of course, for what you mentioned earlier about then potentially going back as the department, going back to importers and say, "We're suddenly starting to get this from this area. We need to work together with you to look into why that has changed, what is happening." There could be any kinds of things behind it. Like suddenly these containers are standing around for a long time in a certain place that hasn't been cleaned up or something like that. And then you can go even further back than the border, at least that's my understanding from those discussions, and try to make it so that things don't even arrive in Australia and don't have to be cleaned up extensively, because if you've brought your defences out even further away from the border.

Steve Peios:

Yeah. Understood. We've just been speaking then about DAWE and the fact that we've got our offices and staff members doing it. What about the public? Is there an intention to progress the concept into the public to be able to use the app and report and that sort of things?

Alexander Schmidt-Lebuhn:

Yeah. In the first instance, currently, the app is being tested by DAWE staff, but that has certainly been a discussion. So whether then ultimately, a year or two later, another version of the app gets made available to the public with enough functionality in there that it will help them detect things behind the border, that is a consideration for the midterm future.

Alexander Schmidt-Lebuhn:

There's a few things to be considered there. And this also goes back to the question, why can't we just use a publicly available identification apps that are already out there, like iNaturalist and so on? The question is then what happens with the observation records that there might be confidentiality issues in there and so on and so on. So that is something, I guess, then for DAWE to really think about, what exactly do we want to give to people, why. But again... Sorry, I'm kind of mixing two things together here. There's first of-

Steve Peios:

No, that's okay. That's what we want to hear.

Alexander Schmidt-Lebuhn:

Yeah. There first of all, that is a reason why you have to think carefully about how exactly do it. And second, the question is then, are you competing with the kind of citizen science apps that are already out there? That is an important question, because of course, there's no sense in having 20 apps competing and then none of them can have a critical user base. But the question is really that if, for example, you find a reportable weed or a pest insect that needs to be eradicated, that is not necessarily the thing that perhaps everybody who reports that wants to see in a publicly available citizen science database. So having this niche of you helping the department, might be a different application than having the publicly available nature enthusiast citizen science app. So yeah, it goes both ways, in a sense.

Steve Peios:

Understood, understood. Is it fair to say that with the more apps that are out there and the more this widespread functionality, does that lead to a lack of consistency as well? Like you said, you can't have a critical user base if it's so spread out, but also if there's an inconsistency in what's being offered to the person, does that cause a problem as well?

Alexander Schmidt-Lebuhn:

Yeah, there's two questions there, basically. I think there is obviously value in having as large an app and as large a user basis to for any specific use case. But I am also very strongly opposed to this idea that there is one app already, so we don't need another one. Because I really do think there are very different use case. It makes a lot of difference whether, for example, you've got a really broad one that is for people having fun on their bush walks, but it's very opportunistic in what kind of image is being taken by people, and they take the same frequently observed species in their garden over and over. They take lots of photos of those. Whereas very specialised products, for example, biosecurity sector, where you really want to be sure these are photos of-

Steve Peios:

That what you're identifying.

Alexander Schmidt-Lebuhn:

Exactly. These are photos of expertly identified specimens. This has been signed off by the expert for being really accurate in what we want to do. This is reducing the likelihood that we're making mistakes, and the data gets collected to the place where we actually need it to do our risk assessments, et cetera, et cetera. So I think just like there were many different identification keys, there will also be different specialised, modern identification tools using artificial intelligence in the future. That's precisely what I'm trying to say.

Alexander Schmidt-Lebuhn:

The second thing about the consistency. So yes, that is a problem that some of the, I don't want to mention any names obviously, but I've tried some of the existing identification apps that are out there and some of them do not provide to me a really good user experience. They're all a bit different, and I would really like to see something that informs you of the confidence that you have so that you can really interactively see how confident you can be or how the identification changes. And that allows you to second guess by confirming against species profile and an example image, or potentially several, as opposed to just you point it at something that has a name and you take it or leave it. That's fun for walking around with your kids on a bush walk or whatever, but it's not necessarily what a professional end user needs.

Steve Peios:

Yeah. Understood. And that's the importance of, as you mentioned before, in the sector that we're working with here in particular, when it comes to biosecurity, it's very, very important that the identification processes are very accurate.

Steve Peios:

Alexander, my last question I'd like to talk to you about is artificial intelligence as a whole, the way that it's benefiting biosecurity here as well. You've talked about the work that you're expanding into with non-European honeybees and that sort of thing. What I'd like to just get gauge from you is, over the next few years, how important do you see AI in terms of benefiting the biosecurity sector and many sectors, I'm sure, that the CSIRO works with, and the challenge that comes with that? How important is AI? Because I've been fascinated to hear the process of the app and the way that it works, but also the broad aspects of its application when we're talking here about biosecurity, specifically other apps that look at things in the garden. But AI as a whole though, when you can just... I guess the concept of what I'm trying to explain here and thinking for our listeners is that you can just hold your phone up against something and work out what it is. It's an AI that's developing a long way from when I was a kid, for example.

Alexander Schmidt-Lebuhn:

Yeah. So I think the key problem here is that you always have to manage expectations. There are the extremes of some people who hear and think that's magic pixie dust and will solve all our problems. And at the other end of the spectrum-

Steve Peios:

Magic beans.

Alexander Schmidt-Lebuhn:

Yeah.

Steve Peios:

I remember those. I remember those.

Alexander Schmidt-Lebuhn:

Yeah. At the other end of the spectrum, there are people who try it out once and it fails to give them the right answer. And they think, oh, that's all nonsense. That's never going to be useful. The truth of course, is somewhere the middle. There are really cases, and that I see that in the projects I've done so far, where it's really quite amazing how it can distinguish species. And sometimes even where I'm not really sure how it does it, where it must see something that I can't immediately put my finger on, but it works.

Alexander Schmidt-Lebuhn:

On the other hand, there will be cases, as I mentioned before, if you've got a group of little, grey beetles, they all look the same until you take apart the male genitals. Then, well, that is not the use case. So it will always be a part of the toolbox. It'll always be just something that will help us to triage, because in the end, there are critical cases where then you want the taxonomist to have a look at it. You just don't want to overwhelm the taxonomist with 200 other cases that the AI could just sort out for you and the inspector already knows what's going on.

Alexander Schmidt-Lebuhn:

But it's always going to be part of the toolbox together with things like genetic sequencing and human expertise, and I don't know, pheromones, for example. There are so many different innovative methods that are currently coming out, also funded by the Biosecurity Innovation fund.

Alexander Schmidt-Lebuhn:

In general, to answer the other part of your question there, I can mostly speak for image analysis, like image classification, object detection. I can't really speak to all the other users of artificial intelligence, like sound detection, for example, and so on. But yes, in my broader field, artificial intelligence is becoming more and more important. And so I'm also for example, part of a CSIRO internal project in our biological collections, again, that I'm coming from, National Research Collections Australia, where we've got a postdoc who's working on extracting trait information from our image specimen collections using artificial intelligence. So that is then part of our collection science research and evolutionary biology and so on. So this is going to become a big part of our toolbox. But still again, one of many, many tools. It's not magic pixie dust.

Steve Peios:

Fantastic. All right, well, I'll put the magic bean request aside then for the time being, Alexander, and we'll wait for this to develop. But look, that's absolutely magnificent. Thank you very much for that explanation as well. I look forward to seeing how AI develops and more than anything, I'd like to say a big thank you to you for joining us today. That's a very interesting insight into the work that's been doing to develop these apps for identification purposes. And also the great work that all of you and your fellow scientists do to keep our country safe and work in innovative ways to help us. So on behalf of everybody here at Detect and Protect, and all of our listeners, I'd like to say big thank you to you for joining us today.

Alexander Schmidt-Lebuhn:

Thank you. It was truly a pleasure.

Steve Peios:

Excellent. Thank you very much. All of our listeners that was Dr. Alexander Schmidt-Lebuhn. Great insights today into the new technology, and we are hoping that it will become operational very soon. We know that these things take some time to get developed, but it's all heading in the right direction.

Steve Peios:

Primary prevention, of course, another one of our key messages today that we pass on in the Detect and Protect podcast. And hopefully in future, everybody can have one of these apps on their phone and look out for things when it comes to biosecurity.

Steve Peios:

Thanks very much everybody for tuning into our podcast again, this time around. The series is really taking off and we've had some great guests on it. And I look forward to our next guests in the future podcast. You'll be able to find out more information on Australian biosecurity on the department's website, or also by visiting biosecurity.gov.au. We'll make some links available in the episode description. And fingers crossed, in future, keep an eye on the website. We'll be able to provide some information about how you can hopefully download the app if we get it to that point in future. But it's all going the right way.

Steve Peios:

Make sure you subscribe to our podcast series to get updates on future topics and learn more about biosecurity. My name is Steve Peios. I've been your host again, and please stay tuned for the next episode of Detect and Protect.