 Hi everybody. I'm Paul Klerkin. I'm a deep-sea shark taxonomist. I've traveled in the Pacific, the Atlantic, and the Indian Ocean studying deep-sea sharks and the title of my talk is A Picture's Worth a Thousand Words because I think despite all the work I've done the most important thing we can do is to get observers and factory workers to document sharks with cameras and start a pipeline of data from the deep-sea to researchers and policymakers. So I've worked on a bunch of boats and this is pretty much what you're dealing with. You know, whether it's a long line or a trawler you have a bunch of fish with sharks mixed in there. So you pick them off sometimes on a sorting table or belt and then you have to go through all these sharks. There are actually even more sharks in the factory and this came up in a single toe and this took me 18 hours to sort through them, identify them, and take the biological data and that's too much to ask from an observer. And this is reflected in some of the data that we get. When I've double checked observers work they can misidentify up to two-thirds of all the sharks. And this is pretty understandable. Shark taxonomy is hard. You can see right here this is actually all one species even though it looks different and has four color morphs. And this is four different species. So this by itself is already pretty confusing. These are a bunch of lantern sharks and they're all different species. And so you can see it's hard to tell apart plus most of these are new species. So even if you could tell them apart you wouldn't even know what to call them. And this is pretty common. Here are some sharks that came up in the same toe and it's six different species they look alike and they're all new to science. So what have we done in the past? We've tried to give observers guides and it's kind of hard. Taxonomy is difficult and that's because taxonomy is sometimes outdated and a lot of the manuals are written by taxonomists for taxonomists and they make them too verbose. So when you're in the factory there's just too much writing and it almost forces the observers to flip through them and look for pictures. And this is not good because they skip all the steps to key out the shark. Also a lot of the time taxonomy is written in terms of morphometric measurements. That's kind of the language of taxonomy. And this shark is kind of described as having a long snout and this one is described as having a longer snout than this one. And that's not very helpful unless you have both the sharks in your hand. You can measure the neck, the top of the head here and get a snout to head ratio but then you're taking multiple measurements and calculating and that's this is not going to happen in the factory. So that's why taxonomy is hard but why is it important? And that's because all the stuff that we need to estimate life history, sex ratios, length distributions, fecundity, maturity, all that stuff that we need for conservation is all species specific. So if you don't identify the shark to the right species not only is that data useless but it can actually be harmful if you try to use the wrong shark to estimate life history. So an example of this is the squalus acanthus and succlii. These sharks used to be identified as the same we just thought they were the same shark and they were managed as the same species but they've done more work and they found out that acanthus the females mature at 12 years old and they have up to 25 pups and succlii the females mature at 35 and a half years old and they have no more than 17 pups so obviously these species have to be managed separately. So our system is obviously currently flawed even though we have observers going out on some deep sea fishing vessels not enough data is making it back to researchers and policymakers but what if we could send observers out with a tablet that identify the sharks for them and not only guided data collection making it faster and easier but also recorded the data and then transmitted this data so that it was readily available to researchers and stakeholders who could update the protocol as they needed new data. And we believe this is possible with a technological solution combining taxonomy genetics information management systems and AI. So we have different phases phase three right here this is the the tablet that will help us with everything but to get there we have to go through two other phases taxonomy cataloging all the species and electronic key and management system and then we can actually go past that to a shark sheet that could collect data from sharks as they're being discarded. So phase one is very basic it's it's going to be the most important phase and it's also going to be the most time consuming phase but it's actually not much different than what ships are already doing it just involves a tentative ID some kind of hall information location stuff depth and whatever life history information that the the observers are already taking sex length maturity. We are on top of that going to ask for a suite of photos which most boats are already doing and a genetic sample and the occasional specimen if it's a new species. So this is very low tech it's cheap it's going to be minimal time and also minimal training but it's going to be the first step towards cataloging the species that are in our deep sea and training the AI. So the photographs are they're going to be needed for identifying the species and also for training the AI so we're going to need a suite mostly full shots so full top full side and full bottom of the shark and some shots of the head and the dorsal fins would be helpful. So for the full side we just need the whole shark from the tip of the snap to the tail with some size references in there usually just a fish board a meter stick. These ones where the shots are not straight on are kind of complicated it's really hard to identify a shark from a picture like this. These might be helpful for training AI later on but right now we're just asking for these straight-on shots and then shots of the head side top bottom where we can see the teeth and then of the paired fins where they're both in the same picture. For these we're going to need one shark per photo a size reference a color chart and then something in the photograph to identify the the specimen usually a whole number and a vile number if there is one. We can use chief standardized cameras and use a standardized labeling system. The genetics are going to be very important for identifying the sharks and also teasing out how many are new species we anticipate there's probably a lot and it's not going to take that long we have these these genetic guns and you just kind of pull the trigger and that pushes the blade through the shark's fin and through the cap of the vile and then the blade actually pops off they're one time use it will pop off in the vile and a new cap is put on so because of this there's no sanitation or cleanup required it's going to be very fast they're also pre-labeled so you can just take a photograph of the label and that can be scanned into the data. So that's step one it's it's going to be the most important and it's also going to be very quick to start up it can be done with the distribution of some protocols and training videos and it could happen pretty quick for very cheap. Phase two is going to make species identification quicker and it's also going to manage the data faster with the information management system. So the first part is an electronic key or e-key as I've been calling it and this will give the observers a number of characters some of them with photographs that they can pick through so it'll be kind of like a key that they can just touch screen on a waterproof touchscreen tablet and this will will give them pliable answers with you know maybe's or unsure and that way that they never actually reach you know a wall where they don't know what to do that can redirect them to a definitions page or a series of pictures or just more instructions just like bag it and tag it in case it's a new species. So this will go to the end where they get a confirmation image of the shark and then they can confirm or or you know see that it's not the shark that they're looking at and then it will put in the species name into the database and then it can start collecting data and this will guide them through species specific data. So we have a shark like this which is very common this is probably the most common shark we have out there and we don't need really that much data from it we can do with just number of males and number of females. This shark is rare and on common we'd probably ask for a fin clip and bag and tag the specimen. So because they'll be spending less time on the species that we know a lot about they can spend more time collecting from the the data deficient species and we'll get more data for less time. So the data coming in will be organized it will all be you know in the same spreadsheet it will be standardized you know everyone can be using meters instead of millimeters instead of centimeters and because it's so standardized it will be easy to analyze and quickly utilized. Step three is creating the AI so this will automatically identify the shark using on-board camera on the tablet. It will fill in the species name it will automatically label the photos and it can also take basic information such as sex and length and then it would guide the observer to take further data measurements and samples. This is expandable with other wireless devices that can be bluetooth sunk to a scale or calipers or a fishboard if we want to take more data and because we'll be taking fewer images everything can be uploaded via wi-fi once it has a wi-fi connection. So this will be highly automated it will be efficient there'll be less of a chance of human error and the data will go straight to researchers and policymakers. Fish AI has been around for a while I actually have three apps downloaded on my phone. The reason it hasn't really developed for the deep sea is because there's this lack of photographs and data to go with them. So from the first two phases we'll have a lot of photographs that will have the ID anyway to be able to use that data for life history and then we can use that to educate an AI model that's already been trained to identify fish we can transfer learn these sharks. The next step would be to create a shark shoot that could identify and catalog data from sharks as they were being discarded. This could be done adjacent to the AI tablet so observers could continue to collect more high resolution data and samples and the shoot could be used for live sharks to minimize the amount of time that they spend out of the water. This photo might look familiar to Susanna Romain. She's been working with NOAA to create this electronic monitoring system. And that's pretty much it we could start this with phase one right away which is going to be low tech just taking images and genetics from commercial fishing boats. We've done this in the past with sea lord and we've got a lot of new species and it's been a successful project and that would go straight into phase two which is using the e-key on the tablet to collect more data and then phase three would be the AI tablet and we get a lot more data with a lot less time and we could take a census of sharks which we could pair with developing technologies such as environmental DNA or EDNA. All this technology already exists it just hasn't really been put together or tailored towards deep sea sharks. I'm happy to take any questions.