So when we look out in the world, we see the amazing breadth of biodiversity. And one of the things that's most inspiring about it, right, is beautiful, ornamentation, colors, extreme displays... and all that has to do with social and sexual behaviour of animals. Which means even though we're inspired by it and it's really interesting to us, it's not actually for us, which is really interesting to understand. How does it come to be? Right. But, to know, what's that big blue feather for? Or what's that spot pattern on a guppy for? We need to understand why those things are important to the individuals for whom it is being produced.

I'm Sam Snow. I’m a postdoctoral research fellow here at the Biodiversity Research Centre and in the Department zoology at UBC. In my work, I'm creating a framework which allows us to ask, in this case the Trinidadian guppies, what's important to you for mate choice decision making?

Guppies are an amazing system to work in. People have been working with guppies for a really long time actually. Ever since about 100 years ago, people realized that the males of the guppies are really variable in their sexual ornaments--their black spots, their orange spots, their blue spots. And since then, a lot has become known about how guppies make their mating decisions and under what context. But a lot of the things that we know about guppies have been, discovered within a contrived, experimental, highly controlled scenario. Now, this can tell us a lot of things, right? This is how we know that females have a preference for male guppies that are more orange. But when we create these controlled experimental setups, it obviates the possibility of us studying the social context in which these behaviours happen. Right?

Because usually guppies are swimming around in, whole group of guppies and a female has many different males to choose from. The males can interact competitively with one another. And in guppies in particular, the females can react to a male's courtship display and therefore produce sexual selection on that Male’s ornaments and courtship behaviour. Or males can also choose to attempt to coercively mate with the females and subvert her ability to choose based on colour patterns or the display behaviour, and mate with her without her consent. The influence that that sneak mating or coercive mating has on the force of evolution on the ornaments themselves is really hard to understand outside of a true social context where all of those behaviours are possible.

In my experimental setup. Basically, I have a fish tank full of water and we put in five females, five males. We just let them do what they do. Now, what makes this fish tank different from all other fish tanks is I have multiple cameras set up around it at different angles. In this case three that allow us to triangulate the position of each fish within that tank. And then we use a machine learning algorithm to find each individual fish for us and maintain its identity and track it through time, over two hours of social and mating behaviour. So for every fish, we know their identity.

So let's say Bob the fish, we know who he is and we know who he is all through time, even if he's crossing, in front of other fish and over and under, because we have this multiple camera angles, we also have information on where Bob's nose is, where Bob's head is, where Bob's tail is, where his genitals are. We're tracking all of these different pieces of Bob's body automatically using the machine learning algorithm. This information then lets us train another machine learning algorithm to extract the behaviours from the data. So we have poses. For example, you see me waving to you. You know that I'm waving hello, but how do you actually know that? You see my hand moving in a particular way relative to my face. You see my mouth. Maybe I'm smiling. Hello, hello. That's all the actual information that your eyes are taking in and processing to understand that, I'm saying hello.

We need to train the computer to understand that kind of information. So when we have a series of points in three dimensional space that are conformed in a particular way, we know that's a mating display and we can say, okay, across these frames that male is displaying to this other females, which we've also tracked, we can train an algorithm to pull out those behaviours for us. Then we can use that for downstream analysis. This kind of approach lets us take an enormous amount of data and make it actually usable in a statistical context. So what we're hoping to do is marshal all this new capacity for data collection, using machine learning techniques to really bring this kind of behavioural ecological study into the future.

I think we're in a really exciting moment right now where we can use these new artificial intelligence and machine learning tools to finally get a handle on more complex and more realistic systems that reflect more the realities of individual animals life experiences in the real world, and in turn, help us better understand the production of biodiversity and the amazing different cool colours and displays and ornaments that we see around us.