Modelling Algorithms From Ant Behaviours

Kathy Nickels interviews Professor Dan Angus, from the ARC Centre of Excellence for Automated Decision-Making and Society.

Kathy: Welcome to the Automated Decision-Making and Society podcast, my name is Kathy Nickels and today we are discussing what an algorithm is and how some of them are developed. Joining me in this episode is Associate Professor Dan Angus from the ARC Centre of Excellence for Automated Decision-Making and Society. Welcome to the podcast Dan! Can you tell us about your research background?

Dan: Sure thing, so I’m a computer science who’s drifted slowly into humanities and social science during my career. So, I started off doing engineering and science at Swinburne Uni and then I did a PhD in artificial intelligence in a very obscure area called Biologically inspired computation. We modelled the behaviour of ants and other kinds of swarming insects to make algorithms that would be able to solve problems. And then I moved up to Queensland where I worked for a decade at the University of Queensland. And there I was developing natural language processing algorithms and information visualization techniques, particularly for use on social media.

Data on conversational data on really kind of any kind of text data. You can get your hands on and I started to find, I guess some purchase for this within, then the School of Journalism Communication. It started up many collaborations and eventually kind of fully drifted across into that then school, which later became the School of Communication Arts and then a few years ago I decided to make the jump across to QUT and I now lead a program called computational communication and Culture, which is in the DMRC, and within that, we explore all those kinds of questions around both computational methods that could be used to study social systems, but also look at the impact I guess of technology on society. And so, bringing in understanding our guess that bridges across both the very technical like how algorithms work and their kind of role and function together with their impacts and the ways they shaped society. And we of course how society shaped those technologies in return.

Kathy: So you mentioned in your bio that you studied ants to work out algorithms – it sounds fascinating, can you tell us more about that?

Dan: Yeah, so so we have to understand. Like when we talk about algorithms, I mean people have their own understanding about what algorithms are and the simplest way to understand an algorithm is as a sequence of instructions or such, it’s like a recipe almost that the computer can kind of move through a sequence of instructions to achieve some kind of output. So, you give it some data or it does something with it and then spits some output the other end.

I’m there’s a whole branch of computer science dedicated itself to looking into nature I guess and trying to mimic you know, processes of like evolutionary processes or other kinds of natural processes around and mimic qualities of that in order to solve engineering problems or even you know, just create, I guess, interesting philosophical inquiry about the nature of what is intelligence or these kinds of questions. So, you think about one of the fundamental technologies that drive a lot of the systems that we’re interested in from the perspective of ADMS, it this technology of neural networks, right? So people might have heard about deep neural Nets or generalized now to kind of just machine learning. Usually when people say machine learning, their meaning deep neural networks or similar systems.

That kind of that -that inquiry dates back to the 50s, where the kind of early connectionless movement was interested in the human brain and the mammalian brain I should say really and how that brain was composed of these neurons and there was connections and they were configured into these networks and look to try and mimic that structure. And it’s really interesting when you just look at that one branch there, there’s a schism in terms of those who tried to model biological fidelity really, really closely. So, tried to create and mimic the exact functioning of the brain, so even creating different structures depending on like how mammalian brain might work, so things closely resembling like the hippocampus or subsections of the hippocampus which is like a really kind of primal part of the mammalian brain and then out to other kinds of regions around. And thinking about how do we build an actual mimicked version of a brain?

To others that were just kind of easily swayed by the idea of it right, and use that to inspire certain kind of mathematical algorithms and ways of processing information, and that’s closer to what the deep neural networks of today are. They don’t really bear much resemblance to the way, a brain works but they kind of loosely inspired by that idea, and that’s very similar to the Swarm intelligence algorithms I did in my PhD. These are things like yeah, looking at how like an Ant colony or specific subspecies of Ant use pheromone markers to help other ants find their way to food and back to nests. And the way that these simple kinds of rules give rise to really complex emerging phenomena.

If you’ve seen videos of starlings kind of flocking, you’ll know what I’m talking about. Where it doesn’t seem to be that there’s like one bird that’s centrally controlling the whole flock, but yet they just kind of move together. It’s like they’re almost some sixth sense at work where they can all turn and move around together in perfect harmony. And it’s these kinds of actually, there’s very simple rules that play there that govern that movement. But when you look back from afar, it’s this emergence of what looks like a very intelligent, coordinated, dynamic at play and so yeah, it’s a fascinating field of study that at its heart, because what you’re talking about is observing that interplay of nature, right? That emergence and complexity of nature. But many of these systems, when I say complex at heart, they’re rooted in what a very, very simple rules that when you kind of layer with enough birds, ants, whatever it might be, give rise to this beautiful harmony. So yeah, I was really taken in by this as an undergrad, and that’s why I ended up I guess doing a PhD in it, but yeah, the algorithms we developed we developed the variant of one of these Ant colony algorithms and used it to apply to things like creating and scheduling kind of vehicles around like career vehicles around a city. It’s one of the algorithms is actually used to map out a water distribution network in South Australia. So anywhere where there’s like routing and scheduling, and those kinds of what we call combinatorial optimization problems where it’s about sequencing things in order.

Is you know, like what order would you put shipping containers on a ship if it has to go to multiple ports, you don’t have to take every container off to access the one at the bottom or if you’re going on a camping trip, you want to make sure you put your tents kind of towards the back of the car and not bury it under all the other bags, particularly if you can arrive. In the rain, these kinds of things, so these are like what are considered optimization problems, and they’re at the heart of a lot of like the applications of AI. But yeah, perhaps we don’t see them right? These systems are pretty much everywhere you use, like a mapping tool like Google Maps to route yourself from A to B. There’s an algorithm called A-Star which governs you know, and it’s one of the simplest algorithms, but they’re actually quite powerful as well that will route you from kind of A to be quite effectively, and these things are kind of underneath all forms of consumer technology and everywhere. But yeah, we don’t really see them necessarily or talk about them.

Kathy: Yeah, I had this similar conversation with one of our data scientists today and he was saying the more we are trying to model computers to behave like humans – The more we’re learning about humans and, how our brain works and how nature works in a way.

Dan: So yeah, so the study of this when you start looking at these systems. So, it’s really interesting the early work in the Ant colony space was very much rooted in the biology right where there was biologists like Denny Borg and others who were looking at ants and their behavior and kind of modeling this, and it led to work that where they created this famous experiment called the Double Bridge experiment where they actually they grabbed a physical ants nest and lay some food like sugary kind of liquid. And they created a tunnel system that the ants could traverse to obtain the food. And then what they did was they started varying the length of the pipes connecting them.

Now even more locally, some of the first work I was involved in when I moved up here to Queensland over at the Queensland Brain Institute, there’s a researcher there Srinivasan, who’s one of Australia’s leading professors in bees. Right now, there sending bees, but Srini was interested in adapting those kinds of the ideas and the way that bees work, right since the world and make decisions as such into, I don’t think that for unmanned autonomous vehicles, right? So can you create like a bee brain to fly a plane.  And they did exactly that, you know, using principles from the ways that bees navigate. So bees sense polarized light and they did experiments. They’re on actual bees to have them navigate through tunnels, change, change the grading on the side of the tunnels into certain kind of patterns that would trigger certain kind of polarized response and then use that to prove that yes, you know bees are sensing polarized light, but then went so far as worked with the engineers to adapt those basic biological mechanisms to try and embed that as advances in technology.

So, you could create like a self-righting plane or something like that based on, you know, certain kinds of advanced sensors. Uh, another example of this was a new camera that we’re playing around with, which was based on the actual, you know, the biological eye. Our eyes don’t sense the world in frames, so photography is based on, and our videos are based on a frame-based logic where you’re taking a complete scene of information. You take one picture. You take another picture and our eyes kind of do the work of blending the scenes together. Our eyes though, work in a very different way in that they sense change. So, when we’re sensing, uh, seen, our eyes are kind of always moving, and what they’re looking for is scanning a room for movement and change in kind of a background scene, you think about the biological beginnings of this right? If you’re down by a water hole you wanna sense the lion kind of stalking you from behind so you don’t get eaten and so there’s a whole new kind of camera out there which is developed around this principle.

So rather than sensing an entire scene at once, it’s got a bunch of sensor array that is just there to pick up change and if you connect this to an actuator, like an eye that moves back and forward, it’s a very informational like informationally efficient way to create some kind of an eye in say a robot because you’re not having to deal with massive detail in huge frames, it’s kind of automatically getting you the most relevant information about where edges and such, of objects might exist because it’s based on this idea of change in the scene. So yeah, it’s these kinds of ideas, and I’m fascinated by it about translation of what I guess you know millions of years of evolution have kind of delivered us in learning from that to adapt into our own processes to do anything from help design better drugs to be able to create more efficient transport networks through to help organize the world’s information. You know whatever it might be, I think that’s the promise of it. But I guess with ADMS we’re also interested in when those things can do potential harm, right? So, where those systems, those I think systems of beauty right maladapted, right?

And that for me I think is the personal stake here with the admin is that as someone who really and kind of really drank the Kool aid earlier. I think as a young undergrad when I started learning about this and thinking, yeah, this is amazing I want to dedicate my life’s work to this ’cause it’s so cool and then later on because I made the jump to humanities, seeing the other side and kind of hearing from the others about the harms and the  ways, it’s not being done well that it’s creating power imbalances and wealth inequality and it’s harming our environment and  these, and I think no we’ve got to do better because I think we need to truly embrace you. Know that story of that. These things have come from a long line of evolution, and we need to tread mindfully through that.

Kathy: So it turns out that nature has inspired the design of many algorithms and mathematical solutions to complex problems. Thank you for talking with us today Dan.

Dan: That’s fine, anytime.

Kathy: You’ve been listening to a podcast from the ARC Centre of Excellence for Automated Decision-Making and Society. For more information on the Centre go to www.admscentre.org.au