Before I started to work with bees, I was fortunate enough to work with Professor Tom Collett at the University of Sussex, one of the world’s top experts on insect cognition, and especially navigation. We studied wood ants, the largest ant species in the UK; you can often find them in sunny woodland areas where they build huge mounds out of twigs, pine needles and so on, which may contain hundreds of thousands of ants. You will often see what look like ant highways near to nests, with hundreds of ants running back and forth with food. They eat insects and other invertebrates, but their main source of food is aphids. The ants don’t kill the aphids but suck up a sweet liquid, called “honeydew” that the aphids secrete (in return for the ants protecting them from predators). As you get further from the nest, the ant highways split into branching networks of smaller paths, each with fewer ants, so that the colony can collect food across a large area surrounding the nest.

 

It’s well known that ants lay pheromone trails, allowing others to follow their nestmates. Because these trails evaporate over time, only trails that are frequently refreshed remain strong and attractive to follow, so the best sources of food are easily found, whereas routes fade away. What’s less well known is that individual ants mainly make use of these trails only when young and inexperienced. An ant that finds a reliable source of food will quickly learn to find its own way and individual wood ant foragers can learn to navigate complex routes hundreds of meters long very accurately, passing through multiple different types of terrain. They have various sources of information available to them – we often talk about a “navigational toolkit” – but, if necessary, can learn to follow these routes using their visual memory alone.

 

Along with Tom and many of our colleagues, I want to understand how an animal with such a tiny brain can accomplish such difficult feats. Partly, this is a scientist’s fascination with a cool and impressive natural phenomenon, but there are bigger reasons to want to know what’s going on underneath the surface to enable ants to do what they do. Finding your way to important locations (like food sources) and safely home again is a problem that’s common to lots of animal species, particularly humans. Understanding exactly how different types of animals approach this common problem can shine a light on the evolution of brains and the process of thinking. It’s quite common to imagine that humans and large-brained vertebrates must be doing a lot of very computationally intensive processing to support the complex, intelligent behaviour we observe, but when you see an individual ant tackling the same problems – and often doing just as well – it becomes clear that there may be far more computationally efficient ways to achieve the same ends. Which brings us on to another reason for wanting to understand insect brains: finding computationally efficient solutions to complex problems is a major goal of AI and computing research. Historically, biology has inspired many engineering advances, and as we learn more about how thinking actually works, there is every reason to believe that animal cognition can inspire advances in artificial cognition. The neural networks that underly all current AI research were themselves inspired by brain architecture, but understanding the specific computations and mechanisms behind intelligent behaviour has the potential to unlock even greater advances.

 

Over the last decade or so, the most influential idea about how insects learn routes has been a concept known as “alignment image matching”. The idea is simple: you memorise the view towards your goal from various points along the route. Later, when you try to follow the route again, you turn until your current view matches the view you memorised and then walk forward in that direction. If you stray slightly from the memorised route, it is not possible to make your current view match your memory exactly, but if you turn until you find the best fit, or the least-bad match, that will often be when you are facing in close to the right direction, so repeating this process will eventually get you back on track. Imagine, for example, that you are trying to reach a location at the base of a mountain. Even if you lose your path, turning to face the mountain will give you the view that’s closest to the one you want, and heading in that direction will bring you closer and closer to your goal. Of course, if you go too badly awry this doesn’t work – if the mountain you see in the distance is not the same one you are looking for, then moving towards it isn’t going to help. Perhaps surprisingly, though, it seems that in many natural scenes there is a fairly large “catchment area”, within which repeatedly heading towards the view that best matches your memory will eventually bring you to your goal.

 

Algorithms based on this principal have had a surprising degree of success in replicating the achievements of navigating ants and given scientists great insights into how ant brains go about learning and following a route. But surely this can’t be all there is driving the apparently complex and sophisticated navigation we see? Although quite a lot of ant navigation can be explained by alignment image matching, there are some observations that suggest there are other factors in play; for example, desert ants can follow a route when walking backwards, dragging heavy biscuit crumbs!

 

Tom, Craig Perl and I deliberately set up a situation in a simple, controlled environment, in which alignment image matching would fail, in order to find out whether and how ants could still follow a route. Instead of walking hundreds of meters through a complex forest to collect honeydew from aphids, we asked ants to walk about 60 cm over a featureless, flat table to collect a mixture of sugar and water from a microscope slide. The table sits inside a large, featureless cylinder and we have shown in the past that if you create a “panorama” by arranging a swathe of black cloth around the walls, ants can learn to navigate to the food, using the panorama as though it was a natural skyline. We can rotate the entire cylinder between tests to make sure that the ants are really relying on the panorama, rather than using the earth’s magnetic field, or the position of the sun, or any other unknown cues.

 

For this study, we made ants wear an “eyepatch” over their left eye, so they could not see anything in the left of their visual field. We made sure that the panorama was completely blank on the right-hand side and only had useful information on the left. This meant that the position of the black shapes on the panorama still specified exactly where to find the food, but whenever an ant was actually facing toward the food, all it could see was blank walls. An alignment image matching algorithm could never succeed in this situation: the “reference image” it tried to match to would just be a pure white wall and so facing in almost any direction (except with the black shapes in view of the good eye) would be an equally good match, so it could never decide which direction to travel in.

 

Our ants needed a little help in the form of a short wooden channel that ensured they were heading in the right direction at the start, but they soon learned to find the food almost as well as two-eyed ants. When we messed with them by moving the black shape on the panorama left or right, the ants’ shifted their paths to the left or right as well, confirming that it was the position of the black shape that was telling them where to go. When we removed the black shape entirely, the results were interesting: the mean direction of all the ants was quite close to the food, suggesting that the little starting channel did give them some information about which direction to go in, but the ants’ routes were much more circuitous and variable than before, confirming that they needed to use the black shape to control their route effectively.

 

This experiment shows that ant navigation is not so simple as turning the body until their current view matches a memory and then walking forwards. One hint that this might be the case can be found in the way ants walk. Far from walking in a straight line, always facing their goal, most ants walk with a distinctive zig-zag motion and actually spend more time looking in directions to either side of the goal than at the goal itself. This behaviour may have evolved to help ants follow pheromone trails or may even be a leftover from much further back in evolutionary history; but it may also play a role in visual navigation, allowing ants to constantly scan the scene in front of them. When we looked closely at the turns our ants made, we found that a lot of the leftward zig-zags ended with the ant facing toward the black shape on the panorama and a lot of right turns ended with the ant facing directly toward the food, even though there were no visible cues to tell the ant where to face. It may be that the ants had learned how large a right turn they needed to turn away from the bar toward the goal direction and were able to use this to keep them on course: seeing the black shape during a leftward turn would trigger a right turn of exactly the right size needed to ensure they were still pointed in the right direction.

 

This probably isn’t a special tactic to deal with being forced to wear an eyepatch but might instead be how ants typically navigate. Of course, under normal situations there would probably be shapes on the skyline on both sides and they would also be able to learn what size of turn to make in the other direction. They could even learn to recognise lots of landmarks on the horizon and remember what size of turn was needed from each one; see my 2016 paper with Cornelia Buehlmann and Tom, below, for further evidence for this idea. I’m unlikely to continue this project further, but in the future experimenters could carefully change the positions of the shapes in the panorama to test whether the sizes of the zig-zags made by ants are tuned to correspond to the size of turns they need to make in order to keep facing their goal.