The COVID-19 outbreak affected a lot of people in a lot of ways. For Project C-Gull, this meant a cessation of fieldwork and a turn towards other methods of trying to understand how gulls and other wild animals perceive their interactions with humans. Our latest paper is a collaboration with the Max Planck for Evolutionary Anthropology’s Tom Holding, a computer scientist with a background in theoretical modelling.
One of the questions we had been asking was “How can wild animals succeed when humans dominate the landscape and yet act in very different ways?” Humans aren’t like any other species: some act as predators while others provide food. It seems as though humans send “mixed messages” to wild animals, and it must be challenging for them to figure out when to stay and when to avoid people.
In addition, wild animals are very diverse, and different species or individuals will react differently from others. It can be difficult to get at broad questions such as ours just by conducting fieldwork, because there are many species and many other influences on animal behaviour besides humans. We therefore felt a modelling approach would provide a good starting point.
How can computer modelling accurately represent real animals? We know that wild animals can habituate to human presence, becoming less likely to flee over time if people are not harmful. We also know they can learn to avoid dangerous things and approach things that provide food. Additionally, we know animals’ speed of learning can vary, some can recognise individual humans, and some can learn by observing others.
We can code these “behaviours” into computer programs. The individuals (which we called “critters”) in our computer model can be thought of as a cell in a database. They are very basic, and their primary feature is the amount of energy they have acquired. These energy values are updated after they are assigned an encounter with a randomly selected virtual “human” from another database, which can cause loss or gain of energy. If the critter avoids an encounter, it loses some energy. The idea is that it has stopped foraging and fled. If the critter stays but the human is dangerous, the critter also loses energy. If the human is someone who provides food, its energy increases.
Our computer model showed that, when animals learn on their own accord and cannot distinguish among different people, it’s likely to be very difficult for them to be able to work out the best way to act. Animals may not flee when they should and vice versa.
We explored different learning rates, which determine how quickly an animal can change its behaviour after a new encounter with a human. It might be expected that a fast rate of learning would be advantageous. Animals that learn slowly are vulnerable to exploitation: consider the extreme case of the dodo, which was wiped out from its home range because it did not appear to acquire any fear of humans.
However, while avoiding dangerous people is a good thing, avoiding people who aren’t dangerous can mean that foraging opportunities are lost. We showed that fast learning could cause a high, and potentially detrimental, level of avoidance behaviour, because avoiding necessarily means forfeiting further learning opportunities. This could have a large impact on the health, survival and reproduction of wild animals, especially as there are now fewer places that are uninhabited by humans.
One way animals could overcome this problem is with individual recognition of humans (IRH). There have been a few studies showing that some wild animals (usually members of the crow family, which are known for their well-developed cognitive abilities) can recognise people who previously behaved in a threatening manner. But how useful is individual recognition really?
We showed that, if wild animals respond to each human as an individual, it could be beneficial when there is a moderate number of dangerous people in the population, but detrimental if there are many. Just a small amount of generalisation – treating humans as a group – would help animals evade the potential danger posed by humans they haven’t encountered before.
We therefore can’t expect that animals would necessarily respond as though they are capable of recognising individual humans, even if they can. The defensive behaviour towards “neutral” people often reported in IRH experiments suggests that even animals that show a stronger response to “dangerous” humans don’t fully discriminate among people, and probably for good reason.
Another way we found animals could overcome the problem of “mixed messages” is by learning about humans through observing others’ encounters. This kind of social learning may allow animals that have learned to avoid people after a negative encounter to exploit feeding opportunities when others explore areas near humans that don’t pose a threat. Social learning could be especially useful when the number of dangerous humans in a population fluctuates over time, such as during and after hunting seasons.
One way animals learn socially is by making alarm calls. These function to alert others to danger. If animals accurately identify a threat each time they encounter one, all in the population will benefit. Therefore, if animals can distinguish among individual humans, and alarm call only on these occasions, their ability to avoid only dangerous individuals will vastly improve. In reality, there will be mistakes (we all make them), but this general pattern may explain why species with individual recognition have been found to spread information about dangerous humans via alarm calls.
Our findings have implications for conservation. By identifying the learning strategies of different species, it may be possible to predict which will be most vulnerable to an increasing human presence. Knowing which species are capable of recognising individual humans and/or socially learning about them may also affect how people choose to run wildlife management programmes.
Our findings also have implications for wild animal welfare, which is often overlooked. Many people around the world engage in direct feeding interactions with wild animals, where they offer food and encourage animals to approach them closely. Some of these animals are species that are hunted or persecuted. If animals do not adequately distinguish people who feed them from people who want to harm them, they run a higher risk of being hurt or killed.
Our paper can be read in full here.