Predator or provider? How wild animals respond to mixed messages from humans

Abstract

Wild animals encounter humans on a regular basis, but humans vary widely in their behaviour: whereas many people ignore wild animals, some people present a threat, while others encourage animals' presence through feeding. Humans thus send mixed messages to which animals must respond appropriately to be successful. Some species appear to circumvent this problem by discriminating among and/or socially learning about humans, but it is not clear whether such learning strategies are actually beneficial in most cases. Using an individual-based model, we consider how learning rate, individual recognition (IR) of humans, and social learning (SL) affect wild animals' ability to reach an optimal avoidance strategy when foraging in areas frequented by humans. We show that 'true' IR of humans could be costly. We also find that a fast learning rate, while useful when human populations are homogeneous or highly dangerous, can cause unwarranted avoidance in other scenarios if animals generalize. SL reduces this problem by allowing conspecifics to observe benign interactions with humans. SL and a fast learning rate also improve the viability of IR. These results provide an insight into how wild animals may be affected by, and how they may cope with, contrasting human behaviour.

Keywords: generalization; human–wildlife interactions; individual recognition; learning rate; optimal foraging; social learning.

Figure 1.

Theoretical optimal strategies to avoid or stay in scenarios of varying danger and reward. Each ternary plot shows the complete parameter space of all possible human population compositions, with proportions (as percentages) of neutral, dangerous and rewarding humans on each axis. For each set of proportions, the energy lost from avoiding an encounter with a human has been subtracted from the mean energy gained or lost by staying. Parameter space shaded red indicates scenarios where the optimal strategy is to avoid interacting with humans, parameter space shaded blue indicates scenarios where the optimal strategy is to stay and white indicates neither strategy is better than the other; (a) depicts a scenario where the cost of encountering a dangerous human is equal in magnitude to the benefit of encountering a rewarding human (here, a change in energy of −1 versus +1, respectively); (b) shows the effect of increasing the cost of encountering a dangerous human by 50% (a change in energy of −1.5); (c) shows the effect of increasing the benefit of encountering a rewarding human by 50% (to 1.5); (d) shows the effect of increasing the cost of avoiding an encounter from 0.1 (shown in (a–c)) to 0.25. Inset in top right shows the direction of each axis.

Figure 2.

Results of model simulations where critters are able to learn to avoid dangerous humans and to stay when they encounter neutral and/or rewarding humans. Models have been run to equilibrium (200 time steps). Each ternary plot shows the complete parameter space of all possible human population compositions, with percentages of neutral, dangerous and rewarding humans on each axis. The cost of encountering a dangerous human is set to be equal in magnitude to the benefit of encountering a rewarding human (i.e. change in energy of −1 versus +1). The cost of avoiding is 0.25. Each row displays a different learning weight. Row (a) shows a low learning weight of 0.1 whereas row (b) shows a high learning weight of 0.9. Columns show different metrics for each scenario. Column (i) shows the population mean probability of avoiding an encounter with a human: blue indicates that critters tend to stay whereas red indicates that critters tend to avoid. Column (ii) shows the mean energy gained at each encounter. Parameter space shaded magenta indicates a net gain in energy and parameter space shaded grey indicates a net loss in energy, with saturation indicating the degree of loss/gain. Column (iii) shows closeness to the theoretical optimal avoidance strategy: the mean energy change at each encounter subtracted from the maximum theoretical energy possible. Blue indicates convergence on the optimal strategy and shades moving towards red show increasing distance from the optimal strategy. Inset in top left shows the direction of axes.

Figure 3.

Heat maps demonstrating how changing the proportion of dangerous humans in the population affects the utility of varying degrees of discrimination (D) in terms of total energy gained or lost over time. In these scenarios, only populations with dangerous and rewarding humans are considered; neutral humans are ignored as they cause no change in energy. Thus, a value of 0 dangerous humans indicates a scenario where all humans are rewarding. Dangerous humans cause a decrease in energy equal to the gain in energy caused by rewarding humans (i.e. −1 versus +1) and the cost of avoiding is 0.25. Because the proportion of dangerous humans in the population determines the absolute possible final energy value, the energy values have been normalized to be between 0 and 1 within each level of human population composition. Values are therefore relative within columns along the x-axis: white squares indicate the worst performing D value for a given proportion of dangerous humans and shaded squares indicate how much better alternative levels of D fare in a given scenario. Row (a) shows a low learning weight of 0.1 and row (b) shows a high learning weight of 0.9. Column (i) shows energy after 500 encounters and column (ii) shows energy after 3000 encounters. N_humans = 100; N_critters = 500.

Figure 4.

The effect of social learning (SL) on critters' ability to adjust their avoidance behaviour following a change in human population composition. After 100 encounters, the human population is reduced from 80% dangerous to 20% dangerous. Blue lines depict critters without SL (S = 0), orange lines depict critters with SL (S = 1), and grey lines indicate the baseline probability of avoiding without SL when the human population is composed of 20% dangerous humans at the beginning of and throughout the time series. Results are shown for critters with a low learning weight of 0.1 (a) and a high learning weight of 0.9, showing the associated effect of learning asymmetry (b). Figure shows 10 repeated simulation runs.

Figure 5.

Time series showing the effect of SL on the ability of critters with IR (D = 1) and a high learning weight (W = 0.9) to reach the optimal avoidance strategy in a population of 80% dangerous humans. Here, three scenarios are shown: one where critters do not socially learn (S = 0; blue line), one where critters socially learn from observing the nature of conspecific interactions alone (S = 1, no alarm signalling; orange line) and one where critters also learn from conspecifics' avoidance of a dangerous human (S = 1, alarm signalling; dark brown line). Lines show critter population mean ± s.e. probability of avoiding at each time step. The optimal avoidance strategy is indicated by a dotted line.