Is bin-opening in cockatoos leading to an innovation arms race with humans?

Cockatoos in Sydney have begun to open suburban rubbish bins, with knowledge spreading to form local traditions. In accessing food, cockatoos spread garbage onto the street, putting them in conflict with people. Protecting bins from cockatoos is challenging as the lid needs to still open when the bin is tipped over into the garbage truck, yet local human residents have innovated various solutions to protect their bins from attack (Figure 1A,B).

To map the local patterns of bin-protection behaviour, we recorded 3283 bins across four suburbs with reported bin-opening activity by cockatoos (Supplemental information). For each bin, we recorded whether or not it was protected (yes/no), and if so: rationale of protection, method, location of the protection device on the bin, material of the device and whether or not it was fixed to the bin (Figure S1A and Table S1 in Supplemental information). Protection levels varied between suburbs from 0% to 50% (Figure S1B). Using Gower distance, a method for hierarchical classification of categorical data (Supplemental information), we calculated distances between different methods based on the above categories (n = 52 unique combinations). Our data were best described by 13 distinct clusters, which we further assigned to levels based on the stage and degree of alteration to the bin (Figure S1A). These ranged from level 2 — no functional alteration to the bin (e.g. a rubber snake), level 3 — unfixed object to prevent lifting (e.g. rock), level 4 — object to prevent flipping (e.g. shoes in hinge), to level 5 — fixed alteration to prevent opening (e.g. an attached weight; representing the greatest investment). Variation in the efficacy of these levels on preventing bin-opening was then verified with personal observations. We observed that ‘no functional alteration’ and ‘low efficacy’ devices (e.g. rocks) could be solved by cockatoos (Figure 1A), while we have not yet observed cockatoos opening bins protected at level 4 or 5.

We investigated the distribution of protection methods using a spatial network approach. Within each suburb, we found that protected bins were geographically more assorted by protection status (yes/no; r = 0.076–0.096) than expected if protection was randomly distributed (all p ≤ 0.003). Protected bins were also geographically assorted by protection type, based on the identified 13 clusters: r = 0.133–0.329, all p ≤ 0.002 (Figure 1C and S1B). With the exception of overall protection in Stanwell Park (>50% protection level, p = 0.692), bins close to each other on street distance were significantly (all p ≤ 0.046) more likely to use similar protection methods than those that were spatially close but out of sight (e.g. neighbours on the street behind). This suggests that people copy protection devices from neighbours on the same street, creating street-based clusters of protected bins and protection methods.

These direct observations were complemented with broad-scale citizen-science. In 2019 and 2020, we conducted online surveys asking residents in the Greater Sydney and Wollongong Regions: in which suburb they lived, what protection device they used (Table S1) and whether/when they changed protection devices (Supplemental information). We also gave them opportunities to add written comments. Of 1134 participants across 401 suburbs, 172 from 51 suburbs reported protecting their bins against cockatoo attacks. Of the respondents that provided sufficient information, 64% reported using social information, of which 60% directly referenced neighbours or people on their street, supporting our empirical findings (Table S1). Further, 61% of those that protected their bins increased their protection efficacy over time (Figure 1D and Table S1). A subset of respondents also commented on their reasons for changing. Cockatoo behavioural change was the most frequently cited reason both for escalation (cockatoos solving the method, 39% of escalation responses) and de-escalation (cockatoos no longer opening bins in the local area, 44% of de-escalation responses; Table S1). For example, one person commented: “Bricks seemed to work for a while, but cockies got too clever. Neighbours on other side of highway suggested sticks. They work.”

Finally, using data from the online survey, we modelled how bin-protection methods changed over years and as a function of how long cockatoos had opened bins in that suburb. First, an ordinal logistic mixed effects regression revealed that each year (Figure 1D) increased the odds that residents used a higher protection level by 3.6 times (p < 0.001), with the time since cockatoos started bin-opening increasing these odds by 3.02 (p < 0.001). The model also suggested that the rate of escalation slowed down as bin-opening became more established in an area (β = −0.11, p < 0.001). We therefore investigated how rates of transitioning between efficacy levels varied over these two temporal axes using a multi-state Markov chain model (Supplemental information). This revealed that changes occurred relatively gradually, with 89% of bins remaining unprotected in the first three years after bin-opening started, and while uptake of protection sped up the longer cockatoos opened bins, escalation to higher protection methods slowed down (Figure 1E). However, the model predicted that escalation, whether from unprotected to protected or from low-medium to high protection, always occurred at higher rates than de-escalation (Figure 1E).

Extending the concept from the study of human cultural evolution, and analogous to evolutionary arms-races, we consider that an innovation arms-race requires re-iterated learned behavioural change in two populations/species (whether escalation or de-escalation), each driven in response to the other. This could occur via individual-level learning, or as a ‘cultural arms race’, if social learning occurs on both sides. Here, we demonstrate that bin-protection and protection types were geographically clustered and socially learnt from neighbours. While we don’t directly measure responses by bin-opening cockatoos, this geographic clustering creates conditions for further innovation and/or social learning. Moreover, people increase efficacy of protection measures over time, most often self-reported as escalation in response to cockatoos defeating previous methods, and predicted by how long cockatoos have been locally opening bins. These results give indirect evidence for at least one re-iteration of behavioural change by local cockatoos. To fully understand these dynamics, further work is needed to establish the underlying learning mechanisms, and to identify the extent to which cockatoos can solve medium and high-efficacy measures.