When does management using artificial intelligence lead to unintended consequences? A case study using ‘smart’ traps

Abstract

Rapid improvements in Artificial Intelligence (AI) are leading to new options to manage species of conservation concern, for example by developing “smart” traps which selectively trap some species over others. To better understand the potential for unintended consequences, we propose incorporating decisions based on AI into existing models of ecological dynamics. We show that density-dependent changes in the effectiveness of AI-based tools can lead to unintended consequences. We illustrate this with models of a pest species in Aotearoa New Zealand, the Ship Rat (Rattus rattus); and a bird of conservation concern, the Weka (Gallirallus australis). Traps are known to harm both species; management that is strong enough to control the rapidly growing rat population risks harming the non-target species. Furthermore, improving the ability of the trap to detect rats can indirectly, and unexpectedly, lead to declines and even extinction in populations of Weka. The extinction of Weka can only be avoided if the probability of falsely identifying a Weka as a rat is extremely low. These results highlight how AI-based pest management can have unintended consequences for populations of non-target species.