Machine learning outcompetes human assessment in identifying eggs of a conspecific brood parasite

Abstract

Avian brood parasitism provides an exceptional system for studying coevolution. While conspecific brood parasitism (CBP) is more common than interspecific parasitism, it is less studied due to the challenge of detecting parasitic eggs, which closely resemble those of the host. Although molecular genotyping can accurately detect CBP, its high cost has led researchers to explore egg appearance as a more accessible alternative. Barn swallows (Hirundo rustica) are considered conspecific brood parasites, but identifying parasitic eggs has traditionally relied on human visual assessment. Here, we used UV-visible photographs of non-parasitized barn swallow clutches and simulated parasitism to compare the accuracy of human assessment with automated methods. In two games, participants and models identified parasitic eggs from six or two options. While humans performed better than chance (72% and 87% accuracy), they still made significant errors. In contrast, the automated supervised model was far more reliable, achieving 95% and 97% accuracy. We think that the model outperformed humans due to its ability to analyse a broader range of visual information, including UV reflectance, which humans cannot perceive. We recommend using supervised models over human assessment for identifying conspecific parasitic eggs and highlight their potential to advance research on evolution of egg colouration.