Summary: A new AI-driven study examines evolutionary differences between male and female birdwing butterflies, offering fresh insight into a long-standing debate between Charles Darwin and Alfred Russel Wallace.
Using machine learning to analyze more than 16,000 museum photographs, researchers show that both sexes play important roles in species-level diversity. Males frequently display greater visible variation, consistent with Darwin’s sexual selection hypothesis, while subtler but significant variation in females supports Wallace’s emphasis on natural selection. Together these findings demonstrate how sexual and natural selection can act jointly to shape biodiversity.
Key facts
- Machine learning examined over 16,000 dorsal and ventral photographs of male and female birdwing butterflies to quantify phenotypic diversity.
- Males more often exhibit pronounced variation in shape, color and pattern, supporting Darwin’s sexual selection ideas.
- Females sometimes show subtler but biologically meaningful variation consistent with Wallace’s focus on natural selection across sexes.
Source: University of Essex
Pioneering AI research probes the under-studied evolution of female butterflies and revisits a classic evolutionary debate.
Published in Communications Biology, the University of Essex study addresses a Victorian-era controversy between Charles Darwin and Alfred Russel Wallace. Darwin predicted greater variation in males driven by female mate choice, while Wallace emphasized natural selection acting on both sexes as a key driver of inter-species differences. Historically, researchers have examined male butterflies more often because their differences are visually striking; female variation, being more subtle, has received less attention.
Dr Jennifer Hoyal Cuthill and colleagues from the Natural History Museum and AI research institute Cross Labs, Cross Compass applied deep learning to museum collections, measuring visible traits—wing shapes, colors, and patterns—across species from Southeast Asia and Australasia. This is the first comprehensive visual comparison of male and female birdwing butterflies across an entire clade.
Researchers trained a convolutional neural network using a triplet-loss approach to learn image embeddings that capture phenotypic similarity. Validation showed these embedding distances can reconstruct aspects of phenotypic evolution and produce measures of phylogenetic congruence comparable to genetic species trees. In short, the AI-derived image space reflects meaningful evolutionary relationships.
Results reveal that males frequently contribute strong visible diversity—extreme shapes, vivid colors and distinctive patterns—consistent with sexual selection from female mate choice. Yet the analysis also uncovers cases where females exhibit greater visible disparity than males, aligning with Wallace’s idea that natural selection can generate and maintain diversity in female phenotypes. Overall, either sex can dominate observed phenotypic diversity depending on lineage and selective history.
For example, the genus Ornithoptera shows high male image disparity, signs of diversification under multiple selective optima in fitted multi-peak Ornstein-Uhlenbeck models, and accelerated divergence including extreme differences found in both allopatric and sympatric contexts. Conversely, the genus Troides displays an inverted pattern with relatively conserved male phenotypes and comparatively higher female disparity, though those female variants often fall within a shared inferred selective regime.
Dr Hoyal Cuthill commented that machine learning is opening new opportunities for large-scale tests of long-standing evolutionary questions, including addressing historically neglected female variation. By quantifying visible extents of evolution across sexes and species, these methods reveal how sexual selection and natural selection together shape the remarkable—and often endangered—diversity of birdwing butterflies.
About this evolution and AI research news
Author: Ben Hall
Source: University of Essex
Contact: Ben Hall – University of Essex
Image credit: Neuroscience News
Original research (open access): “Male and female contributions to diversity among birdwing butterfly images” by Jennifer Hoyal Cuthill et al., published in Communications Biology.
Abstract
Male and female contributions to diversity among birdwing butterfly images
Machine learning now makes it possible to test long-standing predictions about whether males or females show greater inter-species diversity in visible phenotype (disparity), predictions that follow from Darwinian sexual selection versus Wallacean natural selection. Here, ML is used to quantify variation across a sample of more than 16,000 dorsal and ventral photographs of sexually dimorphic birdwing butterflies (Lepidoptera: Papilionidae).
Validation of image embedding distances, learned by a triplet-trained deep convolutional neural network, demonstrates that ML can automate reconstruction of phenotypic evolution and achieve measures of phylogenetic congruence comparable to ranges seen among genetic species trees. Quantifying sexual disparity difference (male versus female embedding distance) reveals sexually and phylogenetically variable inter-species disparity.
Ornithoptera exemplify high male image disparity, diversification of selective optima in fitted multi-peak OU models and accelerated divergence with instances of extreme divergence in both allopatry and sympatry. By contrast, Troides shows inverted patterns with relatively static male embedded phenotypes and higher female disparity, though often within a common inferred selective regime for those females. Overall, the shapes and color patterns most distinctive in ML similarity tend to be those of males, yet either sex can be the major contributor to observed phenotypic diversity among species.