Machine learning tracks female butterfly diversity

I’m prepping a course on AI and sustainability for the fall, so everything AI-related is interesting right now, including this article about using machine learning to look at visible diversity in birdwing butterflies. Historically, male diversity has been studied more extensively, partly due to Darwin’s claims about sexual selection applying primarily to males due to female selection of mates. However, machine learning provides opportunities to review large samples of populations, potentially helping to balance research on historically neglected groups.

In this study, Hoyal Cuthill, Guttenberg, and Huertas (2024) explore how machine learning can support the analysis of visual diversity in order to compare the diversity between male and female birdwing butterflies. They found that males do often have more distinct shapes and patterns, but there is still significant diversity among females and, therefore, they both contribute to the overall diversity. In addition, some subgroups had greater diversity in female birdwing butterflies than males. This supports the view that, at least in some groups, sexual selection is not the main driver of variation.

In keeping with my theme of sustainability, you might wonder what the connection is here. Too often, we make decisions without realizing that the data we base those decisions on (please, may we be basing them on data) is biased. There is often a view that data isn’t biased, but the humans who came up with the research questions to ask are biased. Therefore, the questions can lead to biased data. Previously, I wrote a post about hibernating bats and who does the research. This research is somewhat related but shows how we can use tools like machine learning to look at large data sets and potentially uncover knowledge that would have been difficult to get otherwise. At the same time, it still requires someone to come up with the questions to investigate and track down the data sources to use.