Placing the "Lens of Evolution" on AI: AI Understanding Evolution, Applications Ranging from Drug Discovery to Paleontology

A team from Ruhr University Bochum in Germany has announced a method where they provide AI with a known phylogenetic tree (evolutionary relationships) as "prior knowledge" and train it so that the feature space during learning aligns with these phylogenetic relationships. The key concept is the "quartet" approach, which involves designing a loss function to ensure that the order of all quartets is correct. Once all quartets are aligned, the overall phylogenetic tree is determined like a puzzle. This method can be extended beyond genetic sequences to include structural information and images, known as "phenomics," and is expected to have implications for virtual reconstructions of ancestral forms and estimating the pathways of trait evolution. However, there are concerns that using an incorrect reference phylogenetic tree could amplify biases, and challenges remain in validating the method's applicability to non-sequence data like images, as well as in computational costs. Although its exposure on social media is currently limited, practitioners on LinkedIn have praised its "cross-modality significance," and press releases are spreading. At the intersection of evolutionary research and machine learning, the emergence of a new standard is visible.