University of Michigan researchers, in collaboration with New York University, have utilized artificial intelligence to confirm that rapid bursts of avian evolution in the Passeriformes order align closely with historical climate shifts. Published in Nature Ecology & Evolution, the study leverages machine learning to analyze skeletal data across 45 million years of evolutionary history, validating long-standing theoretical models of adaptive radiation. To process more than 170,000 skeletal measurements from over 2,000 contemporary species, the team deployed Skelevision, an AI vision model developed through a seven-year partnership between the U-M and NYU research groups. Skelevision photographs specimens against a calibrated grid and autonomously measures a dozen key bones in approximately 45 seconds, digitizing thousands of museum holdings efficiently. Lead author Jake Berv subsequently applied a novel statistical framework named Bifrost to model how these morphological changes correlated across the entire avian body over time. The analysis reveals that passerines experienced accelerated body-shape evolution roughly 35 million years ago, directly coinciding with the Eocene-Oligocene transition and a period of intense global cooling. A subsequent cluster of evolutionary slowdowns emerged approximately 15 million years ago, aligning with another major geological shift. These patterns substantiate evolutionary theory, which posits that new species diverge rapidly when exploiting novel ecological niches, before rates decelerate as environments stabilize. Beyond historical timelines, the study identified a spatial correlation: bird communities inhabiting higher latitudes and regions with pronounced seasonal temperature fluctuations exhibit faster morphological evolution than those near the equator. Researchers attribute this to environmental variability acting as a consistent driver of physical adaptation across both temporal and geographic scales. The research underscores the critical value of natural history museums as foundational data repositories. By integrating AI-driven photogrammetry with advanced statistical modeling, scientists can now extract macroevolutionary insights from legacy specimens at unprecedented scale. Senior author Brian Weeks emphasized that early collectors could not have anticipated how digital preservation and machine learning would later unlock these specimens. As contemporary ecosystems face unprecedented anthropogenic climate disruption, these historical baselines provide crucial context for predicting biological resilience. The findings demonstrate that environmental volatility directly catalyzes morphological diversification, suggesting that modern species may undergo rapid adaptive shifts in response to current climatic instability. The study ultimately bridges computational biology, paleontology, and climate science, illustrating how AI can transform static museum collections into dynamic models for understanding Earth’s evolutionary past and future.