Schmidt AI in Science Fellow University of Michigan, United States
Abstract: In the last century, biodiversity has drastically declined due to climate change and other human activities, posing a threat to crucial ecological functions, genetic diversity, and socio-economically important species. To prevent further loss, targeted conservation efforts based on large-scale monitoring of all species are necessary. However, creating comprehensive datasets is challenging and costly, resulting in a lack of accurate observation for many species from which populations and distributions can be inferred. This lack of data is exemplified by the fact that one in six species is classified as data deficient by the International Union for Conservation of Nature (IUCN) Red List. To address this research gap, citizen-sourced inputs from social media and mobile applications have emerged as a powerful tool to increase biodiversity monitoring. However, the use of citizen-labeled species observations is limited by the high rate of misidentified observations, with the manual validation of such large datasets remaining unfeasible. Here, we evaluate a multimodal artificial intelligence approach, harnessing heterogeneous data types including, photograph contents, textual data, and spatiotemporal information to validate photographs of different species uploaded to the social media website ‘Flickr’. The use of this multimodal artificial intelligence validation approach shows high accuracy for rapidly generating validated species observations for a range of species. These artificial intelligence-labeled observations can be used alongside traditional data sources to improve our knowledge of species populations and distributions, to better inform biodiversity conservation.
