A newly developed AI tool for image analysis has the potential to enhance algorithms that examine wildlife photographs, providing insights into how species worldwide are responding to climate change, according to a recent study.
A newly developed AI tool for image analysis has the potential to enhance algorithms that examine wildlife photographs, providing insights into how species worldwide are responding to climate change, according to a recent study.
This advancement could enable scientists to develop AI-driven algorithms that quickly and thoroughly analyze the vast number of wildlife images uploaded online by the public each year.
Such tools could uncover significant information regarding the effects of climate change, pollution, habitat degradation, and other challenges impacting thousands of animal and plant species, according to researchers.
Citizen science platforms serve as a valuable source of data on how various species are reacting to climate shifts. However, although current AI algorithms can identify species in the images submitted, it was uncertain if they could extract additional information.
Now, a global team of scientists has designed a new tool to evaluate the effectiveness of AI algorithms in examining image databases for further insights. This might include aspects such as the diet of species, their health status, and interactions with other species.
The tool, known as INQUIRE, assesses the capacity of AI to draw conclusions from an extensive collection of five million wildlife photos that have been uploaded to the iNaturalist citizen science website.
The research team discovered that while present AI algorithms can answer some questions, they struggle with more intricate queries. These complex questions often necessitate reasoning about minute details in images or involve scientific terminology.
The results emphasize the potential for creating new AI algorithms that can assist scientists in efficiently navigating vast image archives, according to the researchers.
The peer-reviewed results will be showcased at the NeurIPS conference, a prominent event focused on machine learning.
The research team comprises experts from the University of Edinburgh, University College London, UMass Amherst, iNaturalist, and the Massachusetts Institute of Technology (MIT). This work received partial funding from the University of Edinburgh’s Generative AI Laboratory.
Dr. Oisin Mac Aodha, from the University of Edinburgh’s School of Informatics, remarked, “The thousands of wildlife images uploaded online daily offer scientists valuable insights into the habitats of various species on the planet. However, identifying the species in a photo is just the beginning.”
“These images are a potentially diverse resource that remains mostly unexplored. The ability to swiftly and accurately sift through the vast information they hold could provide essential insights into how species are adjusting to complex challenges like climate change.”
Dr. Sarah Beery, an Assistant Professor at MIT, mentioned, “This meticulous data curation addressing real scientific inquiries across ecology and environmental research is crucial in broadening our understanding of what current AI techniques can achieve in these impactful scientific contexts.
“It has also highlighted research gaps that we can now strive to fill, especially regarding complex compositional queries, technical language, and the subtle variations that distinguish categories of interest for our partners.”
