CRII: III: Explainable Artificial Intelligence for Biodiversity Science & Conservation

生物物种的灭绝正在加速。即使有高分辨率的信息，在预测物种灭绝或种群减少方面也往往存在很大的不确定性。生物物种分类的变化是影响生物多样性保护和政策决定的一个关键挑战。分类学是用文字描述的，这些文字提供了使用自然语言处理和机器学习（ML）来澄清物种关系的机会，并通过解决物种分类学的可变性来提供对灭绝风险的新见解。为“分类智能”开发准确和可扩展的机器学习和人工智能（ML/AI）可以帮助支持保护决策的鲁棒性。这一点很重要，因为生物分类学可以将一组生物纳入或排除法律的保护范围。人工智能可以帮助这种分类，并支持保护项目的协调。该项目的目标是开发AI/ML技术，通过预测不同分类学视角下物种分布和风险的不同偶然结果，为灭绝风险提供新的见解。至关重要的是，衍生的见解是人类可以理解的，以安全地将这些结果转化为操作建议。生物多样性数据，包括分类学和地理空间数据，对人工智能提出了独特的挑战，因为它们是异构的，结构复杂的，并且经常变化。该项目旨在通过一种新的方法来解决这些挑战，该方法将自然语言处理（NLP）与相关科学出版物的文本数据相结合，并自动进行归纳和演绎推理，包括定性空间推理，结合分类因子和相关领域结构，以发现人类可理解的保护生物学应用知识。在这样做的过程中，该项目也有可能将人工智能推进到单一应用领域之外。该奖项的研究活动包括在领域专家的帮助下进行数据和知识管理，以及开发和评估上述人工智能技术。该奖项反映了NSF的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Extinction of biological species is accelerating rapidly. Significant uncertainty is often involved in predicting the extinction or population decline of species, even with high-resolution information. Changes in the taxonomic classification of biological species is a key challenge that impacts both biodiversity conservation and policy decisions. The taxonomy is described in words and these words provide an opportunity to use natural language processing and machine learning (ML) to clarify species relationships and provide novel insights into extinction risk by addressing the variability in species taxonomy. Developing an accurate and scalable machine learning and artificial intelligence (ML/AI) for “taxonomic intelligence” can help support the robustness of conservation decision making. This is important because the taxonomic classification can move a group of organisms in or out of consideration for legal protection. AI can help in this classification and support coordination of conservation projects.The goal of this project is to develop AI/ML techniques to provide novel insights into extinction risk, by projecting different contingent outcomes for species distributions and risks under different taxonomic perspectives. It is critical that the derived insights be understandable to humans, to safely translate these outcomes into operational recommendations. Biodiversity data, which include taxonomical and geospatial data, pose unique challenges to AI in that they are heterogeneous, structurally complex, and frequently change. This project aims to address these challenges with a novel approach combining Natural Language Processing (NLP) from the textual data of relevant scientific publications, and automated inductive and deductive reasoning, including qualitative spatial reasoning incorporating the taxonomic factor and relevant domain structures, for discovery of human-understandable knowledge for conservation biology applications. In doing so, this project also has the potential to advance AI beyond a single application domain. The research activities to be undertaken in this award include data and knowledge curation with the help of domain experts, and the development and evaluation of the aforementioned AI techniques.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.