The International Telecommunication Union (ITU) and partners held a webinar to explore how artificial intelligence (AI) could help implement a formal framework and guidelines for the detection and attribution of biodiversity change to support effective policymaking.
The webinar kicked off an AI for Good Discovery Series on AI for Biodiversity. Organized in partnership with the Convention on Biological Diversity (CBD), the series aims to bring together the AI and biodiversity communities to help protect our natural world.
María Cecilia Londoño, Humboldt Institute, and Mike Gill, NatureServe’s Biodiversity Indicators Program, moderated the discussion.
David Cooper, Acting Executive Secretary, CBD, welcomed the series’ goal of motivating creative solutions to support the conservation and sustainable use of biodiversity, in a way that is fair and just.
Cooper drew attention to the recently launched Global Biodiversity Framework (GBF), comprising four goals for 2050 and 23 actionable targets for 2030. He said to implement this whole-of-government and whole-of-society framework, countries must translate it into national plans and targets, mobilize financial resources, and build capacity.
Identifying the need for tools to enable indicator-based progress monitoring, Cooper noted AI’s existing applications in identifying species, pinpointing drivers of biodiversity loss, and mapping areas suitable for ecosystem restoration, among other uses. Calling for representative, high-quality data, he said it will require human intelligence to ensure that AI does not lead to local knowledge being cast aside and that optimization algorithms do not undermine stakeholders and rights holders, including Indigenous Peoples and local communities (IPLCs).
Andrew Gonzalez, McGill University, proposed a framework for the detection and attribution of biodiversity change where AI could play a prominent role by tracking progress and linking action on drivers to outcomes on biodiversity. He outlined the framework’s five steps:
- Causal modeling;
- Observation, or the process of recording elements of biodiversity;
- Estimation, referring to statistical estimates of variables;
- Detection of change; and
- Attribution – the process of evaluating the (in)consistency of the detected change to multiple causal factors with an assignment of statistical confidence to the causal models used to estimate those effects.
Gonzalez suggested the framework could strengthen the bridge between biodiversity science and AI and support rapid assessments of actions required to mitigate human impacts and reduce rates of biodiversity loss.
Among challenges he identified:
- Data inequities, with billions of observations concentrated in less than 7% of the world’s surface;
- Data sovereignty;
- Accessible workflows for indicators; and
- Collaboration for better outcomes.
Gonzalez underscored the need for theory-guided data science that would combine big and small data to guide AI application. He outlined the role on Biodiversity Observation Networks in co-producing monitoring, from biodiversity observations, to detecting and attributing biodiversity change, producing scenarios and forecasts, and supporting policy and decision making and local management. He provided examples of AI uses, including: advancing causal discovery by text mining; enabling a fully automated approach to monitoring; using neural network approaches to predict species interactions; using machine learning to predict large-scale declines in plant biodiversity; and using extreme gradient boosting to predict tree diversity under different climate change scenarios.
During the question-and-answer session, participants discussed bottlenecks in advancing AI in the detection and attribution system, including those relating to AI literacy, data availability, and the disconnect between the human, the analytical tool, and the inferences we make. They expressed support for public-private partnerships (PPPs) to harness tools for biodiversity change, peer-to-peer data sharing, and combining causal inference frameworks in machine learning with traditional and theoretical knowledge.
AI for Good is a digital platform where AI innovators and “problem owners” come together to identify practical AI solutions to advance the SDGs. AI for Good is organized by ITU, in partnership with 40 UN agencies, and co-convened with the Government of Switzerland. [SDG Knowledge Hub Sources] [Video: Detection and Attribution of Biodiversity Change: A Role for AI]