What do Artificial Intelligence and Complexity Theory have to do with a water research institute like Eawag? This tenure talk explores the first five years of the Machine Learning & Complex Systems group, showcasing how these tools advance research, policy, and theory across environmental domains.

The talk begins by demonstrating how machine learning (ML) can reduce animal testing for environmental risk assessment, addressing critical gaps in data availability and study comparability.
It then highlights how ML is transforming biodiversity monitoring through non-invasive, high-throughput observations. Using camera traps to identify birds and microorganisms, ML provides data that is used to support species abundance forecasting and ecosystem modeling through theory of complex systems.

Complexity theory is also applied to understanding ML itself. The talk explores three key topics—community detection, class imbalance, and architecture bias—where theoretical insights lead to practical strategies for training better models, particularly with limited or imbalanced data.

These contributions showcase the transformative potential of ML in addressing environmental challenges, but this promise comes with caveats. The environmental and ethical costs of ML—reliance on massive computational resources, cheap labor, and high energy consumption—pose significant concerns. Additionally, private sector investments increasingly outpace academic efforts, raising questions about long-term sustainability and independence. The talk invites reflection on how scientists and institutions can navigate these challenges to maintain their relevance and integrity in a rapidly changing landscape.