Sustainable AI Deployment and Opportunities for Statistical Practice

As generative AI moves rapidly from experimentation into day-to-day deployment across government, industry, and research sectors, both its areas of application and its consumption of resources are growing. This invited panel will bring together experts in statistics and computer science to discuss practical strategies for environmental and cost sustainability of AI systems while continuing to address impactful societal solutions without sacrificing performance or productivity. Panelists will explore the unique role statistical practitioners can play in guiding model selection, evaluation, and deployment choices; present governance frameworks for sustainable practices for optimizing resources; and discuss sustainability education. Panelists include Dr. Steve Sain, Senior Director, Geospatial and Data Sciences (Jupiter Intelligence); Prof. Cynthia Rudin, Distinguished Professor and PI, Interpretable Machine Learning Lab (Duke); Interim Head, Computer Science and Engineering Dept Mahmut Kandemir (Penn State), working with the Institute of Energy and the Environment; and Dr. Lyndsay Shand, Sandia National Laboratories. Moderated by Liz Mannshardt, Vice President and Director of Statistics and Data Science, Westat. They will highlight current research and implementation efforts in model dispatching and domain-specific model development (e.g., lightweight or fine-tuned models for areas like health care), human-interpretable machine learning for simpler models and practical code for sparse models (such as decision lists, decision trees, and additive models that provably optimize accuracy and sparsity); optimization techniques such as quantization and low-rank adaptation, deployment strategies such as aligning computation with renewable-energy availability across time zones, and sustainable data-engineering practices such as structured metadata. By highlighting concrete, actionable options already available to organizations and describing proper governance frameworks, this session aims to help statistical practitioners and data science leaders make more environmentally and financially responsible decisions in developing and operationalizing AI tools