Abstract:

Many researchers are already experimenting with AI tools for drafting code, synthesizing literature, outlining proposals, or summarizing datasets, yet few feel they have a clear framework for evaluating what these systems are actually doing. Are they sophisticated statistical instruments, overconfident pattern generators, or something that genuinely changes how symbolic information is processed? Without a structural mental model, it is difficult to decide where AI belongs in rigorous scientific work. This two part series is designed for environmental data scientists who want conceptual clarity before adoption.

Part I: The Mental Model You Need (Mar 3)
The first lecture builds a technical and historical understanding of modern AI. We will examine how intelligence has migrated into external artifacts over centuries, from writing and maps to probability theory and forecasting, and then look directly at the mechanics of autoregressive language models. What does it mean that these systems are probabilistic? How does next token prediction relate to Bayesian conditioning and time series modeling? What exactly is being compressed when a model is trained on vast symbolic corpora, and why does scale change behavior in ways that can feel surprising?
By connecting AI to ideas familiar from ecological modeling, uncertainty propagation, and predictive systems, this session provides a clear framework for understanding both capability and failure modes. The aim is not to teach tools but to equip researchers with a principled way to interpret outputs, anticipate errors, and separate fluency from validity.

Part II: From Model to Practice (Mar 10)
The second lecture moves from theory to workflow. Where does AI genuinely add leverage in environmental research? How can it assist in hypothesis generation, coding, data analysis, and scientific writing without compromising rigor? What verification strategies are necessary? How should labs think about authorship, reproducibility, and transparency when these systems are involved? Together, these sessions aim to build AI literacy for scientists. Not advocacy. Not alarmism. The goal is to evaluate a new class of predictive instruments with the same care we apply to models, simulations, and statistical inference, and to understand these systems well enough to decide deliberately how or whether they belong in your science.
 

Speaker Bio:

Dr. Ty Tuff is your friendly neighborhood data scientist, based at the Environmental Data Science Innovation and Inclusion Lab (ESIIL). With deep expertise in processing, analyzing, and modeling data, Ty has led cutting-edge research projects in global institutions and is passionate about advancing the frontiers of AI and environmental data science. Ty's work focuses on turning complex datasets into actionable insights while fostering collaboration across disciplines.