Organizers: Eric Slud
When: Tentatively Mondays at 3PM in PHYS 2122
The topic for fall 2026 is "AI in Statistical Consulting". A breif description follows.
I plan to run the Statistics RIT on the topic of effective use of AI in Statistical Consulting. We will explore what statistical consultants do in reacting to data problems and research questions presented to them for statistical analysis, and how this activity can be speeded up and made more effective by interaction with Large Language Models (LLM's, AI) like Gemini or Claude. Although we are trained in the mathematical statistical aspects of data analysis in relation to theoretically supportable assumption, non-statistician clients are often still formulating their data collection designs and research questions, and may have only general goals like writing a paper with publication-quality statistical analysis on a scientific problem they have been studying. Eliciting the research questions that can be analyzed and answered in a principled way from the data they have or can collect is an activity with a strong linguistic aspect, and therefore one might suppose that LLM's can help. On the other hand, getting meaningful and concise answers from AI on such questions is tricky.In this RIT, we will first explore a little background of what statistical consultants do, and what has been written about it both in the past and also more recently in connection with attempts to develop AI "agents" to replace or supplement old-style statistical consultants. But the main activity of the RIT will be to do statistical analyses interactively using AI, role-playing the refinement of research questions and the choice of statistical method and setting up and doing analyses. As I envision the RIT, the ideal presentation each week will display the interaction with AI and statistical results for a problem based on a statistical dataset. The datasets, to be found by the RIT participants, can be anything from a well-defined dataset in a published paper to a raw dataset (not fully cleaned up and prepped for analysis) coming from a public data-source.There is a web-page for the RIT https://www.math.umd.edu/~slud/RITF26, with some content now but still under construction, which explains more of what the RIT might hope to accomplish and lists some sample AI interactions related to statistical-consulting background, to AI "prompt engineering" in a consulting setting.Please have a look at the website https://www.math.umd.edu/~slud/RITF26, and let me know if you are interested in participating in the RIT, especially if you would like to join but cannot do so at 3pm Monday afternoons. Depending on interest and in-person availability at that time, we might try to run the RIT sessions in `hybrid' format, both in-person and over Zoom.The organizational meeting will be held on Monday September 14, 2026, PHYS 2122. I will speak for 1/2 hour or so on what our prompts and questions to AI about a newly arrived consulting problem should look like, and what one can hope to achieve with the answers.