On April 28, Milwaukee became the hub for the latest breakthroughs in analytics at the Symposium on Data Science and Statistics. It was a gathering of the field’s top academic researchers, and JMP was right at the center of the conversation. On day two, Sarah Sylvestre (@Sarah-Sylvestre) and Sarah Gilyard (@SarahGilyard) had the privilege of representing JMP, presenting a poster to a crowd focused on the latest trends in data science and statistics research.
The statistician’s role in the era of AI
SDSS kicked off with an interesting plenary from Lorin Crawford, a biology researcher partnering with Microsoft, on applying AI and LLMs to single-cell sequencing. Lorin discussed the methods and results his lab is seeing and emphasized the fact that LLM model performance isn’t necessarily scaling with more data. Model performance is saturated at about 1% of the training data for certain biological tasks. This means that LLMs often need more diverse training data rather than more volume.
Lorin thinks that this is where statisticians come into play. Statisticians are needed to help fill gaps in training data by generating synthetic data that maintains the necessary variation and uncertainty structures for the task. Statisticians will be needed to help understand how to stratify the training data to better train LLMs. Lorin believes the new role of statisticians in AI/ML is that of “strategic architect of data-driven systems.”
The theme around the critical role of statisticians in the era of AI was continued into the second day during the plenary panel. Four panelists covering academia, pharmaceuticals, and theoretical physics were interviewed about their thoughts on the current AI landscape. All panelists emphasized the need to adapt to AI or risk being left behind; it is important to be mindful of the dangers. Over-reliance is bad; balance is key. In that regard, statisticians are highly needed to evaluate AI in many ways, including establishing a tolerance for reproducibility; educating AI experts around the proper handling of synthetic data so that it is not blindly handled the same as regular data; and identifying the most statistically relevant data as opposed to simply gathering high volumes of data.
Sarah Sylvestre (left) and Sarah Gilyard presented at the 2026 Symposium on Data Science and Statistics
JMP at the center of the conversation
During our poster session, we presented a survey of newer JMP capabilities, including Bayes Opt, FDE, dynamic filtering, data table updates, Python integration, and the LearnBot/Assistant. We generated several good conversations with potential JMP users during our 25-minute poster slot. Most attendees were unfamiliar with JMP, but a few were on the fence about using it. One attendee has colleagues who use SPSS, while she uses R. She would like a tool to make collaboration with her coworkers easier. After showing her several JMP capabilities, she left with a firm belief that her coworkers would benefit from JMP over SPSS. In another conversation with a professor from Dartmouth, we learned that the professor has students who are not keen on coding in Python and would prefer a tool they can use to build models point-and-click, while still integrating with Python when necessary. After discussing JMP's options for model building and Python integration, she is convinced that her students would prefer to use JMP.
Overall, it was an engaging conference with lots of discussion around how the statistical community can adapt in the era of AI, while remaining true to their craft. Although the attendees and talks were mostly academic in nature, there was significant value in spreading the JMP brand and strong interest in JMP’s newer capabilities.
.jpeg)
