Approaching The Black Box: Teaching The Process of Ai Training Data Creation

The Digital Scholarship and Data Services team within Johns Hopkins Sheridan Libraries aims to guide the responsible use of technology in data-intensive research. Observing increased patron use of commercial generative AI products, and anticipating the rollout of an institutional generative AI platform, the team delivered a series of 3 workshops in Fall 2025 aimed at supporting responsible AI use.

Our approach was informed by the EDUCAUSE Framework for AI Literacy’s four levels of AI Literacy: Understand, Apply, Evaluate, and Create. This presentation will outline an approach to designing the first of these workshops targeting the first level: Understand AI. Students were intended to come away with a baseline understanding of how generative AI tools, particularly those built on large language models, generate responses to prompts.

Drawing from the literature on data & AI literacy, the ACRL Framework for Information Literacy, and my own experience as a former student of statistics & computer science, I chose to root my approach to detangling the technology of large language models in the creation process of the training data they emerge from. Where does this data come from? How can simpler versions of machine learning illustrate the type of information computers can interpret? How do these requirements inform how text is turned into data that computers can use? And finally, what challenges does this process pose to responsible use of generative AI tools, and how can students navigate them?

Those teaching AI literacy to a general audience will come away from this presentation with the rationale for and results of one approach to building a foundation for responsible AI use.