Building to Understand: Examining Teens' Technical and Socio-Ethical Pieces of Understandings in the Construction of Small Generative Language Models

A research team from the University of Pennsylvania, led by Luis Morales-Navarro, Yasmin B. Kafai, and Danaé Metaxa, published a study examining how hands-on AI construction impacts teen literacy. The team conducted a week-long participatory design workshop where sixteen teenagers built their own very small generative language models (LMs). These models were designed to generate creative content like recipes, screenplays, and songs, moving teens from passive users to active creators of the technology.

Using thematic analysis, the researchers mapped the specific 'pieces of understanding' teens developed. On the technical side, this included grasping concepts like training data, model outputs, and the iterative nature of AI design. Socio-ethically, teens engaged with critical issues such as bias in training data, the environmental impact of model training, and the societal implications of generated content. The study contributes a new, theory-backed framework for analyzing how novices build comprehension of complex AI/ML systems through constructionist learning, a significant shift from purely theoretical education.