Promise and Pitfalls of Scale: Working with Machine Learning Models of Text

Dr. Serrano's (Computer Science) presentation is part of a semester of programming organized by the Hanson Center for Inclusive STEM Education on the theme "Can AI Generate IDEA's? Navigating the Promise and Perils of Artificial Intelligence for Inclusion, Diversity, Equity, and Access."

Abstract: Scale is crucial to contemporary large language models. What implications does that scale have for viable uses of these language technologies? Knowing that these models have biases, how can we leverage their scale in a safer way to accomplish nontrivial tasks? In this talk, Dr. Serrano will use a past project to illustrate working through these questions together with qualitative researchers studying power and historical political violence. She’ll walk through how their initial approach evolved to address the limitations of the language technologies they were using at the time, how those models’ technical biases informed their use on the project, and what they were able to use those models to accomplish.