Class meeting: Mondays and Thursdays, 11:45am – 1:25pm
Instructor: David Smith (Office hours: TBA; WVH 356 or via zoom)
A common mode for understanding artificial intelligence systems, from popular fiction to textbooks in computer science, has been the metaphor of an agent that perceives, forms beliefs about, and intervenes in the world. In the past year, some scholars have instead framed large pretrained language and vision models as cultural technologies (Gopnik; Farrell and Shalizi, 2023). Other researchers have pointed out that the builders of large AI models must taken on some curatorial tasks in order to be successful and should learn from archival practice (Jo and Gebru, 2020).
In this seminar, we will read and discuss papers addressing large language and vision models as tools to investigate human language, history, and culture; papers analyzing and auditing corpus creation for model training; and papers exploring and mitigating biases and gaps in the archives of the past. Students will take turns presenting and leading discussion of papers along with the relevant background material. All students will write short reviews of the papers we read and work on writing research papers on a topic of their choice.
There are no official prerequisites; however, it is expected that students have some background either in NLP, computer vision, or other machine learning field, or in working computationally with large collections of text and images in the humanities or social sciences.
Each week, we will read roughly two papers on a common theme. The papers could be tied together by methodology—e.g., model or inference method—or by subject matter.
A list of the first few sets of papers is forthcoming. Further readingas will be added by input from seminar participants.