Abstract:
Large language models are transitioning from static text processors to autonomous, tool-using agents that can plan, execute, and iterate across the research lifecycle. In contrast to a chatbot, agentic AI reads and edits files in your project, runs your code on your computer, fixes errors, and carries out long-running multi-step plans. I review current approaches to using agentic AI to augment routine parts of the social science research process that happen on a computer. I will demonstrate how to use open source software and free or low cost models (avoiding vendor lock in). I will review tools, skills, and model context protocols (MCP) useful for social science research. This approach will be most useful for researchers already using R or Python with small to medium sized data in Quarto or Jupyter notebooks. This is a human-at-the-helm approach (rather than an end-to-end automated paper-writing system) where you are steering the agent and you are responsible for all decisions. I review emerging ethical questions, implications for training students, and how we understand the research process.
Ben Marwick is an archaeologist focussing on palaeolithic technologies and human behaviour in East and Southeast Asia. At UW he directs the interdisciplinary Data Science Minor. https://faculty.washington.edu/bmarwick/
