2025 Fall Seminar Series Seminar: Miklós Sebők poltextLAB Artificial Intelligence Laboratory

Using AI Assistants in Comparative Research: The Case of the Comparative Agendas Project and the Babel Machine

Abstract: The automated classification of political texts remains a fundamental challenge in comparative political research, particularly as the volume of digitized documents expands exponentially. The talk presents two novel AI-driven innovations that extend the capabilities of the Babel Machine, a multilingual transformer-based classification platform developed by poltextLAB. First, we introduce a fine-tuning agent that automates the model adaptation pipeline, enabling researchers to customize classification models through natural language interactions in Slack. By uploading datasets and specifying configurations, users can trigger automated validation, data strategy execution, and GPU-based training jobs without coding expertise. Second, we present RobotAssistant, a multi-model comparison framework accessible via Slack commands that simultaneously deploys classification tasks across multiple large language models, including Claude, GPT, and DeepSeek. This tool enables systematic benchmarking of model performance for Comparative Agendas Project (CAP) classifications and other political science taxonomies. As deployed through the Babel Machine, these tools contribute to the democratization of access to state-of-the-art classification capabilities, enabling researchers without computational expertise to develop custom models and conduct systematic model comparisons. The infrastructure represents a significant advance in making gold-standard machine coding accessible for comparative research, addressing both the technical barriers to model customization and the challenge of selecting appropriate AI models for specific classification tasks.