SODAS Data Discussion #4 (Fall 2024)
Copenhagen Center for Social Data Science (SODAS) aspires to be a resource for all students and researchers at the Faculty of Social Sciences. We therefore invite researchers across the faculty to present ongoing research projects, project applications or just loose ideas that relate to the subject of social data science.
Two researchers will present their work. The rules are simple: Short presentations of ten minutes are followed by twenty minutes of debate. No papers will be circulated beforehand, and the presentations cannot be longer than five slides.
Discussion 1
Presenter: Hjalmar Bang Carlsen
Title: The Performance Accountability Tradeoff in AI Interviews - Comparing Open Source, Open Weight and Proprietary LLMs for Chat Based Qualitative Interviews
Abstract:
In this presentation, I discuss the performance and accountability tradeoff in AI Interviews. AI-interviews use LLM agents to act as interviewers that ask questions and follow-up questions to respondents in manners similar to that of the semi-structured interview. Most recent applications have used versions of Chat-gpt, a closed proprietary model, and arguably the best performing model. Our own efforts in the AI-interviewer project use open weight and open source models. Open models allow for better data security, transparency and model stability, ultimately allowing for a more accountable research practice. Ultimately, I'll argue that for AI-interviews, and social science applications more broadly, open models need to be the basis of our methodological developments.
Discussion 2
Presenter: Yevgeniy Golovchenko
Title: Sanction Evasion and Covert Maritime Networks
Abstract:
In the past decade, a global "shadow fleet" of vessels has emerged to evade sanctions, violating international maritime law and supporting smuggling activities that fund wars, such as Russia's invasion of Ukraine. These fleets, often non-compliant with safety regulations, pose significant environmental risks, exemplified by a recent tanker explosion off Malaysia. Despite these threats, shadow shipping remains understudied. This project conducts a systematic analysis of shadow fleet operations, using geospatial vessel tracking data to build a global oil tanker dataset. Focusing on Russia, which has constructed the largest shadow fleet to bypass EU and G7 sanctions, the study employs social data science methods to explore maritime sanctions evasion strategies. By bridging social data science methods and maritime security studies, the research fills a critical knowledge gap and demonstrates how ship tracking data can enhance social science methodologies for analyzing illicit maritime networks.