Research Scientist

About This Role

This role is for a Research Scientist to join the Digital Twins Team, which runs an ongoing research program to measure and improve how faithfully AI systems reproduce human responses and behavior. The scientist will work directly with the Head Scientist to design and interpret studies, define measurements, and write up findings for internal decision-making and academic publication. The position requires current methodological training in computational social science, data science, survey methodology, human-computation interaction, and experiment design.

Responsibilities

• Co-design and run studies that compare system-generated responses with responses from humans, using both qualitative and quantitative measures
• Design, justify, and hone measures of performance
• Apply and adapt state-of-the-art statistical techniques common to survey methodology and large-scale social data analysis, to benchmark and improve sample- and population-level simulation performance
• Design evaluations and experiments with pre-specified analysis plans
• Design and oversee human rating studies, including rater training, coding rubrics, and inter-rater agreement measurement
• Design data collection from third-party data sources and directly from human participants, including longitudinal and repeated-measures studies
• Contribute to the research agenda for new modeling approaches, including evaluation of fine-tuned models and of model calibration
• Specify analyses precisely enough for an engineer to implement
• Co-author papers, preprints, and presentations

Requirements

• A completed or nearly completed PhD in computational social science, communication, HCI, sociology, political science, psychology, statistics, or a related field
• Peer-reviewed publications on which you were a primary author
• Working command of graduate-level statistical methodology common to social science, including the assumptions behind each
• Experience analyzing large-scale telemetry and behavioral datasets, with attention to the validity of the resulting inference
• Familiarity with machine learning, spanning classical, nonparametric, and LLM-based approaches, including how transformer-based language models work, what fine-tuning does, and how the major model families differ
• Fundamentals of computing: file systems, software libraries, databases, SQL, and tabular data
• Ability to work in an environment where most code is produced through agentic coding tools, and to read and check that code

Nice to have:

• Experience fine-tuning open-weight language models on survey, experimental, or behavioral data
• Experience building or evaluating calibrated confidence or error-prediction models, including deriving decision thresholds from human judgments