Harnessing Artificial Intelligence to enhance the power of experimental social science research to improve the human condition
About our project:
Talking to Machines
INTRODUCTION TO TTM
“Talking to Machines” is a project that aims to leverage artificial intelligence in the design and implementation of experiments. This project is a unique combination of both academic and industry researchers working in very diverse institutional contexts throughout the world. We are an international collaboration with a talented global team of faculty, staff, postdocs, and graduate and undergraduate students. Research results from these studies have been published widely in leading social science peer-reviewed publications.
Through our commitment to evidence-based policies, we engage with national and international governmental agencies to evaluate the effectiveness of current policies and design new programs and tools with the potential to ultimately affect the lives of millions of individuals.
The project is led by Sonja Vogt at HEC University of Lausanne and Ray Duch at Nuffield College University of Oxford. Our primary funding is from the Swiss National Science Foundation (#100018M-215519).
Our main objective is to advance social science research through innovations that harness recent breakthroughs in AI. We believe AI-driven solutions can address many of the challenges researchers face in designing and conducting experiments.
Our Focus
The rapid advances in LLM technology offer scientists unique opportunities to re-think the feasibility, design and implementation of experimental studies ranging from online experiments to large ambitious randomized control trials. The TDM team has redefined experimental “subjects” exploring strategies for incorporating synthetic persona in the design and execution of experiments. We are also building LLM models that help craft the treatments that we implement in experiments – our current focus is on the information content of video interventions that are widely used in experimental research.
Diversity is the guiding theme for much of our work: The global south is the context for much of our experimental research – with this in mind we are developing AI-enhanced experimental tools that are built to reflect the cultural diversity of global south contexts. More generally, we are employing AI to help better characterize the heterogeneity of treatment effects for distinct segments of populations; to generalize estimates to a broader population; and to transport findings to different contexts.
The Talking to Machines project is developing open-sourced platforms that make these AI innovations widely available to the global research community.
Our Goal
We are champions of causal inference and evidence-based policy. The T2M initiative is dedicated to expanding the community of scientists who have access to the powerful tools of experimental research. Advances in AI offer numerous possibilities for increasing the efficiency of experimental research – reducing the costs and enhancing the power of experiments. Our remit is to identify these AI tools and methods; confirm their utility for experimental design, implementation and analysis; and build applications that make these open-source tools as widely available as possible.
To accomplish these goals we leverage the significant ongoing advances in Large Language Models (LLMs) and apply them to the various tools employed by the experimental research community.
Some of the current project themes include:
- Facilitating experimental interactions between AI and human subjects
- Developing digital twin technologies
- LLM-guided augmented RCT data collection
- AI-powered adaptive experiments
- Artificially intelligent public opinion polling
- AI-enhanced design of video treatments
