Synthetic data generated by large language models (LLMs) is rapidly transforming social science research. Advocates highlight its potential for faster, cheaper, less biased, and more privacy-preserving research, while critics warn of significant risks, including perpetuating biases, relying on outdated or overly simplified data, discouraging original data collection, and weakening transparency and data literacy. This session, led by Mikkel Krenchel and Luca Rossi, will explore both the opportunities and challenges, emphasizing how researchers can responsibly leverage synthetic data.

The session will focus on three key questions:

1. Why should we care about synthetic data? Why is it growing and where is it going?

2. When and how should—and shouldn’t—we use synthetic data? How can we reap the rewards without falling prey to the pitfalls?

3. What does high-quality synthetic data look like, and how do we evaluate it? LLMs can be black boxes, so how do we know what we can trust?

By addressing these questions, participants will gain a deeper understanding of synthetic data’s promise and pitfalls, enabling informed decisions in their research designs.

Speakers

Mikkel Krenchel is a Partner at ReD Associates, where he has spent the past 15 years helping leaders make better bets on human behavior. He focuses on how ideas around disruptive technologies shape our lives, advising some of the world’s largest social media companies, telecom providers, electronics manufacturers, and clients across the finance, energy, and industrial sectors. Mikkel has long been exploring the growth of AI tools and synthetic data in research. He and colleague Maria Cury wrote a primer on synthetic data in research that you can find here. His work has been published in Wired, Foreign Affairs, and VentureBeat as well as numerous academic outlets.

Luca Rossi is an Associate Professor at IT University Copenhagen, where he leads the Data Science Section and Human-Centered Data Science Research Group. His work connecting media and communication studies with computational approaches explores how digital technologies and social media impact complex social processes such as participation, activism, politics and, more recently, information propagation. He publishes widely in scientific journals and recently co-authored the The Problems of LLM-Generated Data in Social Science Research, a major synthesis of the use of synthetic data and ‘research participants’.