Tianyi Peng, Melanie Brucks, George Gui, Daniel J Merlau, Grace Jiarui Fan, Malek Ben Sliman, Eric J Johnson, Abdullah Althenayyan, Silvia Bellezza, Dante Donati, Hortense Fong, Elizabeth Friedman, Ariana Guevara, Mohamed Hussein, Kinshuk Jerath, Bruce Kogut, Akshit Kumar, Kristen Lane, Hannah Li, Vicki Morwitz, Oded Netzer, Patryk Perkowski, Olivier Toubia
Scientists and practitioners are aggressively moving to deploy digital twins-large language model (LLM)-based models of individuals-across social science and policy research. We conducted 19 preregistered studies with 164 diverse outcomes (e.g., attitudes toward hiring algorithms and intention to share misinformation) and compared human responses with those of their digital twins (trained on each person's previous answers to more than 500 questions). We establish an empirical benchmark for digital twin performance: Digital twins' answers are only modestly more accurate than those from the (homogeneous) base LLM and correlate weakly with human responses (average correlation coefficient of 0.20). To guide future development, we document five ways in which digital twins distort human behavior: (i) insufficient individuation, (ii) stereotyping, (iii) representation bias, (iv) ideological biases, and (v) hyper-rationality. We make our full dataset and code public as a standardized testbed. Our results caution against premature deployment while laying the groundwork for the transparent, replicable, and iterative science necessary for responsible deployment of digital twins.