Use of Synthetic Personas in Communication Research

Course Description

LLMs allow the creation of synthetic personas (aka simulated humans, virtual personas, AI personas, artificial personas, persona proxies, digital twins, virtual twins, virtual humans) that may be able to simulate human responses to media. Those simulations, if proven accurate, could change how communication research with and about media is conducted. Theory development and practical applications could be accomplished more quickly, more affordably, with more targeted samples, and perhaps even with greater measurement accuracy and breadth. The goal of this seminar is to experiment with this new research option, one that is gaining significant traction in academia and industry, and then evaluate and critique its utility in communication research. Via reading and experimentation, we will seek to develop a point of view about how and whether to embrace synthetic personas as a substitute or complement to current techniques for studying how media are processed and how media shape our lives.

There are numerous published reports about synthetic personas that we will read and evaluate, including articles in Science, Nature, PNAS, and leading specialty journals in communication, psychology, marketing, political science, computer science and business. Many of the reports are focused on how well synthetic personas can replicate identical experiments with human respondents. In addition, we will use the synthetic personas to explore generalization of research results to different persona samples, to new media stimuli and to different research designs and settings. In addition to replicability and generalization, we will evaluate how synthetic personas might be used to mitigate research challenges related to online data fraud, collection of sensitive personal data, IRB constraints, research about special human samples (e.g., children, people with specific medical diagnoses, unhoused people), research about sensitive media content (e.g., pornography, private medical information, personalized financial details), and people that are hard to reach (e.g., geographical dispersed groups, special job categories, unique psychographic profiles).

We will do actual experiments using synthetic personas during our seminar sessions starting with the first seminar session on Sept 24th. Access to synthetic persona tools will be arranged and some may wish to construct their own tools (there are papers that offer guidance). Important technical research about synthetic personas is being conducted by computer scientists and mathematical modelers. We will read a bit of that work, however, our seminar discussions and projects will focus on experiments and tools that use natural language.