Sarah E. Kreps | The Promise and Perils of AI-Mediated Political Communication

Smart replies, writing enhancements, and virtual assistants powered by artificial intelligence language technologies are increasingly being integrated into consumer products and everyday experiences. This research explores both the potential and risks of AI-mediated communication (AI-MC) technologies such as GPT-4—specifically in the political sphere—through a series of experiments designed to assess the possible uses and misuses of AI-MC. The initial part of the research evaluates whether human-AI collaboration can increase legislator responsiveness by studying citizen responses to AI-generated tweets and email correspondence. The findings point to the importance of disclosure, transparency, and human-in-the-loop accountability for AI-mediated political communication. The research then turns to the plausibility of misuse by actors seeking to influence the democratic process. It shares results from a field experiment on legislators, highlights the challenge these technologies present to democratic representation, and suggests techniques elected officials might employ to guard against AI-sourced astroturfing.