Michal Luria (Center Democracy and Technology) and Emilie Grybos (U Pennsylvania Annenberg Communication) have posted “Policy Considerations for Socially Interactive AI Agents: A Systematic Literature Review” on SSRN. Here is the abstract:
Although interfaces that use human communication, such as language and nonverbal behavior, to interact with users have been around for decades, the introduction of ChatGPT and other large language models (LLMs) to the public began a new era of LLM-based artificial intelligence (AI). As a result, the implementation of LLM chatbots and agents quickly expanded across applications and domains. In this paper, we conducted a systematic literature review on AI agents that socially interact with users, to identify research-based policy considerations in the academic discourse. This review provides a critical baseline for understanding policy concerns and priorities, and highlights challenges that remain unresolved. By analyzing 110 peer-reviewed publications that discuss various aspects of agent-related policy, we highlight the most prominent issues and how each uniquely applies to socially interactive agent technology: Data and Privacy, Safety and Security, Human Rights, Agent Rights, and Liability and Responsibility, Transparency and Explainability. We identify key gaps in existing research given current media and policy discussions, and conclude with recommendations for researchers and policymakers alike.