On the Ethical Considerations of Generative Agents

The seminal work by Park et al. introduced the Generative Agents framework for simulating human behaviour using generative language models. Generative agents have the ability to operate independently and creatively, making decisions to reach a goal with only simple suggestions injected at initialisation. To do this, each generative agent maintains a distinct and dedicated set of memories and behavioural parameters that are provided to a generative language model as needed. By recording memories, prompting reflection on them, and carefully planning future actions or reacting to short-term observations, generative agents can produce reasonably realistic emulations of human behaviour and decision-making. In this position paper, we examine the existing literature on the ethical aspects of generative agents, highlight ethical concerns, and offer recommendations for their mitigation.