The Ethics of Speaking (of) AIs Through the Lens of Natural Language

This theoretical essay offers a critical exploration of the ethics involved in interacting with and talking about large language models (LLMs) of artificial intelligence (AI). The discussion is framed within philosophical post-humanist conceptualizations of the ethical agent, which is understood as a sociocognitive assemblage of human-machine interactions at various scales. The central argument asserts that the morality of texts generated by AI cannot be determined by extracting moral properties from the natural language used in the training corpus. There are inherent limits to how much analysing moral language can contribute to establishing a moral theory or a "language of mores" among humans. The essay also examines the ethical implications of current public discourse surrounding the capabilities of LLMs, as well as the ways in which LLM outputs personify the AI model itself. It is proposed that the ethics of LLMs should be approached as an ethics of translating informational patterns of linguistic symbols into multi-layered cultural meanings and vice versa. This includes addressing the opacity of the inner workings of these translations in the model, as well as the public relations practices of the creators. Ultimately, the discussion encourages rethinking the ethical agent as a human-machine sociocognitive hybrid, suggesting the need for a reassessment of what it means to be human and ethical in current AI ethics debates.