From Form to Ratification: Reassessing AI “Communication” After the Facebook Chatbot Episode

This article revisits the 2017 Facebook negotiation-bot episode, widely reported as a case of artificial intelligence “inventing a language,” to ask a narrower and more sociologically precise question: under what conditions should machine output count as language rather than as task-conditioned symbol coordination? Methodologically, the article is a narrative review and conceptual synthesis organized in line with the Scale for the Assessment of Narrative Review Articles (SANRA). It draws on purposively selected work in developmental psychology, philosophy of mind, sociology, and recent research literature on large language models (LLMs). The analysis proceeds in three steps. First, it shows that the Facebook case involved a reinforcement-learning negotiation agent and cannot be generalized to contemporary LLMs without serious qualification. Second, it argues that current debates about AI “understanding” are best read not as a choice between hype and dismissal but as disputes about different kinds of competence. Third, it develops a sociological distinction between form, coordination, and ratification. Contemporary AI systems can generate linguistically legible forms and sometimes support limited coordination, but language in the stronger sense depends on public uptake, normative correction, and institutional durability. The article’s contribution lies in shifting the debate from inner-state attribution to the social conditions under which signs become language.