Abstract
As ChatGPT, a large language model, speaks like a human we set out to test not only whether it follows the Gricean maxims for rational communication, but also its formal and functional language competence when prompted to answer questions about real world topics. The questions posed included ethics, common sense, and issues in different cultural contexts. ChatGPT’s responses systematically reveal that its algorithm does not construct a model of the situation reflecting the information contained in the prompt; it tends to focus on a single aspect. By contrast, human knowledge in the form of social representations is constructed through communication, exchanges of questions and answers, supported by a theory of mind leading to the automatic understanding of the ‘other’, and a mutual understanding grounded in the same cultural setting. We conclude that as GPT does not think like a human it should not pose as one.