Language Use in Humans and AI: Intention and Convention

Most humans speak daily a particular language like English or German. We make certain sounds, thereby use sentences of that language with their meanings and as a result say something, ask a question or tell someone to do something. And we typically do this to perform further acts of social-communicative significance: we make claims, query people for information, give orders and make requests, make promises and give consent. Understanding human language use across the three levels of use, meaningful use, and social-communicative acts is essential for answering most of the central questions in philosophy of language. For example: what are words, sentences, and other expressions? What is meaning and what is the relation between meaning and use? How should we think of particular social-communicative acts like assertions, orders, and promises? Recent years have highlighted the possibility that humans are not the only language-using things. In November 2022 OpenAI released a chatbot called ChatGPT which took the world by storm. Chatbots like ChatGPT respond to user prompts by producing outputs that at least look like texts in English and other languages and are very impressive and usable. This generates the strong appearance that chatbots use language. But the current consensus is that LLMs lack mental states like beliefs or intentions and knowledge of language. Given that human language use is widely taken to depend on such states, many find it natural to draw the skeptical inference that chatbot language use is a mere illusion. Understanding whether this is true is important for deciding on how we should think of our interactions with them and for assignment of moral and legal responsibility when their outputs are problematic. This project will study human and AI language use across the three levels. The guiding hypothesis in the case of human language use is that views that attempt to answer questions about language use by appealing only to intentions or only to conventions or rules are problematic for analogous reasons. The most promising view at each level is one which combines intentions and rules in the right way. The role of the intention is to make it the case that the rule applies to ones lower-level act (e. g. a sound, use, saying) the role of the rules applying is to make it the case that it counts as the higher-level act (e. g. a use, meaningful use, social-communicative act). The guiding hypothesis in the case of chatbot language use is that the skeptical inference from their lacking mental states to their language use being an illusion is too hasty. Even if chatbots dont have intentions, it might be something else that makes it the case that the relevant constitutive rule applies to their outputs, making their lower-level acts count as higher-level acts.