Designing Artificial Reasoners for Communication

In order to endow a conversational agent with sophisticated social intelligence, machine learning (which is prominent in LLM-based systems like Chat-GPT) is not enough. Logic-based reasoning and decision-making is needed. We need formal languages as well as reasoning and planning algorithms based on them for modeling and endowing the agent with intentional communication, theory of mind, explanatory capability and norm compliance. We identify some requirements that such languages should satisfy as well as a number of challenges regarding their combination and their integration with machine learning methods.