Thomas Manning
In the last decade, there has been a growing interest in advancements made in artificial intelligence via modern natural language processing techniques utilising artificial neural networks in the form of large language models. This has led to the emergence of conversational technologies like ChatGPT, which mimic aspects of the conversational processes that occur between humans. However, it is contended that the model of conversation proposed above is inadequate as a model of human-like conversation. Already, far simpler models are found in the context of autonomous agents and multiagent systems theory (AAMAS), whilst more primitive are better representations of the types of primordial interactions found before the development of conversation proper. But whilst these developments are on the right track regarding replicating the emergence of actual conversational systems, there already exists a preexisting methodology so far not utilised in the context of AAMAS that could be used which better replicate the kind of conversations we humans have with each other. Thus, conversation theory (CT)—which is a cybernetic theory of concept-forming and concept-sharing via conversation—is suggested in this paper as a systems-oriented and cybernetic design methodology that could be used to design and model conversational interactions in AAMAS. Unlike contemporary approaches to the design of conversational and conceptual systems, conversation theory is specifically concerned with the following: How the formation of coherent structures of self-reproducing topics—i.e., the concept—via conversation permits the convergence of shared understandings through positing and testing for analogy relations holding between one participant’s concept with another’s concept. One key benefit of utilising this approach for conversational design in the context of researchers working in AAMAS is that its insights are built upon (currently underutilised) literature from educational psychology and early childhood studies. Because of this, it is held in this text that conversation theory has untapped potential as a methodology for designing and analysing conversational interactions that involve mutual conceptualisation between discursive agents.
KEYWORDS: conversation theory, discursive agents, concepts, autonomous agents and multiagent systems