Ai and Emergent Abilities

Perhaps the most remarkable attribute of Large Language Models is their capacity to develop ‘emergent abilities’, that allow them to engage in often highly convincing, human-like conversation, translate languages, solve complex puzzles, and even design. These are totally unexpected abilities, acquired by Large Language Models in areas in which they have not been explicitly programmed, that arise implicitly as a result of their exposure to massive quantities of data during their training process. But are these ‘emergent abilities’ connected in any way to the theory of ‘emergence’ - a term used to describe non-linear, bottom up and unpredictable behaviors, that we find in complex, multi-agent systems? This paper examines whether we can understand these emergence abilities within the context of the theory of ‘emergence’. In conclusion, while there are clear parallels between the two terms, we do not know enough about either term to say conclusively whether they are related.