Is ChatGPT the way toward artificial general intelligence

Abstract The success of the conversational AI system ChatGPT has triggered an avalanche of studies that explore its applications in research and education. There are also high hopes that, in addition to such particular usages, it could lead to artificial general intelligence (AGI) that means to human-level intelligence. Such aspirations, however, need to be grounded by actual scientific means to ensure faithful statements and evaluations of the current situation. The purpose of this article is to put ChatGPT into perspective and to outline a way forward that might instead lead to an artificial special intelligence (ASI), a notion we introduce. The underlying idea of ASI is based on an environment that consists only of text. We will show that this avoids the problem of embodiment of an agent and leads to a system with restricted capabilities compared to AGI. Furthermore, we discuss gated actions as a means of large language models to moderate ethical concerns.