Position: Levels of AGI for Operationalizing Progress on the Path to AGI

We propose a framework for classifying the capabilities and behavior ofArtificial General Intelligence (AGI) models and their precursors. Thisframework introduces levels of AGI performance, generality, and autonomy,providing a common language to compare models, assess risks, and measureprogress along the path to AGI. To develop our framework, we analyze existingdefinitions of AGI, and distill six principles that a useful ontology for AGIshould satisfy. With these principles in mind, we propose "Levels of AGI" basedon depth (performance) and breadth (generality) of capabilities, and reflect onhow current systems fit into this ontology. We discuss the challengingrequirements for future benchmarks that quantify the behavior and capabilitiesof AGI models against these levels. Finally, we discuss how these levels of AGIinteract with deployment considerations such as autonomy and risk, andemphasize the importance of carefully selecting Human-AI Interaction paradigmsfor responsible and safe deployment of highly capable AI systems.