Actionable Aspects of Agency

This paper proposes a comprehensive classification scheme for identifying agentic entities based on six essential aspects: perception, predictive capacity, decision-making capability, goal-content integrity, resource input and output, and manipulators. By systematically examining these aspects, this paper aims to distinguish true agents from nonagentic entities, including advanced tools and collective entities. The classification scheme addresses practical implications, such as legal liability and moral considerations for AI models and the potential identification of extraterrestrial life forms. Perception is identified as the foundational aspect, enabling the acquisition and processing of environmental information. Predictive capacity and decision-making capability allow entities to anticipate outcomes and select optimal actions, respectively. Goal-content integrity ensures sustained commitment to intended goals, while resource input and output highlight the necessity of resource management. Manipulators are essential for the physical realisation of intentional actions. The paper emphasises the interconnectedness of these aspects, demonstrating that their integration is required for an entity to be classified as an agent. This framework provides valuable insights for diverse fields, from legal and ethical discussions surrounding AI to philosophical debates on personhood and agency, enhancing our understanding of intelligent entities across various contexts.