Towards a Comprehensive Ontology for Requirements Engineering for AI-Powered Systems

Context and motivation: Artificial intelligence (AI) provides computer systems problem-solving and decision-making features mimicking human behavior. As AI becomes widely adopted, AI-powered systems become increasingly ubiquitous. Requirements engineering (RE) is fundamental to system development, including AI-powered systems, which provide novel RE challenges. Question/problem: Developing means for addressing these challenges, which include increased need and importance of specifying and addressing social requirements, (e.g., responsibility, ethics, and trustworthiness); achieving a comprehensive understanding of all RE aspects, given the substantial growth in the diversity and complexity of requirements and the emergence of newand often contradictory ones; and, employing relevant methods and techniques that are suited for addressing these challenges. Principal ideas/results: We propose anRE4AI ontology as a first step toward addressing the above challenges. The development of the ontology was based on a meta-synthesis of relevant publications for identifying recurring themes and patterns, resulting in a set of themes categorized into RE stages, topics, stakeholders' roles, and constraints that formed the developed ontology. Contribution: The ontology provides a systematic and unambiguous representation of the accumulated RE knowledge about the system, including requirement themes, relationships between requirements, constraints, and stakeholders needed in the RE process. This ontology provides the basis for a complete AI RE methodology (AI-REM) framework that will incorporate methods to develop and manage AI-powered system requirements.