A Case for Competent AI Systems $-$ A Concept Note (2312.00052v2)

Abstract: The efficiency of an AI system is contingent upon its ability to align with the specified requirements of a given task. How-ever, the inherent complexity of tasks often introduces the potential for harmful implications or adverse actions. This note explores the critical concept of capability within AI systems, representing what the system is expected to deliver. The articulation of capability involves specifying well-defined out-comes. Yet, the achievement of this capability may be hindered by deficiencies in implementation and testing, reflecting a gap in the system's competency (what it can do vs. what it does successfully). A central challenge arises in elucidating the competency of an AI system to execute tasks effectively. The exploration of system competency in AI remains in its early stages, occasionally manifesting as confidence intervals denoting the probability of success. Trust in an AI system hinges on the explicit modeling and detailed specification of its competency, connected intricately to the system's capability. This note explores this gap by proposing a framework for articulating the competency of AI systems. Motivated by practical scenarios such as the Glass Door problem, where an individual inadvertently encounters a glass obstacle due to a failure in their competency, this research underscores the imperative of delving into competency dynamics. Bridging the gap between capability and competency at a detailed level, this note contributes to advancing the discourse on bolstering the reliability of AI systems in real-world applications.