FAT-CAT—Explainability and augmentation for an AI system: A case study on AI recruitment-system adoption

초록

Because artificial intelligence (AI) recruitment systems exhibited discriminatory decisions in recent applications, the adoption of such systems in industry has raised doubts. As equity has been emphasized in AI decision-making frameworks, the non-explainability issue regarding the high performance of AI methods has become prominent. Therefore, scholars have focused on human–AI augmentation in which humans consider equity and AI supports the consideration. As a result, explainability is highlighted as a new capability of AI methods for an ideal decision. In this regard, this study proposes the so-called fairness, accountability, and transparency (FAT)-complexity, anxiety, and trust (CAT) model that describes the path from explainability to AI system adoption considering augmentation, assuming that the capability of the AI decision maker to explain the basis of its decision and interact with the human decision maker is crucial for AI recruitment system adoption. We found that explainability and augmentation are two key factors in AI recruitment system adoption and assessed that their importance will gradually increase as recruiters will be asked to use such AI systems more commonly. Moreover, this study conceptualized the role of an augmented relationship between humans and AI in decision-making, in which they complement each other's limitations.

키워드