Data-Centric Artificial Intelligence (link.springer.com)

A new framework for data-centric artificial intelligence (data-centric AI) has been introduced, shifting the focus from model-centric approaches that primarily enhance AI model performance to systematically improving and engineering data quality and quantity. This emerging paradigm, advocated by AI leader Andrew Ng, emphasizes the importance of selecting and refining the right data rather than merely increasing its volume. The paper delineates key characteristics that distinguish data-centric AI from traditional model-centric methods, providing a comprehensive overview of its implications for the Business and Information Systems Engineering (BISE) community. The significance of this paradigm lies in its potential to revolutionize how AI systems are developed, addressing a critical gap in existing practices where data quality is often overlooked. By advocating for systematic data work—such as refining datasets and addressing blind spots through thoughtful data extension—data-centric AI aims to improve machine learning outcomes in real-world applications. This inclusive approach not only fosters collaboration across various research fields but also proposes important methodologies and tools necessary for implementing effective data engineering practices, ultimately leading to more reliable and efficient AI systems.