Core ML Concepts · intermediate · concept 38 of 311 · Updated Nov 4, 2026

🧭Data-Centric AI

Data-centric AI is the practice of improving a system by systematically improving its training data while the model code stays fixed. Andrew Ng popularized the term in 2021. It matters because most production gains come from fixing labels, filling coverage gaps and removing noise rather than from a new architecture. In Ng's steel defect inspection example, making the labels consistent raised accuracy where model changes did not.