Machine learning (ML) has triggered major changes across a great number of computing fields. People’s lives today are full of ML-driven services: eerily accurate recommendations, ability to automatically tag one’s friends in photos, and well working translation systems, for example. This keynote talk presents how ML technology upends the computational thinking (CT) consensus in computing education. It begins by presenting why and how a number of classical “CT1.0” concepts need to be re-thought for the “CT2.0” (machine learning) era, from control structures and problem-solving workflow, to correctness and notional machines. Based on a series of classroom interventions on teaching machine learning to middle schoolers, conducted by DIGS RC at University of Eastern Finland, the talk also presents how classroom pedagogy shifts between CT1.0 and CT2.0.