CT 2.0

CT has been the central rallying point for K-12 computing education at least since the early 2010s. Many teachers, school administrators, and policymakers have joined the movement. A consensus has emerged over the conceptual landscape of CT. Meanwhile, machine learning (ML) has triggered some major changes in many sectors of computing. Children’s lives today are full of ML-driven services—take TikTok’s spot-on recommendations, social media’s automatic tagging of their friends in photos, and targeted personalized advertisement, just to mention a few. Children cannot learn to think about and design ML technology from learning classical programming. ML is poised to upend the CT consensus. Look at some of the changes ML has already triggered in computing. It has enabled greatly improved speech and image recognition, powerful recommendations on streaming services, autonomous navigation of cars, super-human performance in board and computer games, and even alternative-reality “deepfake” videos. Most advances in topics above are due to hardware evolution to non-traditional, special purpose architectures, new algorithms such as convolutional neural networks (CNN) or generative adversarial networks (GAN), and new objectives and measures of success. We will show that several key CT concepts, including debugging, problem-solving workflow, correctness, and notional machines, are insufficient for ML and need to be extended. Moreover, ML introduces new concepts including neural networks, curating and training data, and reinforcement learning that are not part of CT at all. All these changes challenge the traditional views related to teaching CT in K–12. ML is not the only emerging technology appearing in the computing landscape. Quantum computing and biological computing are not far behind. We need to start rethinking how CT must evolve to anticipate and meet these challenges.