Yann LeCun, Facebook AI Research & New York University, New York, NY
Deep learning has caused revolutions in computer understanding of images, audio, and text,
enabling new applications such as information search and filtering, autonomous driving,
radiology screening, real-time language translation, and virtual assistants. But almost all these
successes largely use supervised learning, which requires human-annotated data, or
reinforcement learning, which requires too many trials to be practical in most real-world
situations. In contrast, animals and humans seem to learn vast amounts of background
knowledge about the world through mere observation and occasional actions in a selfsupervised manner. Making progress in self-supervised learning is the main challenge of AI
for the next decade. Success may result in machines with some level of common sense. But
they will be built around deep learning architectures that are considerably larger than current
ones, requiring vastly more powerful hardware than what we have today.