Why is it difficult?

We are surrounded by autonomous bots and machines that help us in many ways in our daily lives. Can we make these machines more thoughtful and intelligent and as efficient as humans and animals? Human intelligence comes from observing the world from an early age and a common-sense about physical phenomena such as gravity and prominence. On the other hand, a machine trained with supervised learning would require tons of data to learn some specific task and might still not work as efficiently as humans. Self-supervised learning aims to learn machine common sense by observing subtle patterns in common tasks and activities but requiring orders of magnitude more data. It is hard to build machines that work like humans or animals with similar goals and efficiency.

What is the impact?

Solving this challenge would lead to the development of machine common sense. And these machines could help humans in better ways. Many applications of machine common sense in natural language processing, speech recognition, and computer vision will cause a revolution in the next couple of years. Like humans, machines could communicate naturally with people and learn from new experiences and physical phenomena. Machines could simulate a model and devise a sequence of actions to take, for example planning the trajectory of a rocket.