Deep thinking and quick learning for viable AI
Seng-Beng Ho · 2016
Despite the vast progress made in artificial intelligence (AI) over the years and the recent renewed interest in it because of some major breakthroughs in methodologies seemingly signifying its general viability, there are still important gaps that have to be filled to enable the construction of truly general and adaptive intelligent machines. This paper points out that a useful general learning machine must not only be “general” in the sense of being able to learn to solve many different kinds of problem, it must also be able to learn to solve them rapidly, and continue to learn rapidly in an ever-changing, non-stationary environment. The paper reviews the current limitations in certain popular learning methods, such as reinforcement learning, and proposes new methodologies that address the gaps. A new paradigm of deep thinking and quick learning is proposed for a future direction of research to produce truly general and adaptive intelligent machines.