Can We Teach Machines to Think?

Recent advances in machine intelligence have allowed artificial systems to achieve near-human levels of performance on tasks that involve classification of sensory information, such as object recognition. But humans can do much so more – we can think and act creatively, make detailed plans for an uncertain future, and engage in the reasoning about cause and effect that underlies progress in science and technology. I will address the question of how we might build machines that display this level of intelligent behaviour. I will argue that building machine architectures that incorporate concepts from cognitive psychology and neuroscience – including hierarchially ordered sensory systems, attention, working memory, episodic memory, and mental simulation – may be key to achieving this goal. I will offer examples from prominent recent advances in machine learning and AI. Finally, I will discuss the promise and pitfalls of this research for human progress in the 21st century