Stephen Fleming: Metacognition

It’s human to know oneself. We are able to self-monitor, understand our cognition, and recognize gaps in our knowledge. This is called metacognition—we think about how we think. We can think of it as self-awareness or the ability to understand the state of our knowledge. In this episode, we talk with Stephen Fleming, Professor of Cognitive Neuroscience at University College London.

Steve has recently published a book on this topic called Know Thyself. We wanted to explore a number of ideas with him—metacognition as uniquely human, as an important skill, whether machines need to have some form of awareness and the issue of agency and machines.

In his book, Steve proposes that we either put self-awareness into machines, thereby reducing our need for self-awareness, or design interfaces that increase human self-awareness. If we have self-aware machines then we risk losing our own. If we want better self-awareness we must prioritize how an AI metacognition can show us its uncertainty and error. This second route is more likely to result in humans retaining autonomy. It preserves the human role of wrestling uncertainty, seeking explanations, and making sense of the world. Perhaps the biggest insight for us regarding agency and AI is how to think about it as having a structure. We talk about how true autonomy is aligning our choices and wants and how machines play a role.