The self-teaching brain
Our goal is to uncover the learning algorithms that subserve biological intelligence and to discover how they are implemented in the brain. We take for granted that biological intelligence results from neural information processing, that neural information processing is based on the transmission of action potentials through synapses, and that learning is realized through synaptic plasticity. We are inspired by two key observations: Firstly, we know that biological learning unfolds in ways different from mainstream machine learning that relies on learning from large labeled datasets. Second, we discovered that the engagement of the brain during play can result in unexpected and profound cognitive benefits. This proposal describes an untravelled route to the learning algorithms of the brain that runs through the no-man’s-land between synaptic physiology, systems neuroscience, cognitive neuroscience, theoretical neuroscience and machine learning. Our approach focuses on the self-teaching abilities of the mammalian brain and covers and connects four major topics: (1) the objective functions that govern synaptic plasticity, (2) the teaching signals through which learning is steered, (3) behavioral mechanisms of self-teaching, in particular play behaviors, (4) the brain states that engage self-teaching behaviors, in particular the brain state of play. The BrainPlay grant will study self-teaching abilities from synapses to brains, from computational theory to action video games. As gaming has been shown to be highly beneficial for human brain function, we are intrigued by how little we know about what is going on in playing brains and how the brain state of play shapes learning. Engaging the latest theoretical and technological breakthroughs, BrainPlay will reach far beyond mainstream neuroscience and embrace and elucidate playfulness and self-teaching as important components of the brain's learning algorithms.