John and I discuss how understanding behaviors first will improve understanding how brains generate those behaviors, the state of neuroscience, plus the role of philosophy in science, lots of resources along those lines, and lots more.
Tim and I discuss his work suggesting grid cell-like coding could provide a map for us to navigate abstract concepts in our minds – a cognitive map. We talk about the discovery of place cells and grid cells and the work suggesting they contribute to spatial navigation. We also touch on how Tim’s work is related to meta-reinforcement learning (a la Matt Botvinick, see episode 21), and how grid cell codes might contribute to AI systems moving forward.
Marcel and I discuss his recent work using generative adversarial networks to decode brain activity to reconstruct images people saw, his work to restore vision to blind humans by stimulating early visual cortex, general AI, and even — shutter — consciousness a bit.
Melanie and I talk about the limitations of artificial intelligence in its current deep learning state (a la her New York Times Op-Ed), what AI needs to proceed toward general AI, what complexity is and how it relates to the fields of AI and neuroscience, and plenty more.
Matt and I discuss his neuroscience and AI research at DeepMind, including how AI benefits from neuroscience, his work on meta-reinforcement learning to create systems that learn more efficiently, how meta-reinforcement learning might be implemented in a neural circuit involving the prefrontal cortex and the dopamine system, how theory of mind might be implemented in machines to help them understand each other and to help us understand them, and a lot more.
