This podcast discussed the current dominance of neural network research in AGI. A theory-based approach to AGI is suggested which can only work for insanely powerful or impractical machines.