Isn't it the case that human one-shot classification tasks can rely on a pre-existing corpus of concepts that span a richer and broader area? For example, some

Isn't it the case that human one-shot classification tasks can rely on a pre-existing corpus of concepts that span a richer and broader area? For example, someone who looks at a segway, won't just see the pixels of a two wheeled vehicle, and that it is axially unlike a bike, but they may have also read articles about it, and know that two wheeled vehicles without peddles or mechanisms for riders to stabilze them, imply the device has a self-stability function. Thus, a human being could group a segway or unicycle segway type device, with BB-8 style self-balancing/gyro balls that people can ride as chairs. But to do so, they are relying on reasoning about how the device must work, not just image recognition. I mean, I welcome all of these advances at getting better at recognition tasks, but I still feel like we're producing "Fast Brain" AI, not "Slow Brain" AI.