ANN-based AI needs more macro architecture
yes... fortunately it's still mostly just hype, and some say the machines beating humans at various contests are often ugly kludges. I do fear the actual arrival of a capable human-level AI. And, it seems to me that the AI people (including the deep learning folks) are still somewhat clueless as to *how* consciousness should eventually emerge. No matter how much pattern-recognition power you throw at it, that alone won't result in a motivated, focused thinking machine. I would expect to see some cunning "macro architecture", possibly involving some cognitive loopy/feedback circuit, and generally macro-topologies faintly similar to a Von Neumannean CPU: search the huge memory for concepts, select one, keep it in temporary storage, search for further concepts, select some, keep them in temporary storage, search for more, run like that in cycles, with some general drives / motives / goals / aux inputs behind this basic loop (working as a filter in the selection process), with some generative/action outputs etc. Something along those lines, just possibly an order or two more complicated, multi-layered and whatnot. I actually feel that many of the basic concepts and functionality of this "sapient macro-architecture" could be tested / proven on a much smaller scale.
Biting the hand that feeds IT
