The computational efficiency in brains comes from spatio-temporal sparsity and recurrent feedback, features typically missing from âoedeep learningâ AI. The difference between analog and digital is merely an implementation decision. Airplanes donâ(TM)t flap their wings, and itâ(TM)s unlikely that analog computing will make a comeback. In advanced semiconductor processes, analog circuitry is much larger and less well optimized than transistors used for digital circuity, partly because that is what has been optimized for, but also due to increasing variation in manufacturing which makes analog circuits difficult to calibrate, tune, and debug. There will definitely be applications for neuromorphic systems, but I predict they will be digital. Like Intelâ(TM)s Loihi. Thereâ(TM)s is much work left to do in figuring out useful computational structures.

Starship has 100X more payload capacity to the lunar surface than Apollo. The Lunar Module of the Apollo program could deliver only about 1 ton of payload to the lunar surface â" 2 people, life support, and a rover. The confusion (which Iâ(TM)ve corrected in other articles) is that Saturn V launched 50 tons âoetowards the moonâ. This included the entire command and service module and lander. But barely any of that could be delivered to the surface of the moon. Starship will drop off 100 tons of whatever you want, then fly back to earth for another load. Of course that requires multiple refilling launches of a rocket bigger than Saturn V, so the capacity increase makes some sense!