[fallingfrog]

The article seems to be about how it is that the brain can process so much information at so slow a cycle speed, but it doesn't really address power efficiency at all. Even assuming 100% parallel operation, our current chip designs use something like a factor of a million more energy to do the same amount of computation. I wish I knew why- it's not like we're ignoring power efficiency. Depolarization of an axon must be incredibly power efficient.

[dlwdlw]

I think it's about context. If I say "order pizza" a ton of information is transferred via shared environmental context. I'm not transferring information directly but manipulating contextual levers to map certain types of information.

An example is if I say something to you in say japanese, I also need to teach you japanese in the same phrase unless there is shared context. computers run on a very impotent form of language, logic, that is rigorous and general purpose but has no "escape hatch". (basically the sentiment of Godem, Escher, Bach with regards to rigorous systems. )

[fallingfrog]

Oh ok you're talking about data compression. That's also not what efficiency means, in a strict sense.

[fallingfrog]

A litte online research reveals that the depolarization potential is on the order of one or two millivolts, whereas silicon has a bandgap of 700 millivolts. That's plausible; since power is proportional to voltage squared, a 500 fold voltage improvement would equate to a 250000 fold power improvement.