[ExI] Scientists Discovered ?Mini-Computers? in Human Neurons?and That?s Great News for AI

[Stuart LaForge]

Quoting John Clark:


> Landauer's law can't be responsible for the high energy usage of the  
> brain, dendrites are not nearly small enough or fast enough for it  
> to be important. At body temperature Landauer says you could erase  
> 10^11 bits a second and use only 2.8* 10^-10 watts of energy.  And  
> remember signals in the brain only travel at a few hundred meters a  
> second. Even our best microprocessors are much too large and much  
> too slow for Landauer's law to be important, although in about 20  
> years that could change and we'll have to start thinking about  
> reversible computing.

Landauer's law is responsible in that it is defined as the lowest  
possible energy it would take to erase a bit in a 100% efficient  
computer. That is why it is expressed as greater-than-or-equal-to.

That it costs energy at all to erase information is a consequence of  
Landauer's law. It is like a thermodynamic limit and it costs  
biological brains more to erase bits because they are kludgy  
naturally-evolved processors that uses selective movement of ions  
across membranes rather than a simple current of electrons to perform  
calculations.

Action potentials happen in solution and involve moving entire atomic   
species around in potential fields instead mere electrons so it is  
bound to be many times more energetically costly than an electronic  
computer.

Likewise using an abacus to calculate your mind instead of a  
compartmentalized ion concentration gradient would take even more  
energy and be even slower than your naturally-evolved brain.

Perhaps I should have specified that a combination of Landauer's law  
and thermodynamics is responsible for the brain's high energy usage.  
But I admit I could be wrong. Seems like it should be a testable  
hypothesis.

Stuart LaForge