[extropy-chat] Question on computational power of brain

[Anders Sandberg]

Jef Allbright wrote:
> We've seen estimates of the computational capacity of the human brain
> based on estimates of the number of synaptic connections and their
> nominal speed.  But this doesn't take into account an understanding of
> the usage patterns of those connections -- certainly they're not all
> working flat out.

The problem is how to define a computation. A neuron that is not firing,
is it representing a long string of zeros 00000000, not participating or
somewhere in between?

Cortical neurons seem to be in "up" states perhaps less than 10% of the
time, many probably only get activated so they can signal and participate
in certain specific conditions. The average activity is pretty sparse,
with firing rates <1% or so.

> Given that we know the energy consumption of the brain, and (assuming
> we know) an efficiency factor for this biological machinery, can we
> arrive at a reality check for the effective, rather than total,
> ballpark computational equivalency?

We don't know the J/ops efficiency of the brain. If we knew, it would be
very cool. The kind of estimates we usually do are bottom up, and pretty
vulnerable to assumptions.

Most of the energy cost comes from maintaining the membrane voltages
rather than the neural signal themselves. So the efficiency is pretty low.



-- 
Anders Sandberg,
Oxford Uehiro Centre for Practical Ethics
Philosophy Faculty of Oxford University