[ExI] Why space tech isn't cutting edge

[Anders Sandberg]

On 19/11/2012 21:30, Adrian Tymes wrote:
> Neurons are not as tiny as modern CPU interconnect
> wires, therefore single-atom events aren't nearly so
> disruptive.  Being highly parallel helps too.
>
> On Mon, Nov 19, 2012 at 1:01 PM, Giovanni Santostasi
> <gsantostasi at gmail.com> wrote:
>> How the brain does it?
>> How do we reboot from such high energy events? Is the brain affected by such
>> high energy events?
>> Is the brain more stable because of its high parallel architecture?
>>

Parallelism certainly helps. But the brain is likely full of 
error-correction too, since there is so much neuronal noise: action 
potentials do get lost along the axon, synapses fire a random number of 
vesicles as a response to a signal, ion channels open and close 
randomly, the number of ions and molecules involved in a cellular 
compartment can be down to a few hundred or worse. A cosmic ray ionizing 
something so a cell fires randomly will not be too different from the 
normal noise.

An action potential is around 10^-20 J, and we get around 2 nucleons 
with more than 1.6*10^-10 J energy per square centimeter steradian per 
second ( http://www.int.washington.edu/PHYS554/2011/chapter9_11.pdf ) 
Not certain about absorbption coefficients for such particles in the 
brain, but at least in space there will likely be a few misfirings per 
second due to local bumps.


BTW, I recently came across something *awesome*: Hinton et al.'s 
"Improving neural networks by preventing
co-adaptation of feature detectors" http://arxiv.org/pdf/1207.0580.pdf  
They reduce neural network overfitting *a lot* by randomly omitting 
neurons during training, making a far more robust network. Maybe this is 
a good reason for the brain to occasionally lose signals.


-- 
Anders Sandberg,
Future of Humanity Institute
Philosophy Faculty of Oxford University