[ExI] Smallest human-equivalent device

[Anders Sandberg]

On 2013-10-11 10:39, Eugen Leitl wrote:
> On Fri, Oct 11, 2013 at 09:49:30AM +0100, Anders Sandberg wrote:
>
>> merging. But other software will disagree. And the uploads in my big
>> data center will not care that they are sharing underlying hardware
>> and software as long as their data and processes are kept in
>> separate virtual machines.
> This is one of the rare cases where you're mistaken, Anders.
> All the other points are spot on.
"Any lapses in Omniscience are the price I pay for being implementable." :-)

> Context switches make things slower, if your system is
> 100% loaded. The more state you have to save and restore,
> the slower. In a simulation, it's all state.
>
> This is why supercomputers use batching, running until
> done, and never multitask between jobs.

Good point. And shifting around 10^21 bits non-locally is a hassle, so 
you do not want to move your uploads between servers too often. I think 
Robin makes the same point in his manuscript.

Still, I wonder how much can be done using superposed states. SIMD taken 
to the logical extreme.


> Fast and energy-efficient means there's risk.
Energy efficient also means less error correction (for a given speed).


-- 
Dr Anders Sandberg
Future of Humanity Institute
Oxford Martin School
Oxford University