[ExI] The Parallel Man

Tue Oct 11 16:03:53 UTC 2011

On Tue, Oct 11, 2011 at 07:04:22AM -0700, Ben Zaiboc wrote:
> Eugen Leitl <eugen at leitl.org> wrote:
> 
> > This is a classical in vivo incremental uploading
> > scenario.
> 
> Hmm.
> 
> > Notice that the hardware will have considerable bulk and
> > power
> > footprint, and will utilize numerical methods we do not
> > yet
> > have.
> 
> OK, possibly.

If you're using one probe for each cubic micron (arguably, not
enough by far) you'll need 10^15 of those. Notice that these
are active probes, with local communication and computation for
crosscorrelation measurements. You'll be processing at least 10^18
data points per second. Your raw data uplink (=out of your noggin
to an external piece of hardware) rate is exasample/s, 
uncompressed. You're looking at I/O rate of at least PByte/s,
processed. Unless you want to do it all in situ, which is what
you probably don't want to do -- you've already more than doubled
your brain volume, after all, and look roughly like

http://hydrocephalusfoundation.org/wp-content/gallery/hydrocephalus-babies/hydrocephalus%20kid%2015.jpg

And that is the *easy* way of doing it, not the hard way like
you're suggesting.

>  
> > Sorry, this is bogus. Instead of active grid probe spaced
> > every micron
> > apart or less and in-situ online processing you're suddenly
> > dealing
> > with an unphysical load of I/O. I/O is OPS/s, it takes J/s,
> > and the
> > infrastructure for it is both bulky and invasive.
> 
> I'm not sure if I'm just not expressing the idea very well, or if I don't actually understand what the idea is in the first place!
> 
> Forget the 'shutting off' idea, that just muddies things.

You're building a shadow infrastructure in situ, building a computational
model capable of tracking (and short-term forecasting) of relevant
dynamics of underlying wetware. Once you're done, you shut down
your primary (to avoid diverging), and live happily ever after.

> I am admittedly assuming that it's neural spikes that are important, and disregarding things like chemical gradients.  Maybe that's naive, maybe not.  It's still a worthwhile experiment, I think, even if it just established that the assumption is wrong.

That particular rabbit hole goes damn deep, the longer you look into it.

>
> At the risk of people rolling their eyes and saying "Yes, we *know* what you mean, Ben, but it still won't work!":
> 
> Imagine each neuron has a molecule (or molecular assembly, trans-membrane protein, whatever), that changes state when the neuron depolarises, in a way that is detectable to an external scanner (so this is a passive signal, no energy is needed for transmission), and in a way that uniquely identifies the neuron.  A model of neuron firings could then be built up in an external system that is very accurate, and could be said to be a passive copy of the functioning of the brain.

Yes, this is what we're talking about. You span the coordinate
system by smart probe spacing.

> Also imagine a second mechanism, maybe an ion channel, that can reliably depolarise the cell when it's activated.  This also knows which cell it's in, and only activates in response to a specific coded signal from outside the brain (oscillating magnetic field or whatever).

I recommend you look into how MRI works. Then compare the
numbers with those estimates I posted above. As NMR voxel
size and gradient rate formation is rate-limiting, you're
unfortunately SOL.

> So, would these combined mechanisms constitute a practical two-way neural interface?  (forget mind emulation, etc., just concentrate on the idea of an interface).  I understand that the external equipment would be bulky, and it might not be practical except in a purpose-built room, with the subject immobile in a special apparatus, etc.

Interface is one thing, uploading is another thing entirely. 
You'll do fine with brain surface pickups if all you want to
do is I/O.

> Or, would just one of them work?  Even a one-way interface would be well worth the trouble.
> 
> > 
> > Meanwhile, people are dying, and turn to carbon dioxide.
> > Unnecessarily so.
> > A classical case of misplaced priorities.
> 
> Sorry, you've lost me altogether here.

Many die, few are cryopreserved. All the thinking you're doing
now won't save you from becoming carbon dioxide. It would seem
such an awful waste of human life.

>  
> > Neurons are not the objects you're looking at.
> 
> Is this a Jedi mind-trick?  Not sure what you mean.  I shouldn't be looking at neurons?  This technique won't be looking at them?

You should be looking at tiny parts of neurons. The unit of
computing is more a synapse with an associated piece of the
dendrite tree.

> 
> > Don't spend too much time on this. It's not going to work.
> 
> Well, it's my time to spend.  I want to at least understand /why/ it's not going to work, if it's not.  And maybe thereby get an idea of what might work.

Ask your actuarial table when you're going to die, then think
whether above will help you. I'm fairly sure it won't.

-- 
Eugen* Leitl <a href="http://leitl.org">leitl</a> http://leitl.org
______________________________________________________________
ICBM: 48.07100, 11.36820 http://www.ativel.com http://postbiota.org
8B29F6BE: 099D 78BA 2FD3 B014 B08A  7779 75B0 2443 8B29 F6BE