Neuroprosthetics, immortality, & the Frankenstein scenario (was: Monkey brai

Eugen Leitl
Tue, 11 Jun 2002 18:24:34 +0200 (CEST)

On Tue, 11 Jun 2002, Russell Turpin wrote:

> Think chips, not computers. There is a lot of research

Computers are made from chips. Latest AMD has one order of magnitude 
higher power density than your kitchen oven plate. If you need crunch, 
you're going to burn power. Only a fraction of this is used for powering 
I/O pads.

> right now in how to make these kinds of things so they
> draw power from the body itself. I don't know how the

Implantable fuel cells metabolizing glucose have been prototyped. However,
we're talking about power density (power dissipated/volume). It's
irrelevant where it comes from. Of course, you won't be able to get even
10 W without making the drain very obvious. (If the implant and your brain
are fighting for serum glucose, your brain would lose).

> retina implants are powered, but they don't require
> a scheral portal. Remember: the end result of this

We're talking high bandwidith I/O. This has not been done before, and be
it because we don't have the technology. (Microelectrode arrays have 10^2,
not 10^6 pins, and they don't contain any DSP onboard, it's all external).

> prosthetic is to enervate neurons. That takes miniscule
> amounts of power, maybe no more than switching a
> transistor on a chip. Even the latest artificial heart,

Switching transistors on a chip takes a lot of power, if it needs to be 
done quickly, and there are a number of them.

> which has to do orders of magnitude more work, gets
> its power without a skin portal.

Where they don't use pneumatics they use an induction coil (specifically
for the reason to avoid having a portal in place). Would you like to sleep
with a recharging vest? Run around with a fuel cell vest, just to power
your implants?

> In normal use, there is no need for I/O through
> the skin, but just from nerve to prosthetic to
> nerve. Now 'tis true the prosthetic will need an I/O

Self-contained and self-powered cyborg implantables do not exist. We'll 
get them sometime, but we're decades away from that.

> interface for initial diagnostics and programming. One
> possibility is a temporary skin portal, which is not
> nearly so troublesome as a permanent one. I suspect,
> though, that this will be done wirelessly, without

NIR is an option, but the bandwidth is limited. Beaming power into the
body with induction coils is *ugly*. You don't want to go there.

> incision, and with the external unit supplying
> whatever additional power the prosthetic requires.

You should really check out how latest implantable hearts are powered, and 
what huge inconvenience it causes.
> >Meaningful I/O (>10^6 channels) to the CNS is a lot
> >of hardware.
> I don't think so. How many lines are in even a small
> microprocessor these days? Don't think about getting

Microprocessor? We're talking interfacing to the CNS. Make that rather 
>>10^6 channels.

> those signals out to some kind of device and then
> plugging a cable into the nervous system. Instead,
> think about getting nerves to attach themselves to
> channels on a silicon pad. Think that is ridiculous?

I think that this is what we have been talking about the whole time?

> It has already been done in the lab. Initially,
> these prosthetics won't have 10^6 channels, but
> 10^2. And yeah, that's not enough to walk or type.

10^2 with lots of DSP is enough for a cursor, and is probably enough to 
generate a text stream, albeit slowly.

> But when the alternative is complete paralysis of a
> limb, even a few simple movements and sensations are
> miraculous. Now yeah, that's far away from the end

I agree. But we were discussing augmenting technologies for nondisabled 

> result I described. But once these kind of
> neuroprosthetics start being used, Moore's law
> will kick in. Each new generation will double the
> number of neural connections.

I agree in principle, but for the reasons I cited it will be decades
before you can get an implant job that will make a difference as a normal
healthy adult.