[extropy-chat] Singularity heat waste

[Martin Striz]

On 7/14/06, Robert Bradbury <robert.bradbury at gmail.com> wrote:
>
>
> On 7/14/06, Martin Striz <mstriz at gmail.com> wrote:
> > Heat might only be an issue if computation continues to be performed
> > by semiconductors.
>
>
> No, no, no.  This is not the problem.  You can "compute" effectively for
> *free*.  This area was resolved several decades ago by Bennett with some
> contributions from Landauer, Bremermann and Bekenstein.  It has nothing to
> do with whether or not one uses semiconductors.  It has to do with whether
> one is destructively erasing bits (throwing away information).  That is what
> produces the heat.  This is the difference between nonreversible computing
> and reversible computing.  In nonreversible computing you throw away lots of
> bits and produce lots of heat.  In reversible computing you run the
> calculation forward, save the result (producing a small amount of heat),
> then run the calculation backwards restoring things to their original state.
>  As has been pointed out you can't do such a calculation entirely for free
> (at least not quickly) -- but the energy lost to the computation process
> itself is many orders of magnitude below the energy lost when you erase
> bits.

Very well, let me clear up a few issues that I have.  I will offer the
caveat that my training is neurobiology, genetics and biochemistry, in
no particular order, but not computer science, so the information that
you've offered here is new to me.  Let me ask you a few questions:

1) Why do we use silicon for computation in most microprocessors, as
opposed to some other substrate?

2) Is there a more efficient substrate that we could use, what is it,
and why don't we use it?

3) If neurons are nonreversible (and they are), why are they so much
more efficient (in your view)?

> Merkle and perhaps others (Fredkin?) have shown that you can design
> reversible computing circuits using existing semiconductor fabrication
> methods.  However, the chips that have reversible capabilities are likely to
> require more gates and/or operate more slowly -- so they will not be
> implemented until chip manufacturers exhaust all other methods in their bag
> of tricks for minimizing or removing heat produced in current nonreversible
> designs.

4) Why do they require more gates and/or operate more slowly?

> Neurons can only be considered semi-reversible designs.  You don't have to
> regenerate the Na+/K+ ions in the brain but you do lose the energy needed to
> recharge the neurons after they fire.  That wasted energy shows up as heat
> and the rest of your body functions as a radiator for the brain.

No.  It takes a goot bit of energy to reinstate the action potential.
It doesn't show up as heat though because chemical transformations are
efficient.  Silicon literally RELIES on internal resistance to do
computation, which produces waste energy by its nature.  What I'm
asking you is why we use that to do computation in the first place.

> > Neurons can do a lot of computation within minimal
> > heat loss.  Your head isn't hot due to neuron inefficiency.  It's kept
> > hot on purpose because enzyme kinetics are optimized for 37C.
>
>
> Actually, the metabolism in the liver and perhaps intestines probably plays
> a more important role in keeping one at 37 deg.  Their fundamental raison
> d'etre is to produce the glucose which in turn is used by the brain to keep
> you alive long enough and figure out how to make copies of those genes you
> are carrying around.

Yes, neurons don't warm your head up.  They don't produce a lot of
waste heat, as I said. Your head is kept warm through fluid conduction
thanks to the vasculature.  The point is that there's a reason for
that: enzyme kinetics.  If you drop the system below ~34C, enzyme
rates slow down enough that the system goes kaput.  If you take it
above ~40C, enough enzymes start denaturing that that system also goes
kaput.  They are remarkably temperature sensitive (trust me, I have
loads of empirical evidence to back this up :), and that's why
thermoregulation matters so much to warm blooded species.

>  As arctic ground squirrels can be cooled very close to
> freezing and still restore themselves to normal functioning in the spring
> you would have a hard time convincing me that 37C is necessary for most of
> the enzymes in the brain.

Non sequitur.  Many species (such as cold blooded ones) have evolved
various optimizations for temperature tolerance, but that doesn't
extrapolate to humans or most mammals.  Golden mantled squirrels are
an exception.  You can cool a squirrel but it won't do much.  It won't
eat or drink or mate or clean itself.  It'll be more or less comatose.
 Can't do that for people, though.

> More likely 37C just happened to be the
> temperature that most of the enzymes performed well at without having to
> devote excessive energy to keeping the body cool or reaching the limits on
> resources like water for evaporative cooling.

Cart before the horse.  In the vast design space of protein chemistry,
you can design enzymes to function optimally at virtually any
temperature between 0 and 100C.  Arctic species function at close to
0C.  Thermophilus aquaticus functions at 60C in hot water geysers.
The question of why mammals have converged on 35-40C is an interesting
one.  I personally hypothesize that it has to do with fighting
pathogens, but one way or another, it's dictated by environment, not
the other way around (not due to internal proteomic considerations).
In other words, our enzymes don't just "happen to do well" at that
temperature.  They are under selection to specifically do well at that
temperature.

> > Presumably a Singularity event would produce novel computational
> > substrates, so there's not way to predict post-Singularity energy/heat
> > budgets.
>
>
> There are perhaps half-a-dozen novel computational substrates in the works
> from DNA (chemical) logic to Spin-logic to photonic logic to quantum logic
> to Drexler's rod-logic, etc.  They do *not* require the singularity.  They
> require that they demonstrate sufficient advantages over the current path to
> justify what will presumably be a very large investment that would allow
> them to produce significantly better results than the current path.  My bet
> would be that only robust nanotechnology has the potential for such an
> investment payoff.  *After* the rapid growth phase of the singularity era,
> presumably all of these paths will be explored to see if they provide some
> unique benefits but at that point they will just be icing on the cake.

Martin