The sigmoid is here (Singularity, no)

[Keith Lofstrom]

The Singularity - the rapture of the nerds - is mostly
a blindered cherry-picking cult. "Artificial general
intelligence" (AGI) is indeed advancing - but with a
faster growth of resources consumed versus advances made.

Here in Oregon, with some of the most advanced silicon
chip factories in the world, the computing story is
layoffs, dividends missed, technologies stalled, and
stock prices dropping ... for Intel, and for other
lesser-known semiconductor companies. That is the
physical substrate beneath AGI.

Photolithography has performed miracles over the decades,
from the 6 micrometer linewidth mosfets I created in the
1980s to today's 6 nanometer linewidths (with smaller
FET channel lengths - details omitted). But those tiny
linewidths require highly energetic X-rays, which damage
masks and chips. Scaling has collided with a horribly
expensive brick wall.

The heatsink systems have grown more sophisticated as
well - a significant fraction of system performance
growth is the growing system power consumed.

Oregon also has some of the world's largest data centers.
Google's data centers in The Dalles, Oregon consume much
of the Columbia River's hydropower, and dump their waste
heat into that river, while providing way fewer jobs than
were promised to the state when they located here, after
kicking out local manufacturers employing more people.

A.I. algorithms have gotten better, granted. My friend
David, who used to improve and manage data center cooling
systems, lost his job to automation, but only after he
developed the problem-solving methods for those cooling
systems. Note - David adapted techniques from the
submarine nuclear reactor systems he operated for the
U.S. Navy. Technology analogizes and then plagiarizes.

REAL technological development isn't described by a
mathematical singularity - divide by zero is ignorance,
not innovation. Innovation begins with observation, not
obsession - looking in new places, not repeating old dogma.
Sorting through the debris of past failures, and creating
something surprising from them. My own inventions exploit
phenomena that others dismissed as defects.

What other defects can we exploit? SBSP has many.

Space solar power does not need to deliver watts to the
power grid. Instead, we can dispense with the grid and
connect space watts to computing IN SPACE, emitting
waste heat directly into the 2.7K void, not into the
Columbia River and Earth's atmosphere.

Space-Earth signalling won't have the bandwidth of
fiber-optic bundles. Fortunately, there are many
computing tasks that are gigawatt-heavy, gigabit-light
(such as experimental AGI) that can "easily" move to
orbit ("easy" compared to the electrical generation
systems that power them).

Note that we will lose the space bandwidth we have
if we pollute it with terawatt interference sources.
Ask a radio designer about "intermodulation".

The thinned silicon of an advanced multicore micro-
processor ( 1cm x 1cm x 40μm ) masses 10 milligrams, and
consumes 200 watts. That's one kilogram per 20 megawatts,
one tonne per 20 gigawatts, vastly less mass than the
hardware (SSPS transmitter and terrestrial rectenna)
required to move gigawatts from space to ground.

Diving deeper, the REASON that microprocessors are
multicore is to share an expensive heatsink and cooler.
If the heatsink is the 2.7K void, it makes more sense to
spread out the cores, then parallel and granulate the
calculations they perform, combining results at the end.
The cores can be on the backside of the PV arrays, which
can be thinner than kitchen aluminum foil.

Most important calculations can be performed in parallel;
an accurate global weather model describes a system that
"moves" at the speed of wind, not the speed of light.
A parallelized weather model could be computed on systems
spread across the width of the Earth's orbit around the
Sun, if it weren't for interconnect signalling losses.

We can easily do weather calculations on city-sized
arrays of much smaller objects, and efficiently connect
them with microwaves.

Delay isn't the problem. If it is, you are using the
wrong computing algorithm, or solving the wrong problem.

Practically speaking, a compute cluster sparsely occupying
a "100-microsecond-round-trip" region of space (15 kilometer
diameter) also occupies a 240 gigawatt beam of light from
the Sun - which (at 8% light conversion efficiency to 20
electrical gigawatts, 20 GWe) could power a metric tonne
of thinned computation silicon.

There is room inside the radius of the Moon's orbit for
2 billion such computing spheres - one for every family
on Earth. If we extend out to the region of Earth's
gravitational "dominance" (the Hill sphere, 1.5 gigameters
diameter) that's one 20 GWe compute sphere for every
human, plus their dogs and cats.

Radiating waste heat directly into the interstellar void,
not into my favorite river and biosphere.

I think we can get by with far less ... we have so far.

When Peter Glaser "invented" the space solar power satellite
(as did Tsiolkovsky and Goddard and others before him),
computing was mostly single transistors in metal cans,
megawatts doing less than my pocket calculator does.
Of course Glaser did not make the kilowatt-to-terabit
"connection", but we do not have that excuse today.

The enabling tool for Glaser's SSPS was the myth of the
rapid-cycle space shuttle, which collided with reality and
a myriad of pesky mass-encumbered details to become the
slow, expensive, and dangerous "space transportation system".

As will other winged/wheeled shuttles, if and when paper
evolves into hardware. Which in turn evaporates the
environmental benefits, in the real world where hydrogen
is made from methane, not electrolysis. Electrolysis
greatly increases the distance to the goal posts, which
are already far beyond our practical reach.

----

Perhaps space-powered computing will also accumulate
disabling details ... fortunately, details like radiation
damage and bit-flips have solutions. Space computing won't
solve rapid-cadence compute problems, but neither will a
rectenna fit in your back yard, nor in a military forward
firebase. Beware of all-encompassing mythical "solutions".

Civilization uses a vast toolbox; our task as technologists
is to replace a few of the most dangerous and expensive
tools with cheaper/better/safer tools.

Not pull miraculous claims out of our poop chutes.
Seemingly, that includes my poop chute, though less fluff
and more quantified details are at http://server-sky.com

Keith L.

P.S.: I should be busy moving that server to Debian,
and making http into httpd like the Kool Kids.
If Debian Linux adepts are reading this, let's talk.

--
Keith Lofstrom kei...@keithl.com