One of the hottest topics
in futurology concerns the possibility of what is called a
technological singularity. Those who imagine a singularity postulate
that computers will keep getting faster and smarter, following the
trajectory of Moore's Law (the rule that every two years there is a
doubling of the number of transistors that can be packed on a circuit
board). Before long, it is argued, we will be able to fit the
computing power of a hundred human brains on a single desktop
computer. Presumably this will lead to the emergence of computerized
superintelligence. This emergence of superintelligence (in what is
called an “intelligence explosion”) is referred to as the
singularity.
The term singularity was
popularized by Ray Kurzweil in his book The Singularity is Near.
Kurzweil predicted that this singularity would occur around the year
2045. His book contains quite a few logarithmic diagrams plotting the
growth of computing power over the past twenty five years. Extending
the trend line a few decades into the future, he estimates that
within a few decades the total brainpower of all computers will match
the total brainpower of all humans. Singularity enthusiasts postulate
that soon this “intelligence explosion” will lead to computers
that are far more intelligent than humans. Singularity enthusiasts
also imagine there will before long be a merging between computers
and men, allowing people to have their minds connected to computers
or enhanced by computers.
If someone tries to cite
possible limits to how small silicon chips can be miniaturized,
advocates of a singularity will mention other promising technologies
such as quantum computing and biological computing, which may well
allow Moore's Law to continue for many decades, with computers
basically getting twice as fast and powerful every two years.
However, there is a huge
bottleneck that well may mean that such a singularity does not occur
anywhere near as quickly as its advocates predict. The bottleneck is
software. Large advances in computer intelligence require equal
progress on two different fronts: the hardware front and the software
front. To create a computer as intelligent as a human being, you
would need not only hardware vastly better than anything available
today, but also software thousands or millions of times better than
anything available today.
Unfortunately the annual
progress rate of software is much slower than the annual progress
rate of software. Software development progress does not at all
follow any rule of progress as dramatic as Moore's Law.
At what rate of progress
is software improving from year to year? There is really no exact way
to answer this question. Any answer is a subjective judgment call.
In his book Kurzweil
estimates that software is improving at a rate of doubling in power
every six years. But he provides no reasoning to back up this claim,
and it seems that he just kind of picked the number out of a hat. As
someone who has worked in software development over the past twenty
years, I can say that from a development standpoint it doesn't seem
like software is four times more powerful than it was twelve years
ago. In 1997 programmers would develop programs mainly by using
compilers with graphical user interfaces, the internet, object
oriented languages, and class libraries. That's exactly how
programmers develop software today.
But to be generous to
singularity enthusiasts, let's suppose that figure is correct. If
software doubles in power or excellence every six years, it will
still mean a huge and growing gap between our future advancement in
software and our future advancement in hardware.
To see how big this gap
is, let's see what figures we get if we double
hardware power every two
years, and double software power every six years.
This table uses a number
of 1000 as an arbitrary starting point.
What we find is that by
the year 2043 computer hardware power has increased by a factor of
32,000 times, but computer software power has increased by only 32
times. The end result is that computer hardware ends up being 1000
times more powerful than computer software.
What does this mean in
practical terms? It suggests that the technological singularity will
not occur anywhere near as quickly as singularity enthusiasts
imagine. We will not at all have anything like superintelligent
machines (or even computers as smart as human beings) if they are
using software that is only 32 times better than today's software.
We probably won't have computers as smart as human beings until we
have software that is many thousands of times better than today's
software.
This gap between the fast
rate of progress of hardware and the slow rate of progress of
software is called the software gap. The software gap may mean that
you won't see any singularity in your lifetime unless you are young.
There are currently singularity enthusiasts in their fifties who
imagine that they will be able to escape death by uploading their
minds into computers or robots, after computers and robots become as
intelligent as people. I have no such hope. I don't think the
software will be ready before I die.
To see Paul Allen's
argument about the singularity (similar to mine), use the link below:
No comments:
Post a Comment