From: Charles D Hixson (charleshixsn@earthlink.net)
Date: Thu Mar 30 2006 - 12:35:48 MST
On Thursday 30 March 2006 09:01 am, Joshua Fox wrote:
> Singularity work requires us to look at future trends, such as
> accelerating increase in raw computer power towards human levels. But we
> cannot judge the accuracy of our predictive techniques except by
> waiting. One approach is to look at predictions from the past and see if
> those predictive techniques were successful. Kurzweil tells us of the
> success of his predictions from 10 or 20 years ago, but I'd like to see
> more of this sort of retrospective.
>
> Can anyone point to me a webpage collecting predictions from decades ago
> about the general direction of technology ? I'm not interested in
> soundbite predictions, but rather in an overview of deeper long-term
> thinking of the past, like the original Moore's Law article or /The Year
> 2000/ by Kahn and Wiener. Preferably this overview would not just
> present the old predictions, but also analyze the accuracy of the
> prediction techniques used.
>
> Joshua
I don't know of a reliable central source of the information.
Trend curves have a very long history of being unreliable. They're just
better than most other models when you don't understand the system being
predicted. Don't put much faith in the projections, and don't put ANY faith
in the precise predictions. Still... what are the alternatives? Modelling
is based on understanding the system being modelled. Trend curves just
require a simple variable which is CHOSEN to be observable. (Accurate models
usually require modelling changes of variables which are quite difficult to
observe or understand. And even then, models are ALWAYS
oversimplifications.)
You might look into the multitude of articles on "Moore's Law" (check Google).
That's a pure trend curve based projection (if you don't work inside of Intel
or AMD) that's proven quite successful in the past.
What Kurtzweil does is collect together a bunch of trend curves, and presume
that an ensemble of trend curves is more reliable than a single one, which
may or may not be true, depending on the fluctuations in the underlying
system. Some changes will affect all of the trend curves simultaneously.
(The classic example is a devastating war, but plague could also count. If,
e.g., H5N1 turned out to be much more deadly & contagious than anyone
predicted when it mutated, or if HIV mutated into a form passed by simple
contagion, or...) Also, just because the trend curve is proven wrong
doesn't mean it's been slowed or stopped. Unforeseen future discoveries
might exert an equally strong force in the more sharply upwards directions.
And there WILL be unforeseen future discoveries.
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:56 MDT