🚀 go-pugleaf

-From: arromdee@jyusenkyou.cs.jhu.edu (Ken Arromdee)
-Subject: Re: Comet Collision
-Date: 30 Oct 92 18:16:16 GMT
-Organization: Johns Hopkins University CS Dept.

-In article <Bwxqxw.sq.1@cs.cmu.edu> amon@elegabalus.cs.qub.ac.uk writes:
->You are guilty of the classic fallacies of extrapolation. The first  
->fallacy is an overestimate of short term advance. The second is the  
->underestimate of the long term. Both are due, in part, to linear  
->thinking, straight line extrapolation. Technological capabilities and  
->knowledge are accumulating exponententially. If you make a linear  
->extrapolation in a particular field, you will tend to "draw a line"  
->of some slope that seems "reasonable". But the problems invariably  
->turn out to be more difficult in your field than you actually  
->believed.  So the exponential will for some period of time grow  
->slower (lesser slope) than your line, but when it finally does cross  
->it...

-There was a non-fiction article in Analog (a science fiction magazine) around
-1960.  The author pointed out exactly what you do here: lots of ideas about the
-future are extrapolated linearly.  He graphed the maximum speed that human
-beings could travel at, as an example, and showed that it increased faster than
-even an exponential.

-The curve also showed, however, that (by extrapolation) humanity would surpass
-the speed of light by the year 1985.

-I believe that in the 1980's he (or someone, anyway) wrote another article
-recognizing the existence of S-shaped curves that eventually level off.

-The moral is that some barriers can't be found by extrapolation from points
-before the barriers have any significant effect.  It's legitimate to argue that
-advances require solving a difficult or impossible problem; just noting that
-you can graph past that point is not an answer.

-(How do you conclude that technological capabilities are accumulating
-exponentially, anyway?)

There's a limit to growth of knowledge that doesn't seem to be getting as
much attention as it warrants in light of the seriousness of the problem:
the huge amount of information now available is pushing the limits of the
traditional library system.

If I need a specific item of physics or technology for a calculation, I can go
to the library and look through the Handbook of Chemistry and Physics,
Eshbach's Handbook of Engineering Fundamentals, and various encyclopedias.
If I can get the information there, then all's well, and I can often find it
in a few minutes to an hour. However, there is a considerable amount of
information in those publications that is not adequately covered in the
index, so I may have to think of all the related topics and run an
exhaustive search through them, which could take hours. If the information
is not there, I have to look for the specialized books and periodicals that
cover the topic (we do have a computerized card catalog, which helps
considerably with this). The specialized books tend to have pretty poor
indexes, so I often end up having to read a considerable portion of each
book to see if what I need is there. Sometimes the book is not available 
locally, and it may take days to weeks to find a copy. (Assuming I don't
get fed up and postpone the entire line of investigation.)

When I finally find the information, it is often expressed in terms that are
highly specific to that particular field of study, since specialized books
are mainly intended for specialists. So I may have to study the subject for
days just to convert the information to a form that I can use.

The end result is that I may have to spend hours to weeks just to assemble
the information I need, before I can begin to work on my own design or
calculation. That's the status quo, and researchers seem to accept it more
than complain about it. But it does slow down further advances considerably,
and the problem continues to get worse as more information is added.

I think much greater use of computer technology can aid substantially in
dealing with this problem. Data bases, pattern searches, and greatly
improved indexing offer considerable hope. For instance, when the Grolier's
Encyclopedia came out on CD-ROM, magazine articles claimed that it had an
exhaustive index generated in advance by computer, which is actually larger
than the text of the encyclopedia itself. With aids such as these, research
time can potentially be cut tremendously. (Anyone know whether the CRC and
Eshbach's handbooks are available on CD-ROM yet?)

Another change that's very important in the long run for cross-pollination of
different fields is to make the notation and terminology more uniform, and
to include more explanations in the text - with conversion to electronic
format, the pressure to condense the text as much as possible to minimize
the number of printed pages is greatly reduced. Ideally, much of the
information in specialized fields would be usable by non-specialists
for peripherally related projects.

Putting reference materials on CD-ROM is a considerable step forward, but
perhaps an equally important advance is the development of huge centralized
data bases that can be accessed by multiple users. Having multiple terabytes
on-line at all times can nearly eliminate the task of sorting through stacks
of CD-ROMs. Fully implementing this may require changes in policies toward
copyrights and royalties.

Additions can also be made to the field of indexing. A permuted index (?)
allows the researcher to find every instance of a term in the text. It's
also possible (though very difficult) to create a "search tree", in which
the researcher enters one or more keywords or phrases, and the system
responds with a suggested range of topics that are relevant. (For instance,
to go back to the encyclopedia example, entering "Thomas Jefferson" might
bring a response of "President", "US History", etc., and entering "Thomas
Jefferson" and "Monticello" might produce "architecture".) This might be
an interactive system, in which the user is led through a sequence of topics
to find what is needed. (Encyclopaedia Britannica has one volume that
essentially does this. I don't usually find that implementation particularly
helpful, but at least it's a step in the right direction, within the limits
of what can be done in a printed format.)

For the present, and at least for specific topics, the computer networks
are a valuable resource. It's not too difficult to select interesting
posts and file them by category, after which they can be searched by topic
or by pattern match. Even without a regular data base program, "grep" is
usually fast enough to produce useful results.

And of course the frequently-asked-questions list is very helpful, though
I'd like it to be much bigger than it is, and to include a section for
speculative proposals and calculations, so at least people can see what's
been discussed before, and in how much detail. (Maybe a section for 
calculations and proposals that have to be "refereed" by independent checks
before they can "graduate" to the main body of the list.)

John Roberts
roberts@cmr.ncsl.nist.gov