biep@cs.vu.nl (J. A. "Biep" Durieux) (08/26/87)
[Is this what they call chopsticks? I stick out my neck and you chop off
my head? :-)]
[Warning: look above at the "Lines: "-line!]
[ Sarge, you have put an article on the net which expired before I got
to answering it. It was a reaction to (among other things) my question
what you were hiding under the label "empirical". Do you still have it?
(I haven't archived this discussion. I believe Mattew Wiener has?)]
(1) Sarge Gerbode complains I am using "heavy" words for simple concepts.
He is right. For me, "methodology" just means "total method", but then
as expressable in a rigid formalism. So "creativity" (too vague) and
"screwdriver" (outside the scope) cannot be (part of) a methodology.
I use the "heavy" word to stress the formal-ness, "means" and, to a
lesser extent, "method" being too broad terms.
(2) Then several persons have complained that what I describe isn't metho-
dology (at all?). That is possible, but please, in that case:
- provide a better definition (or rather strict description), and
- tell me where *my* stuff belongs.
(3) If I only want to do planning and problem solving, you don't need to
work on automating science, according to Bob Myers and Jeff Dalton.
The latter also suggests that automating science has been tried and
failed.
Well, I don't know whether the latter is true: take the AM program by
Doug Lenat, or its successor EURISKO. AM was told some rules of set
theory, and from there started playing around, inventing the concept
of equivalence-class-accorting-to-cardinality, from there cardinality,
from there numbers, addition, multiplication, division, factorisation,
prime numbers, and unique prime factorisation. After the program had
been adapted, so that it could adapt its own heuristics, (and after
renaming it to EURISKO), it also has become world champion in the
game "trillion credit squadron", about fleet design, has invented
several new (now patented, I believe) smart chip designs, found a
non-optimality in the interlisp system it was coded in (some day Lenat
found EURISKO had redefined the EQUAL-predicate, to make itself
faster - it has rewritten a lot of its own code to that effect too),
and more. Given that this program is a complete stand-alone, not
the result of many people during many years working on similar
programs, I think the result is promising. Or take the BACON family
of programs, that by interpreting measured data and suggesting
new experiments have "redone" the advent and the building of
modern physics and chemistry, notwithstanding the sometimes large
measuring errors in the data they were fed. (Family, since every
time the program was made stronger and more general, and got a new
number). The important thing is, that both programs make extensive
use of the same rules to cope with different situations.
Then, of course, the results of early computer translation were very
promising too.
About the statement that automating science isn't necessary for
planning and problem solving: I think the problems are almost identical.
P&PS is a subject which behaves a lot like doing science: the central
part of it is modelling (with abstraction and generalisation), a
quite hard problem. A simple example of a difficult question is
"How does one proceed in finding a representation for some phenomenon,
given a medium?". Everybody knows how important good notation is
for a scientist, and for the same reasons it's important for
a problem solver: how much difference doesn't it make to have a map
spread out before you instead of a list of "A at coordinates x,y",
when you want to plan a trip. So if you need to do a lot of trip
planning, you'ld better start trying to draw a map, or invent the
concept of a map first. For those who want to try: what would be the
best representation for sounds in a speech-recognizing computer?
Both the known data and the sentence-to-be-recognized should be
well-representable.
Jeff has an argument I couldn't follow, to the extent that
- science is creative
- we are able to do planning and problem solving
- so automated science is not necessary for P&PS.
Do you mean we can do *automated* P&PS? I think we are a far way from
that; outside toy worlds like block worlds we aren't doing very well
yet. But perhaps you don't mean that, but then I don't see your argument.
Please explain.
(4) Bob thinks it's silly to make a distinction (as to "philosophical"
vs. "scientific") between the work a scientist does when he applies
rules (or creativity, or whatever) and when he states new rules of
conduct.
I make that distinction, since the first of these activities is
concerned with the world he seeks to know (model, whatever), and the
second with (very superficial, but nevertheless) epistemological
notions. I suppose Sarge will agree with me on this distinction.
I must, however, admit to having been sloppy. The techniques for
forming new rules of conduct in new situations may be completely
described, i.e. there may be a methodology for it. In that case
the scientist is still not doing (his or any) science, but I would
hardly call it philosophic, as he certainly won't convince himself
of the philosophical reasons for adapting the notions he does, etc.:
he just accepts the rule as correct, as he would do with a mathematical
rule. I think I would neither call that philosophy, nor science.
(5) In a similar line, Sarge wonders whether I want to make the Ph/Sc
division identical to non-mechanisable/mechanisable. I don't think
so. In the first place, there are many, both mechanisable and non-
mechanisable things around, which are neither ph., nor sc., and not
just the things mentioned in (4), but lots of things from "real life",
from love to chatting with the neighbours. I do however think that
something which is mechanized cannot be philosophy, as one important
aspect of philosophy is the continuous checking of results got
against ones "inner feeling". [I am barring the idea of men-replaced-
by-robots here, as I just cannot imagine what that would be like.]
In science this is much less the case, there a possible replacement
is "reproducibility" *outside* of the original scientist.
A division I am interested in is the one between the "basic" rules
of conduct, and everything (both other rules of conduct and the
conduct itself) that is derivable from it. I do, however, not a priori
put labels like "philosophy" or "science" on either set.
(6) Bob gives me two things to try to fit into rules: The idea of Conti-
nental Drift to explain similar-looking coastlines and the idea of
Evolution to explain the existence and distribution of animals.
Unwittingly he has given me the easier task, explaining things. I
think recognizing two coastlines to look similar is much harder
(provided you weren't told to look for it). I had started to write
an answer to this, but things got *very* long, and I have decided
to take some time to work on it.
(7) Bob also wonders whether, as most global geologist work in the framework
of (the idea of) Continental Drift, CD should be seen as a methodology
according to my view. No, I don't think so. However, the idea of
working within a general framework of some somehow convincing hypo-
thesis *is* a methodological rule (whether a good one or not).
--
Biep. (biep@cs.vu.nl via mcvax)
Is the difference between a difference of degree and a differ-
ence of sorts a difference of degree or a difference of sorts?