SUO: RE: automating abduction?
Jon:
Or, as the English have often exclaimed:
O Times! O Daily Mirror!
Once we get an automated procedure that can (a) create a set of hypotheses
among which to pick, (b) apply the pragmatic maxim to those hypotheses as
some kind of cost/benefit analysis, and (c) modify the set of criteria used
in the cost/benefit analysis when the use of said criteria proves to be too
costly and not beneficial enough -- well, then we can all retire and let the
machines do the thinking for us. Granted that we have formalized and
computerized a lot of what we knew about the "logic of discovery" half a
century and more ago, have we really moved any closer to the goal of
creating that kind of procedure? Given that granted, which covers a lot of
work I am sure, I doubt it.
The relevance of this remark to our concerns in this forum is that since (as
I take it) the gap between such procedures and any reasonably creative and
informed human hypotheses generator/evaluator is great, we should continue
to rely on the humans. In the context of creating upper ontologies, the
hypotheses we are concerned to generate and choose among are proposed
generalizations of a set of concepts, i.e. supertype tables to subsume a set
of lower level tables, object classes to subsume lower level classes.
We can indeed build tools to help us with these efforts, but these are not
the same as tools to take over these efforts. Automated supertype generators
cannot yet replace the hard work human beings do. The human ontologists
working at Cyc are not about to be replaced by some creature arising from
the cog sci labs.
I will try to get to the papers you cite. Thanks.
-----Original Message-----
From: Jon Awbrey [mailto:jawbrey@att.net]
Sent: Wednesday, September 03, 2003 9:09 PM
To: Tom Johnston
Cc: SUO
Subject: Re: automating abduction?
o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o
Tom,
I remember the times -- let's not talk about the mores --
and these remarks were relieved against the backdrop of
discussions about "discovery procedures", in particular,
algorithms for "grammar induction" that were envisioned
to be capable of "systematically searching the space of
all possible hypotheses" to arrive at a fitting theory.
Chomsky simply recognized this as Peirce's problem about
"giving a rule to abduction", instead of induction per se.
As experiments on computer and human subjects mounted up,
and rode off in all directions at once, most people, well,
all but the most diehard behaviorists, began to appreciate
just how poverty-stricken the available stimuli actually
were in comparsion to the combinatorial vastness of the
hypothesis spaces in all but the most trivial settings.
I remember being puzzled, along with many others, about
all this, and it was some time before I realized, simply
by taking seriously what I had been reading in Peirce all
along, that Peirce did "give a rule to abduction", and
its name was none other than the "pragmatic maxim".
There's some links to relevant papers on
abduction in computational settings here:
http://carbon.cudenver.edu/~mryder/itc_data/abduction.html
http://www.dipf.de/projekte/Paed_Sem_HCI/Online_Artikel_Peirce.htm
Jon Awbrey
o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o
Tom Johnston wrote:
>
> Who else but Chomsky could have gotten away with an "accounts for"
> that has exactly the explanatory force of Moliere's "dormative power"?
>
> o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o
>
> John, Rich, Tom, et al.
>
> More old notes that bear on this topic --
>
> | The human mind is a biologically given system with certain powers and
limits.
> | As Charles Sanders Peirce argued, "Man's mind has a natural adaptation
to
> | imagining correct theories of some kinds .... If man had not the gift of
a
> | mind adapted to his requirements, he could not have acquired any
knowledge"
> | (ed. Tomas, 1957). The fact that "admissible hypotheses" are available
to
> | this specific biological system accounts for its ability to construct
rich
> | and complex explanatory theories. But the same properties of mind that
> | provide admissible hypotheses may well exclude other successful theories
> | as unintelligible to humans. Some theories might simply not be among
the
> | admissible hypotheses determined by the specific properties of mind that
> | adapt us "to imagining correct theories of some kinds", though these
> | theories might be accessible to a differently organized intelligence.
> | Or these theories might be so remote in an accessibility ordering of
> | admissible hypotheses that they cannot be constructed under actual
> | empirical conditions, though for a differently structured mind
> | they might be easily accessible. (Chomsky, ROL, 155-156).
> |
> | Noam Chomsky, 'Reflections on Language',
> | Pantheon Books, New York, NY, 1975.
>
> o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o
o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o