-----Original Message-----
From: Helmut Enck-Radana <[EMAIL PROTECTED]>
Date: Wednesday, August 08, 2001 03:51 PM
>At 20:21 08-08-01, Tom Fawcett wrote:
>>This is an interesting topic but it seems unrelated to artificial
>>intelligence using Perl.
>
>One of the traditional topics of AI is to find ways to generate
>implementations from specifications. While this is not possible in general,
>there are enough situations where executable specifications are only too
>inefficient to be useful. Let's say you write a specification for your
>application in a specification language which is restricted such that it is
>possible to implement an interpreter for this language. Even if you used a
>low-level language to implement this interpreter, the execution of the
>specification could be too slow. But you write this interpreter in your
>preferred high-level language (which could perhaps be Perl) and apply a
>supercompiler to this interpreter together with your specification as
>input. Imagine the resulting program would be efficient enough to be used
>instead of a hand-coded program. This would be a solution to a classic AI
>problem.
>
>I am interested in comments of the supercompiler specialists on how
>realistic/unrealistic my scenario is.
>
>-- Helmut
Hi, Helmut,
Your scenario is in the mainstream of the use of supercompilers, and you
formulated it
precisely. Up to now we have dealt with the case where the specifications
are those of an algorithm,
but we think of trying some AI applications. I thought of a general
criterion of how effective
the supercompilation will be. It is this: if the role of specificaiton
remains essential at all stages
of the construction of a solution, then we can expect that the effect of
supercompilation will be big;
otherwise the radical improvement is unlikely. Specification of an
algorithm obviously
satisfies this criterion; but theorem proving from the first principles
does not. Indeed, specification
here is the list of axioms, while the progress towards the proof depends on
the total list
of true statements, in which the axioms constitute a progressively smaller
part, relatively.
Best regards.
Valentin Turchin.