Re: a simple question

On Dec 24, 2003, at 8:48 PM, Drew McDermott wrote:

>   [Bijan Parsia]
>>  Sure, lots of semweb reasoning will be done by random Perl and Python
>> scripts (or fairly cheap prolog hacks). But there's some sort of
>> difference between acknowledging that and wanting to privilege 
>> "certain
>> kinds of processing mechanisms" for interoperability purposes. And
>> interop is somewhat the name of the game, I'm pretty sure.
>>
> I'm not thinking of little scripts and stuff.  I'm thinking of big
> black-box algorithms, such as heuristic programs for bidding in
> combinatorial auctions.

I was thinking that lots and lost of little scripts add up to a pretty 
velvety box.

If we look back at how "large scale" websites (and the software that 
backed them) developed, there was a lot less methodology than one might 
have hoped for. Or not.

>> [snip nice paragraph that seems straight out of Critque of Pure 
>> Reason]
>>
>> So, Drew, is there any evolution in your position in CoPR and that
>> paragraph? In your experience?
>
> Not much evolution, if any.

Didn't seem like much.

>     [me]
>>> If you really stand by this, then there really is no difference in 
>>> our
>>> positions.  The assumption set, in this case, will include an
>>> assumption that "the algorithm did not err on this occasion."  How
>>> would one check that without reopening the original question?
>> [snip]
>>
>> Er...isn't the difference that you think the assumption isn't 
>> checkable
>> (in fact) whereas Pat thinks that it is?
>
> Pat's paragraph is subject to multiple interpretations.  If he means:
> an algorithm might cut all sorts of corners, but must in the end
> produce a proof of its conclusions to accompany those conclusions,
> then the assumptions would be checkable.  I've read it a couple of
> times, and I can't tell if he meant that or not, especially in this
> sentence: "Nothing in the semantic specification of the language
> requires that all reasoners only perform valid inferences."  Maybe Pat
> himself will tell us.

Fair enough.

However, I was sorta referring to the *specific* assumption "the 
algorithm did not err on this occasion". If the algorithm must produce 
a verifiable proof of its conclusions...you still don't have a check of 
the *not erring* assumption *if* it's not an error to produce an 
erroneous proof (since the conclusion, after all, might well be 
entailed).

Cheers,
Bijan Parsia.

Received on Wednesday, 24 December 2003 21:06:01 UTC