Re: A plea against central planning (was: Why not import everything?)

>Jeff Heflin wrote:
>
>>First of all, I think the problem with the Cyc ontology is it is too
>>big. Semantic Web ontologies should be small and modularized.
>>
>>Second, in order to avoid making every user learn logic and study every
>>ontology, I envision "ontology certification authorities." These
>>organizations will consist of logicians who make sure that an ontology
>>is correct, summarize it and give it a seal of approval. Users can then
>>freely pick and choose these ontologies with some confidence that they
>>will behave as they expect. Note, users are still free to create their
>>own ontologies and to use uncertified ontologies. Anyway, I think a
>>variation of this solution is viable.
>>
>>
>Jeff,
>
>  If we can't make use of large ontologies like Cyc or TAP, which 
>could provide wonderful launchpads for bootstrapping the whole 
>enterprise, we should seriously re-examine our approach.

I agree wholeheartedly. But Jeff may be partly right also. Think of 
millions of small modules on web pages all pointing to a few huge 
Cyc/TAP-like providers of 'agreed meanings'. Then think what it would 
be worth to point to a provider of agreed meaning which was just a 
little bit better at supporting inferences than the free ones are.

>I do understand that some AI communities have reservations about 
>aspects of  KBs such as Cyc, but they do represent a significant 
>commitment of resources which should be usable by the SW.
>
>  I would also like to plead against the use of  concepts such as 
>"ontology certification authorities" staffed  by logicians. If the 
>infrastructure we build is so hard to understand that we need 
>professional logicians to approve it before it is safe for 
>consumption by others, it will be sad. I also don't see a bunch of 
>logicians understanding medicine or law or ... well enough to 
>provide a seal of approval of an ontology of that topic. I can just 
>imagine the state of Kansas approving an ontology which set the 
>value of pi to be 3 ;-). But seriously, an approach that relies on 
>this kind of certification process as part of the infrastructure is 
>antithetical to the net ... it reminds me too much of central 
>planning.

Let me suggest an alternative vision that might combine the best of 
both views. I agree with Guha that the idea of a 'central authority' 
is antithetical to the Web. But a certification authority does not 
have to be a central authority: it can be a *service*. Consider the 
Good Housekeeping organization, for example, which is such a service. 
It tests things for safety and offers a public warranty to those it 
finds acceptable. That warranty is worth a lot to the manufacturers 
and to the customers, largely by virtue of the public perception of 
the GHI as being objective and reliable. There is an interesting 
three-way system of trust and economics at work here between the 
producers, the consumers and the warranty providers. The trust which 
C has in W means that W's warranty increases the value of P's product 
to C, so P is willing to pay W a considerable sum to display W's 
brand. This is a business model which depends crucially on the 
payments P makes to W being insulated from the trust the C places in 
W's objectivity, a delicate point which deserves some careful study. 
But a similar model could apply here, where the W is the expert 
certification auth....er.... expert certification *service* - and P 
is the ontology provider. One can transfer the entire picture over 
almost intact, where P is the owner of an ontology which C is using 
to 'borrow' some meaning, and W's warranty makes P's ontology worth a 
little more so he can charge a higher fee for using it. This only 
becomes a business model when there is a viable economy of semantic 
information and when reliability of inference is worth something, but 
I don't expect the SW to take off until such an economy becomes real 
in any case. When it does, it will take off so fast that we will not 
know what happened, because it will be possible to make money at a 
rate defined by a small constant times a factor which is proportional 
to Moore's law. It really doesn't matter how small the constant is.

Pat
-- 
---------------------------------------------------------------------
IHMC					(850)434 8903   home
40 South Alcaniz St.			(850)202 4416   office
Pensacola,  FL 32501			(850)202 4440   fax
phayes@ai.uwf.edu 
http://www.coginst.uwf.edu/~phayes