What use the AstroOntology

[Eric Saunders]

Hi

I think there are several, quite seperate ideas and assumptions here which 
need making explicit, if we want to be sure we are all talking about the 
same thing.

Two approaches to codifying knowledge have been discussed. In the 
classical philosophy of ontologies (as championed by Ed) there is a kind 
of 'perfect' external world of atomic facts which the ontology seeks to 
accurately represent. Once the back-breaking labour of building this 
knowledge base is complete, the hope is that all useful queries become 
analytically solvable. This idea has dominated predicate logic reasoning 
approaches in orthodox AI for more than 50 years.

The fuzzy, tag-cloud approach of Flickr et al. takes a totally different 
approach. The knowledge structure is not built by design or careful 
construction. Instead, the continuous amalgamation of millions of 
autonomous individuals making personal, biased decisions allows 
differentiation of links based on popularity. If a link between two 
concepts is useful, then it tends to be spontaneously generated by more 
people. The tag cloud simply sorts by popularity. This is the same guiding 
principle that Google uses - if your page is popular, then that means many 
people want to access it. Therefore, it is important.

It seems to me that this list is in general working to the following 
specific assumptions:

* that we require a relatively precise, static specification of knowledge
* that such a specification is tractable and meaningful
* that such a specification will help us do better astronomy
* that modifications such as aliases and refactoring of the ontology will 
allow this to be maintained, and make it a 'good fit' to the subject 
domain

Whether these assumptions are reasonable is a matter for discussion on 
this list. My personal view is that static ontologies are likely to be of 
limited usefulness in a field where concepts are ill-defined, in a 
constant state of flux, and are often subjective. I question the 
assumption that knowledge in astronomy is made out of atomic facts. And I 
don't think the maintainability problems will go away by liberal 
sprinkling of aliases or probablistic 'might-be-a' relationships. As an 
earlier poster pointed out, the strength of ontological reasoning is 
derived entirely from the rigourousness and precision of the concepts 
defined. Sacrifice that, and all queries will return the equivalent of 
'here's some things you might be interested in' - which is where we 
already are.

It's not my intention to be negative or confrontational. I appreciate that 
these are works in progress, people are committed to different approaches 
etc. It's likely there excellent counter-arguments to the issues I've 
raised. I think they should be made explicit. A bit of clarity would 
certainly help me understand the viewpoints others here hold!

Cheers

Eric

---------------------------------------
Eric Saunders
eSTAR Project (http://www.estar.org.uk)
Astrophysics Group
University of Exeter



On Mon, 5 Mar 2007, Ashish Mahabal wrote:

>
> This is getting very Godelsk.
>
> The aliasing mechanism provides for consistency. We could thereby miss some 
> truths. We can only hope that we do not miss any interesting ones.
>
> Flikr, blogspot are inconsistent systems in that A would define SN as 
> something and B would define SNe as something else, but C may presume them to 
> be the same. All will be happy, but at least one will be wrong.
>
> -ashish
>
> Ashish Mahabal, Caltech Astronomy, Pasadena, CA 91125
> http://www.astro.caltech.edu/~aam aam at astro.caltech.edu
>
> Gods do not protect fools. Fools are protected by more capable fools.
> -Luis in Larry Niven's Ringworld
>