« TheirGoogle II | Main | TheirGoogle III: reverse engineering the Google site »

Are ontologies irrelevant?

I know someone who once used two different Pantone colors on an architectural object in order to make it look as if it was one color. This was done to keep a client happy. The client had asked for a particular Pantone color, which my friend had used - but the light was such the client saw two colours and hence complained the object was two colors (yet it was one, and according to Pantone, the right one). To remedy this required covering to object with two other Pantone colors.

Despite the ambiguity, the client got the desired colour in the end.

Anyone that's ever had to match paint to a color chart or get a website just so across browsers has experienced something like this. Anyone who's ever looked at a Constable or a Cezanne and thought they saw the world has experienced this. So much for Pantone, CSS, or linseed oil.

In the same way Pantone can't protect you from sunlight, using unique names (URIs) is not going to stop ontologies breaking into smithereens against the world that is the case for web software. The history of AI in the 1970s and the history of mathematical logic in the 20th Century offers compelling evidence that proper names and denotational semantics are not sufficient for the semantic web, and we'd do well to get over it.

Nor, does it matters whether two RDF agents are talking about the same thing. Philosophically it matters, but this won't get us any running software - computationally it's verging on the irrelevant to agonize what a URI points at. What matters is whether two or more URIs are interchangeable in a given set of graphs. RDF alone can't provide solutions to that problem, it only allows for the definition of the graphs and expression of the recognized functions on those graphs.

One good way to deal with this in order to get some running code out is with probabilities or statistical techniques which are fed the RDF graphs and allowed to compute whether the URIs line up. Hybrid reasoners have been popular in AI and robotics for over a decade. Failing coding with probabilities, just make a neat hack that is coherent for a given domain - this will be almost always be as good if not better than the current ocean of code will jokingly call business logic.

November 5, 2003 09:19 PM


Trackback Pings

TrackBack URL for this entry: