The Semantic Web Considered Harmful?
Today, there is tremendous activity and promises under the auspices of the Semantic Web covering the academic, research and industrial sectors. This includes numerous accounts of the benefits that this set of new technologies brings to these communities, and ultimately, the end user - underpinned by the RDF and OWL languages. These benefits include the rhetorical “semantics” and much vamped “reasoning” features.
I was part of the original group of people that meet W3C at MIT one cold December in 1996 to persuade them to look at a common metadata framework for the web. We succeeded and the first RDF Working Group commenced in 1997 and the rest is history. However, almost immediately, a strong push towards formal knowledge representation was evident. This community saw a new opportunity to re-invent the discipline. A new name, a new beginning, a new market opportunity.
Very little has been written that challenges the approach and relevance of the Semantic Web. Not that this new technology has taken us backwards - but there is a serious lack of critical analysis of the prophesied advantages and determining if these are valid and reasonable outcomes for the web metadata community. The alternative to this is that we spend significant resources and time leading in a direction that we may not want to head - or - have not learnt from past similar experiences.
Lets look - very briefly - at the two benefits already mentioned; semantics and reasoning. The funny thing about Semantic Web is the almost mythical saying that the Semantic Web is good because it capture the “semantics”...obviously...as they make the constant comparison to its poorer cousin, XML Schema. In the former, its all about “semantics and ontologies” and the latter its just datatypes, no semantics, just datatypes with some structure thrown in for good measure.
In the below example, no one has yet to explain why the former somehow capture “semantics” and the latter does not:
<rdfs:Class rdf:ID=“Person” />
<xsd:complexType name=“Person”>
Ironically, in the poorer cousin, I could create an extension of the Person datatype for a new Student datatype, inheriting all the parent attributes. But that’s just not semantics!
If we take a brief look back at history, an interesting parallel could be made with SGML. SGML is complex and difficult to develop applications for. In fact, there are very few applications that implement full SGML. Like most complex things, people make it smaller and hence simpler - a profile. In this case, the resultant HTML profile hit the right balance and we now have trillions of pages of the stuff.
I believe the same will happen for the Semantic Web. It is far too complex as it is today and exceeds its ideal functions. It will be profiled and a much smaller subset of RDF/OWL will emerge that satisfies 95% of the web community needs. We are starting to see this happening now as the splinter microformats community charges forward with simple HTML hacks to embed metadata. It may not be pretty but it works. The Semantic Web is the new SGML and we await the natural process to evolve.
Could the Semantic Web be harmful? It could if it continues to steer resources towards an unobtainable goal in web-scale terms. Such resources could be used to provide more grass-roots level solutions. Like SGML, the time has come to profile the Semantic Web, get rid of the dead wood, and (re)-address the metadata needs of the web community.