HTML and CSS are used as a presentation markup language. Adding "meaning" to those is approaching it backward. First, mark up the document / data, using XML or RDF (for this argument the preference doesn't necessarily matter). Then, use XSLT to provide a transform into the presentation language of your choice.
I guess we could argue that XHTML _is_ XML at its core, though, so adding attributes to add "meaning" might be doing what I said, anyway.
As far as 'why are the big companies agreeing to this, why isn't everyone consulted?' argument - hey, at least we're making progress. If one large group of pages is marked up in a standard way, then that's a lot of "meaning" that can suddenly be extracted and used in our apps / widgets / mashups / things-to-be-named-later-when-someone-thinks-them-up. In that sense, I'm all for it; let's get started!
There's also a mechanical precedent for this: think back to when the auto industry in the US decided to standardize the "markup" of wheels with a 5-bolt lug nut pattern, spaced the same way - suddenly wheels were interchangeable, and making special chromed add-ons for people was economically more feasible too. They didn't consult a body of international automobile enthusiasts - Ford, GM, and Chrysler pretty much drove that one.