"When you have a large pile of stuff, you need a way to identify it. The more meaningful the names, the worse they scale. For example, if you want to make a photo of a rabbit findable by anyone across the Web, calling it "rabbit" or even "rabbit_305464" (because there were 305,463 rabbit photos posted before yours) may make it easier for English speakers to find it, but it makes it harder for those in other languages. Plus, while you think it is a photo of a rabbit, someone else may think it's a photo of a pet or dinner. A better solution is to take the semantics out of the identifier so that multiple semantics can be layered on top: "Ah, you mean photo #F345A90875264D3425! The one that Deb Franklin calls 'Rabbit' and that Jean-Paul Lingerie calls 'Lapin.'" (Yes, the alpha characters imply a particular alphabet, which may be a problem.)
We could wait for authorities in each domain to issue the numbers, but we'll make more progress faster if we accept that multiple interest groups within a particular domain are going to issue UIDs."
Towards the end, David also introduces a new potential user group for thinglinks.