Mary Hodder is talking about itags. An itag is a tag + author identity + CC license + media object. Media objects can be text, photo, video, or audio. Trusing tags means trusting the maker of the tag. By uniquely identityfy the object, the tag, the author, and the licensing; the itag can live anywhere.
The goal of all this is to put tags and objects together so that they can be included in places like feeds. "I-tagging would remove the requirement for a tag to be coupled with the originating URL (blog post URL) because identity would be inside it."
The assumption of rel-tag is that the tag is the object. The assumption of itag is that the tag is the subject and the class attribute is the object. Still because backwards compatibility is important, itags use the same syntax as rel=tags, and just adds a class attribute. The class attribute is a URL. There are other variants for licensing itags.
One thought I have in looking at this is that the licensing style of markup would be an effective way of adding identity rights to marked up identity data.
As a meta comment on the presentation, Mary used NovaMind to gather the relvant issues surrounding here central premise and then presented by following the branches. I found it to be effective, but the fonts could have been bigger. I had trouble reading it in the second row. I pity the people in the back.