I think their hypothesis is that it's not the "what" (interacting with content) that prevents phone users from interacting with the world, but the "how" (head tilted down, eyes focused on a small rectangle, peripheral vision almost completely unused).

With an AR experience (in their hypothesis), the overlay nature of the content being displays does not prevent you from interacting with the real world.

Example - you are at an event and a little popup window appears above the head of the person you are walking toward, using a multi-modal AI assistant to remind you of their name and the context of how you met. This feels different from trying to subtly pull out your phone and figure out how to use some combination of linkedin, email search, etc., to figure out who they are.

Alternatively, how many times have we seen phone users trip on the sidewalk or walk into something because they are so engrossed in what's on the screen?