Basically people you really know are 1st degree, people they really know, and so on depending on how far you want to go. So with a given user you could measure how far away you are in the web in trust, how many interconnections there are, and begin to sus out AI agents. Each person in that network has a reputation cost in being tricked. Eventually there could be two principle parts of the web - a bot side where they are verifying each other, and a human side, with some but not tons of connections between the two sides. It’ll get really important when bots are earning money and are smarter than humans.