To be honest, I don't find anything surprising about this, as what he describes is a traditional "small world" network (high clustering coefficient with short path links). The strong links within clusters efficiently amplify information, and the weak links between members of different clusters serve to transmit across the graph. (The fewer sparse cuts needed to decompose the graph, the less effectively you will be able to transmit data across it.)

For example, Milgram (1969) found in his famous letter-delivery experiment that the strongest predictor of success when sending a letter across ethnic lines was whether the first jump between ethnic groups was a "strong" or "weak" link -- weak links were significantly more likely to end in a successful transmission. This is almost certainly because ethnicity tends to be an accidental (due to uncontrollable historical factors) signal of socioeconomic group; weak ties are more likely to be across socioeconomic boundaries.

There may be more to the article than the press release indicates; Centola is a former MIT professor with a background in computational modeling, so I assume he brings something new to the table here.

Does politics count? Parties emerge in parliamentary democracies across a great many variations of the idea.