<p>Many animal groups display self-organized collective patterns of motion. Human crowds also offer various examples of self-organization, including flocking behavior similar to that seen in fish schools and bird flocks. While previous studies on human flocks or swarms have investigated the influence of individual vision on local interactions, the global properties and the change in interaction network at the group level remain elusive. Here, we show that human swarms can transition between multiple collective states, and their interaction networks are not fixed in time, similar to other animal groups. We tracked trajectories of participants walking together in an experimental arena and observed that they can exhibit well-organized collective behavior. In particular, human swarm composed of fifteen individuals showed transitions between two ordered states: polarized and milling groups. Individual noisy movements may contribute to these transitions. Indeed, we found that individual participants continuously change their relative positions with respect to their neighbors in a directed group, allowing for interactions with various neighbors. We expect that our findings provide insights to understand fundamental mechanisms underlying behavior in human crowds and nonhuman animal groups.</p>