“Observation of Phase Transitions in Spreading Activation Networks”, Jeff Shrager, Tad Hogg, Bernardo A. Huberman1987-05-29 (neural net)⁠:

Phase transitions, similar to those seen in physical systems, are observed in spreading activation networks. Such networks are used both in theories of cognition and in artificial intelligence applications.

This result confirms a predicted abrupt behavioral change as either the topology of the network or the activation parameters are varied across phase boundaries.

…Our results clearly show that the predicted phase transitions can be observed even in relatively small spreading activation networks. Moreover, the existence of such transitions has immediate implications for the predictions of memory models and the behavior of artificial intelligence systems which incorporate learning. Specifically, instead of the fairly smooth transitions in performance that have been generally assumed in these situations, we have shown that abrupt transitions can be expected. More generally, these experiments show that statistical models provide a useful way to understand the behavior of large systems. They also emphasize the dominating influence of topological properties. These are particularly important implications for the behavior of any network with a dynamic topology.