Based on a complexity analysis of mutual information transmission of EEG developed by us [Xu J, Liu Z, Liu R, Yang Q (1997) Physica D 106: 363-374], dynamic processes of the complexity of mutual information transmission in human brains were studied. To diminish possible problems due to coarse graining preprocessing, some new measures of complexity were used. The results show that, just before and after generalized seizures, the complexities of almost all information transmission between different brain areas drop significantly; there is also a temporary decrease of complexity when subjects shift their attention. The above facts suggest that there is a transient decrease of information transmission complexity when brain state changes occur suddenly. Mental arithmetic tasks activate the left temporal lobe to exchange more information with other brain areas. The results hint that the methods used here might be an approach to observe quick processes in the living brain.