Recent empirical results on the structural properties of large social and information networks demonstrate that these networks are particularly 
ill-suited for analysis with many traditional machine learning and data
analysis tools.  At root, this has to do with the fact that the relationship between structures that may be interpreted as geometric and
structures that exhibit empirical signatures of quasi-randomness is  
substantially more complex in large social and information networks than
it is in many more traditional classes of data.  After briefly
summarizing these empirical results, as well as how they were obtained, I 
will describe some of the implications of these results for the 
algorithmic and statistical modeling of large informatics graphs, as well 
as for performing machine learning and inference on these graphs.