Humans, being social animals, enjoy a tremendous survival advantage from their ability to recognize and remember large numbers of faces. Consequently, there are special neural systems dedicated entirely to this task. There is even a recognized disfunction of these systems, prosopagnosia, which makes the subject unable to recognize or remember faces, even those of his nearest and dearest.
The idea behind these flash cards is to exploit the natural capacity of the human brain for remembering faces to enhance learning and recall. The hypothesis is that concepts associated with human faces will be better remembered and recalled than concepts without facial associations. It remains to be seen, through research, whether or not this is demonstrably true. If so, the technique of remembering concepts by facial association could prove to be even more powerful than that of remembering by architectonic association. (Cf. Jonathan D. Spence's The Memory Palace of Mateo Ricci for examples of this technique.)
If the concept proves successful, the applications extend far beyond a simple set of paper flash-cards. The number of possible human faces is in the billions, and computer generation of faces has already reached the photorealistic level, as the popular face generator website well demonstrates. The cartoon-like faces on this page were generated with the more limited Human Identikit, but the number of possible faces is still enormous. The faces of human characters in the popular game The Sims are generated using an evolutionary algorithm, and a software package for soing so is available here. These applications may pave the way for a computerized learning program which uses facial associations to enhance recall of verbal, mathematical, or other abstract concepts. Such a program could generate faces in real time for association with certain datasets, in accordance with its memory of what sorts of faces are best remembered by a particular student.
I am by no means the first to suggest exploiting the innate human capacity to recognize and remember faces as a convenient means to represent and recall information. Herman Chernoff, in a June 1973 paper entitled, "The Use of Faces to represent Points in k-Dimensional Space Graphically," in the Journal of the American Statistical Association is, to my knowledge, the first to propose the idea. Statisticians, like Chernoff, face a daunting problem in the graphical representation of systems with many (say, 5 or more) different variables on the 2-dimensional surface of a page or screen. Chernoff's proposal, essentially, was that the various features of faces could be generated mathematically by the variables of the dataset. Isolating anomalies, even in complex systems, might then be as simple as looking for the "strange face in the crowd," as it were. Edward R. Tufte's groundbreaking The Visual Display of Quantitative Information has lately drawn renewed attention to Chernoff faces. Tufte points out that a Chernoff face need not have symmetrical features and that, not only does this approximately double the number of variables a single face can represent, it might also--because facial symmetry has been proven to correspond to attractiveness--allow us to ferret out anomalies by looking for particularly "ugly" or "beautiful" faces.