This seems to me like it would be an interesting hypothesis for a an experiment, but a bad idea to suggest/encourage in any "real" (IRS) capacity. Academically the idea is interesting, but the thought of this being actually implemented is a little scary.

... but you could say the same thing about a lot of other areas of research, too. Face detection at Facebook's scale, for example: really cool academically and potentially really scary in application.