A millisecond's thought will tell you that Google engineers aren't looking at every email (or a substantive fraction). That doesn't mean they aren't looking at any email, which is all that is necessary for that interpretation.
The latter interpretation, if I understand it correctly, is that Google engineers randomly poke at people's emails and in this case just happened to get lucky. Even stipulating thousands of Google engineers doing that, that doesn't sound statistically plausible; the base rate of child pornography among Google Mail users is extremely low.
That was my thoughts as well. But I don't think most people think that way, especially given the misleading wording of the article. I hate to contribute to the "rest of the world is stupid" theory, or pretend that I'm immune from such things, but in this case I really do think the vast majority of people won't evaluate things in a probabilistic way like that. Most people just aren't trained to think in terms of odds instead of emotions (and it constantly annoys me.) At least that is my experience, and it may not be representative.
Anecdote, I go swimming in a river known to have Caiman (small alligators) downstream and everyone thinks I'm a huge risk taker. Nobody comments on the 3 hour drive to get there and back in heavy rain and high speeds. It's obvious, mathematically speaking, which is the riskier activity.
The base rate of child porn could be low, but the predictors for who might be likely to be a pornographer could be fairly high. With the social graph being a primary predictor, but also, say, things such as browser profile, IP, search and browsing history, locations, communications, and other information. Which Google has on individuals in spades.
So: should Google be performing predictive analysis to make precrime forecasts? Has Google been doing this?
