There are probably more comments I ought to address, but I'm busy and this particular subthread was one that most interested me so I'm slacking off and replying to just this for now. Sorry in advance.
"Of course you're expected to ignore overarching generalizations when speaking to individuals! Don't be blind. Should we call every Israeli we meet a war criminal because a plurality voted for the recent governments? Is it helpful for a female rape victim to hold contempt for and despise every man they come across on the assumption that they too are rapists?
Huh? What on Earth are you talking about here?"
In both of your examples, the generalizations are flawed because there exist confounding factors which mitigate the strength of the generalization. Among the subset of Israelis who voted the most recent government into power, there may well be a group of war criminals; but voting for the most recent government does not prima facie make you a war criminal. There are other factors that make you substantially more likely to be a war criminal, like committing war crimes. The assumption fails not because humans cannot be generalized (they most certainly can be), but because the specific generalization made does not make a pertinent logical tie to the characteristics of the individual(s) in question.
The rape example is perhaps clearer. Let's assume we know the rapist is male. Among the subset of men there exists a subset of male rapists; but being male does not prima facie make you a rapist. There are confounding factors that make someone more likely to be a rapist, like committing rape against other people.
Both examples fail because the tie between the group in question and the characteristics in question is too weak to be useful. If you go around assuming that all Israelis are war criminals or all men are rapists, you'll be wrong well over 99% of the time.
In contrast, let's take a rather straightforward example. According to the INCREDIBLY scientific Ask.com (ha), in 2050 we can expect roughly 5% of the world population to speak English. (http://www.ask.com/world-view/percentage-world-speaks-english-859e211be5634567?qo=questionPageSimilarContent) However, I live in my sleepy podunk hometown, Baton Rouge, LA, and in 2050 I can expect roughly 85% (probably undershooting it, tbh) of the people I encounter in BR to speak English. If I were to run into a human randomly selected across the entire world, it would be a terrible assumption that they speak English. If I were to run into a human randomly selected from Baton Rouge, LA, it would be a very strong one.
That's all that pattern recognition is. It's recognizing that certain people with certain characteristics are statistically significantly more likely than random chance to have other characteristics. It allows me to make useful heuristics that save time in my decision-making and allow me to make more informed decisions.
It goes without saying that race and sex are characteristics which imply similar statistically significant correlations with other characteristics. Yet your post would seem to assert that creating similar heuristics is morally wrong:
"To make an assumption about an individual based on their sex is just as sexist as assuming that a given black person likes rap is racist."
Why is this? The only error here is in making incomplete heuristics that don't account for other factors. This is, again, why you should readily admit your errors when you make them and apologize if your errors offend. But I see no moral error with creating the heuristics in the first place. This is why I asked where pattern recognition fits in your understanding of racism and sexism, because both of those words are typically used in the context of condemning a particular practice, and I think it would be incredibly foolhardy to condemn pattern recognition as a practice, considering that it's not only a useful process when done right, but is additionally something we attempt to do automatically, even subconsciously, all the time.