We live in a world full of statistics, but our every day experience would give us guesses about cohorts of people anyway. Some people "look like a Kyle", whatever that means, for instance. I've never figured out what I was noticing when I thought someone looked like a certain name – maybe nothing, maybe an amalgamation of previous people I'd met with their features. Considering how strange that idea is, I'm sure if we knew the reasons we thought it, we'd laugh off the idea. And yet, there are some definitive Chads out there – for some reason I can develop moderate confidence in something like that.

In a Bayesian sense, we ought to have non-zero updates upon learning any new information, which means we probably walk around making very small assumptions about people all the time. Little things we'd throw away given any information at all – that person looks fit, that person might read, that person seems confident, that person seems sad – all pulled from weak guesses and tiny signals we don't invest in because they aren't reliable. A single look, or a word of conversation, would likely wipe the slate clean, but even if people were utterly inscrutable, you'd have the microscopic spurious correlations we walk around with.

Of course, real statistics are different – they're sharable, which isn't quite true of "seems like" intuitions, and are the result of efforts that should make them more likely to reflect reality. But that extra care and authority means we have a responsibility to disaggregate carefully. I've mentioned how unimpressed I am with graduates of 'elite' universities – perhaps the graduates are more useful to the world in aggregate, but if someone mentions their fancy degree to you, they're probably a little stinker. The statistic makes no statement about the individual, really – or, if it makes a statement, it's a statement so weak we ought to not make it to each other. And you can quickly bump into a Simpson's paradox when you split the statistical groups into 'people who talk about it' and 'people who conceal it'. Bragging about something like a degree is mostly useful for those who can't prove their competence any other way.

This idea, which I guess people have been ignoring in their breathless consumption of statistical information, is why I'm mostly disappointed in Taleb's "IQ is Largely a Psuedoscientific Swindle". I don't really disagree with most of his points, I'm just deeply saddened that a 'public intellectual' has such poor statistical literacy. You aren't supposed to learn much information from someone's IQ score. It's for aggregates, like checking if early childhood nutrition helps brain development by comparing IQ scores for your treatment group against wider community averages. A single person's IQ really isn't even supposed to be interesting. And of course it doesn't strongly correlate with income! If it did, we'd just use income as the metric and save ourselves the testing.

Taleb is someone who has written a book about statistical models, and he's deeply confused about what we ought to expect out of them. IQ isn't a swindle, and the only thing psuedoscientific in that article is the expectation IQ should tell you a lot about the individual involved.