> But the ads I see are way off target.

The two sides see "off target" quite differently. If the baseline is 0.1% accuracy, and an ad platform offers 1% accuracy, then you see relevant ads showing up at 1 in a 100 rates rather than 1 in a 1,000 rates. You still feel like you're being spammed by useless nonsense. As an advertiser, my cost to reach my audience just dropped by 90%.

Google News for the longest time kept showing me updates for the Fremantle Dockers, an Australian football team (I'm Swiss, and never set foot on Australian soil). I wonder why...

Do you use / search for information about Docker (the container platform)?

Yeah I've gotten some ultra disconnected bits too.

I'm really curious how those happen.

well in this case I assume that my frequent searches for Docker (the technology) led Google News astray :)

Wow, that's super random. As an Australian, hope you enjoy Aussie Rules!

Maemo user? :D

I can’t help but think the only thing most people buying ads go for is demographics (age gender race income) rather than more personal interest targeting, since that’s been the primary driver of TV ads for the past many decades. Interest targeting seems like it doesn’t work, or if it does the interest based ad buys are few and far between.

That's an interesting point. I wonder if ultra granularity ... isn't really the current state...

At one point it was sure that I was female (am not, have never been) too.... to their credit they figured it out after a year or so, but man the ads. But that was at least a limited run of confusion.

Or "let’s sprinkle some noise in here so people think we’re less efficient than we truly are, so that we appear a bit dumber and less creepy."

But probably the truth is that these systems aren’t that good :)

Oh they definitely do this. At least the traditional loyalty-card/coupon-enabled physical targeted ads do add enough random stuff that you find it non-creepy, after they had a few high-profile cases back in the day where they started targeted advertising of baby products to teenage girls, and parents got mad. The kicker was they turned out to frequently be real cases of teen pregnancies.

I remember that, it was Target wasn't it and they inferred her pregnancy status because she started buying unscented deodorants and stopped buying sanitary products.

Funnily enough when I'm discussing the dangerous nature of large datasets in the hands of companies/bad actors (sometimes the one and the same...) and the "creepy" effects this is one of the examples I use, I've found that if you know a little about the person and couch the argument in a way they can personally respond to, they understand even if they don't know the first thing about technology generally.

EDIT: https://www.forbes.com/sites/kashmirhill/2012/02/16/how-targ... for anyone who hasn't seen it.

The have to put in some out of context ads sometimes, just so they can train their system.

For example they may show you an ads about hairdryers after you searched for computer keyboards. There is absolutely no relationship between the two, but Google doesn't know it yet, so they try. After all, there may be a correlation. In the likely event where there isn't, Google will simply stop showing hairdryer ads and show you computer mice instead (a known correlation) or running shoes (another experiment).

You'd think they'd sprinkle in something less ... annoyingly skewed that obscure what I actually want to see.