The unbelievable stupidity of targeted advertising
I am fairly cagey about what I let track me on the Internet. I use Firefox, I have extensions to further block tracking, and on my computer I also use Malwarebites. I do my searches with Start Page and DuckDuckGo. I decline everything I'm allowed to whenever I get GDPR pop-ups. I regularly check and adjust privacy settings where I can.
But I still get targeted advertising, and lately it has become pretty fucking hilarious.
I didn't screenshot this one, but repeatedly, on the stupid Cat Game I play on my phone, I have had very poorly made ads for dental implants in Turkey. Like, they are still images that were obviously very cheap to buy. Someone is relying on only showing this shit to people they think will be specifically interested.
So why does an algorithm think I am someone who would specifically be interested in dental implants from Turkey?
I'm going to tell you two things that will make you go 'Oh yeah, that's why,' and then I'm going to explain why no human would reach that conclusion.
I have been making a lot of toots on Mastodon about my dentist and doing dentist searches. I have also been searching for maps to places in Turkey.
Clearly a person interested in dental care and going to Turkey, right?
Except this is the context:
My toots are ranting about my ableist dentist and their anti-mask policies that mean I haven't been able to go to the dentist in ages because I am medically vulnerable. They've been very dentist negative and talking about my considerable trauma and anxiety about going to the dentist. I recently lost a filling, so had no choice but to make an appointment. Before I did, I searched for LOCAL private dentists, to see if I could find one with better safety practices.
I had to use Edge to look at one website because it did not work in Firefox. Whatever tracking Edge does must think THE one and only thing I am interested in is dentists.
At the same time, I have been listening to audio lectures on the History of Mesopotamia and the History of the Persian Empire. The courses don't come with maps, so every time they mention a place name, I am running searches on those places for maps so I can visualise where they are and what's going on (I have learnt a lot of geography! I'm probably retaining it better than the history!)
A lot of places in Mesopotamia and ancient Persia are now in modern day Turkey.
Whether it's poor compliance to GDPR (likely - i don't think objecting to (il)legitimate interests stops anything) on the websites hosting the images, or Facebook tracking me when I'm logged out, or crawlers scraping my toots, SOMEHOW I have been tagged as REALLY interested in places in Turkey and probably maps of Turkey, as someone who was going to Turkey might be.
But I'm not. In fact, a human being would infer that someone who's too sick to go to a dentist that doesn't mask is *really* unlikely to be travelling anywhere.
And once you know that most of my searches are about places in fucking Babylonia and Assyria, it's obvious that I'm not even mostly looking at the parts of the maps that are in Turkey.
A human being would NEVER try to target me with dental implants from Turkey.
But whatever auto-buys targeted ad space on Cat Game thinks I am the fucking jackpot.
And then today.
Sweet kittens.
This is the exact same style of ad as the dental implants, btw. Same still image that's the wrong shape to view on a phone. It's a different product, but somewhere along the way, it's the same company that thinks I, an unemployed chronically ill person, would travel to Turkey for dental implants, who also thinks I would want a Rolex that can go under water.
If I hadn't had the first one I probably wouldn't have twigged why I was targeted for this. It's such an unlikely product for me. I hate watches and am aggressively uninterested in expensive ones. So why, why, why...?
I'm pretty sure 'Submariner' is the key.
A couple of days ago I made a rant in which I talked about Margaret Cavendish predicting submarines. I possibly even tagged it with the word 'submarine'.
I have also been talking a lot about my heart rate monitor. But because that's a lot to type out, I often just refer to it as my 'watch'. There was even a post about which wrist you wear your watch on in which I explained in the tags that I hate wearing watches, so I have to keep changing which wrist I wear my watch on.
The confluence of someone interested in submarines and watches could plausibly be considered to correlate highly with the interest in buying a submariner watch by an algorithm.
But to a person it's fucking obvious that I actually hate watches and would never be interested in one that doesn't monitor my heart. And the sum total of my interest in submarines comes from posts about literature that only mention them as one among many signifiers that a classic work is science fiction.
This kind of junk advertising that tries to sell people stuff they hate is what powers most of how our economy works these days. This kind of tagging content as expressing an (assumed positive) 'interest' in something is behind the vast majority of 'machine learning'. I've seen under the hood of how these things work. Just being on a page that contains something tagged as indicating interest in something is taken as useful information.
You visit a page that contains an address, and a computer adds that into the likelihood that you're interested in the city mentioned in the address. You're not. It's just the contact details for the company that sells the rubber dog toy you just bought.
On the vast, generalised scale, you can get interesting info from stuff like this. You can see trends in populations and industries, and tiny, irrelevant info like mentions of cities you're not buying tickets to vanishes into nothing. It's not that the overall practice has no value.
The problem is that the vast majority of advertising that tries to target specific people with specific products is fucking useless. Because meaning is determined by context.
I can't remember if it was Quine or Davidson that pointed this out (Davidson was Quine's student and they were both obsessed with interpretation, translation, and meaning): you can't translate single words meaningfully. You have to translate sentences (do a search on 'gavagai' to find out more, I've talked long enough and don't want to go deep into the indeterminism of translation; it's not as scary as it sounds, though - there are bunnies). Meaning relies on context.
And as Davidson further developed the thought, to be meaningful, the context needs to include embodied, successful communication between minded beings about a shared external world. We have nothing approaching a machine that can do that. And until we do, machine learning is going to keep trying to sell dental implants to people who hate dentists and are learning about Babylonia.
Because machines can't know what teeth and Babylon are, let alone their relations to disabled former academics lying in baths.
2 notes
·
View notes