Paramotoring for the Paranoid: Google’s AI and Relationship Mining
My son approached me the other day with his best 17-year-old sales pitch: “Dad, I need a bucket of cash!” Given that I was elbow deep in suds doing the dishes he neglected to do the night before, I mentioned that it was a singularly bad time for him to ask for anything.
Never one to be dissuaded, he plunged ahead with the reason for the funding request. He had stumbled upon a series of YouTube videos about paramotoring, and it was love at first sight for him. He waxed eloquent about how cool it would be to strap a big fan to his back and soar with the birds on a nylon parasail wing. It was actually a pretty good pitch, complete with an exposition on the father-son bonding opportunities paramotoring presented. He kind of reminded me of the twelve-year-old version of myself trying to convince my dad to spend $600 on something called a “TRS-80” that I’d surely perish if I didn’t get.
Needless to say, the $2500 he needed for the opportunity to break his neck was not forthcoming. But what happened the next day kind of blew my mind. As I was reviewing my YouTube feed, there among the [Abom79] and [AvE] videos I normally find in my “Recommended” queue was a video about – paramotoring. Now how did that get there?
There could be an easy explanation for this, of course. It’s no news that Google reads our email and sends ads based on keywords using AI techniques. [Mike Szczys] recently wrote a piece about Google’s somewhat “weak AI” that does a good job of reminding him when to pay a bill by reading the billing statement emails, but fails to realize he has actually cut the check because it hasn’t figured out how to close that loop. Google’s semi-creepy email reading would be the obvious explanation here, if only my son had sent me a link (as an aside, I think a parent saying “Send me a link” has become this generation’s version of “We’ll see.”) But he didn’t, so email clearly wasn’t the vector that Google used to target me.
A text perhaps? Nope, we didn’t exchange any texts on this topic. In fact, until I started writing this article, I had never in my life typed the word “paramotoring” into any kind of device. So it’s pretty clear that Google wasn’t looking over my shoulder while typing. The only thing we ever did was talk about the topic, so that leads to the possibility that Google is listening to our conversations.
I had my phone in my pocket during his initial pitch – did Google hear us talking about paramotoring and decide I needed to see a video on the subject? Some people seem to think it’s possible, and if so, it’s a little creepy. I know that Google’s assistant is always listening for me to call out “OK Google” and respond to the request. I’ve gone through my audio settings and found plenty of queries, like when I asked for the Superbowl score, or just last night when I asked Google how much protein is in four ounces of chicken breast during dinner. But there’s no record of the word “paramotoring” being uttered. Clearly my phone heard us talking, otherwise it wouldn’t have been able to respond to my other requests. But if it did listen in on our conversation without divulging it, that would be troublesome.
Like Father, Like Son
Luckily, there’s a simpler way to explain this incident, although in my opinion it’s not much less creepy. It could be possible that Google made an association between my son and me based on our contact lists, decided I would be interested in the same videos as he is, and threw a paramotoring video into my feed. That wouldn’t be too hard to program, and it’s actually pretty slick, although I’m not sure either of us wants the to know all the details of the other’s video preferences. Sometimes a little mystery is healthy for a parent-child relationship.
Could Google be digging even a little deeper, though? Could Google’s AI have determined from my son’s single-minded pursuit of paramotoring videos that he was ready to buy but lacked the means, and mined his contact list for the most likely person to have access to sufficient funds? If so, that’s pretty sophisticated stuff. It would need to discern personal relationships, figure out the financial situation of the parties, and target the pitch in an infinitely more subtle way than my son’s “bucket of cash” approach.
Have I inadvertently provided Google with the information needed to make that leap? Probably. It would be pretty trivial to figure out from my contact list who my wife is, and who my kids are. We all have the same last name and the same physical address, and from the context of emails it’d be easy to figure out the family hierarchy. Google clearly has access to my financials, too, at least indirectly.
Like [Mike] mentions in his article, Google reads our billing statements for things like car and auto insurance, and could easily infer from the amounts due what your car payments would be, and how much of a mortgage you’re carrying. From there it’s an easy extrapolation using typical debt-to-income ratios to figure out how much I make. My ZIP code provides valuable demographic information that would make it easy for Google to do a reality check on its estimate of our household income.
Google even knows my spending habits – surely it could not have missed the Amazon receipt for the new laptop I bought last week. Put that together with hundreds of Amazon purchases through a typical year, and Google must have a pretty accurate picture of me. Can all this put together be enough to build a map of my family so that Google can put the virtual touch on me for a $2500 paramotor? Sure seems like it.
To be honest, I don’t know which is worse – mining my emails and social media to tag me as the financial head of the household, or outright eavesdropping on me. The eavesdropping just seems like a brute-force approach, though, and I really don’t think that was the vector in this incident. If it was, though, there would be a simple fix – I could just shut up about the things I don’t want the Hive Mind to know about. Putting together an accurate and perhaps unflattering picture of someone from the bits of digital flotsam of our online lives is far more subtle, though, and far more difficult to avoid just by clamming up. That makes it potentially far more powerful, and I think far creepier.
Filed under: Curated, Featured, Interest, rants
from Hackaday http://ift.tt/2ndAtEt
via IFTTT