TECH

Facebook Is Paying Closer Attention To Your Posts Than You Might Think

The social network sometimes makes startling suggestions with its search function.

Facebook recently suggested that I search for my fiancee's little sister "licking batter."

I am neither kidding nor surprised, but yeah, it's kind of weird.

At the very same time, it also recommended that I search for a dear friend with the keyword "nudist" attached. This was a bit surprising, as I've only seen him fully clothed, but hey, new year, new you.

Here's why each of those two things happened: Facebook is deeply invested in developing superior "machine learning" techniques. It wants its software to recognize things about data -- your words and photographs -- that make sense to people but not necessarily computers. Think about the way you talk and read facial cues, for example: How do you get a device to work with language and expressions in a way that feels human? That's tricky, and suggested search is just one small part of Facebook's overall machine learning efforts.

Lord only knows why Facebook decided late Monday night was the time for 'licking batter' and 'nudist.'

There's really nothing new about this feature, though I can't remember a time when it startled me more. Essentially, it generates suggestions according to what you've searched for before and what you've done on Facebook. I probably have searched for both of the people Facebook suggested to me, and I've definitely interacted with their posts before.

Facebook itself has been candid about its machine learning mission in the past. For example, its head of artificial intelligence, Yann LeCun, once discussed algorithms that would allow Facebook to recognize when you look intoxicated in a photo you're about to post and suggest you do otherwise. Obviously Facebook hasn't implemented this feature, though it could happen some day.

When curiosity got the better of me and I did tap to search for little sis licking batter (there is no possible way to scrub my cache hard enough), I found that she had posted a photograph of her cooking over Thanksgiving, captioned in part "I am licking batter off my hands." (The "nudist" post was a cool diorama my friend's partner made, depicting naked people frolicking in nature.)

It seems simple, but it's not. Facebook was able to recognize the action in a sentence -- "licking batter" -- and recycle it as a term and put it in front of me as a bit of information it imagines I might be compelled to click. It means Facebook knows who I care about, can recognize a verb and an object and put them together in a suggested search. This is complex enough that there's a whole body of research surrounding whether computers can ever really understand human language.

Lord only knows why Facebook decided that late Monday night was the time for "licking batter" and "nudist." Many people previously "liked" both posts, perhaps signaling to Facebook that they might grab my attention.

(Not actually my fiancee's little sister.)
(Not actually my fiancee's little sister.)

Google uses similar technology in its new "Photos" app. The software is smart enough to recognize individual people, identify types of animals and so on. It's not always perfect, but it's pretty darn good. Anil Sabharwal, Photos' lead product manager, told The Huffington Post last year that the app would eventually be able to recognize an individual's facial expressions and correlate them to emotions -- in theory, you could one day search for something like "happy photos of grandma" and get exactly that.

Chances are, there are noble enough justifications for all of this: A Google or Facebook service that can operate as a sort of "second brain" could be a blessing to people with degenerative ailments who struggle to recall past events, for example. Less dramatically, you might just feel kind of nice if you're able to search for "pictures of me laughing."

But realistically, it could someday be a powerful way to market to individuals. When you're bored, you scroll through Facebook. The social network may have grand ambitions to deliver Internet access to far-off places via drones and lasers, but when you get down to brass tacks, it's a place for a lot of people to willingly -- eagerly -- deposit a lot of data that a lot of other people mindlessly look at. Facebook already hauls in cash based on demographic data -- who knows what tomorrow brings if (when) it attempts to manipulate your actions with more abstract data like your captions and photographs?

We don't advocate for the tinfoil hats. Facebook has done a lot of good with its information! And it's all data you provide on purpose. Still, it's always worth considering how the apps and services you use every day evolve on a continual basis. Even your toss-off jokes in an innocent status from weeks ago are fodder for someone else's experience, data that folds into other data in that big brain in the cloud.

CONVERSATIONS