[underwater]

I'm sure Facebook would like to do what you're saying, but I think that they're a long way from having those capabilities.

They have billions of users, billions of ads, and trillions of pieces of metadata. Sifting through that to produce a guess like "User 1234 burned themselves and would be interested in product 5678" would be an amazing and scary piece of AI.

Not to mention that Facebook are limited by the ads they can sell. The manufacturer of the burn cream doesn't give Facebook a bucket of money and free reign. They choose how and where the ads are shown. There is currently no option for "Users who recently burned themselves".

[tfinniga]

Last year there were 29,000 categories that could be targeted. If an ad sales rep wanted to close a deal, how difficult would it be to add one more for recent burn victims, or people who recently bought a specific product?

From the links that m_ke posted, facebook already has ties to loyalty card information, so it's very possible that they didn't need to do any inference, they just had the data directly.

In any case, the main point is that facebook doesn't need to listen to what people are saying, it has a ton of other data streams that could explain the stories in the link.

[hellweaver666]

What would the point be in advertising a burn cream to someone that just bought a burn cream? It's not like it's a common reoccurring purchase (like Toothpaste for example).

Like the article says, it's probably just the Baader-Meinhof phenomenon* at work. It's like when you buy a new car - you suddenly start seeing that same model of car everywhere.

* https://www.damninteresting.com/the-baader-meinhof-phenomeno...

[lolc]

I think the proposed logic here is "a person like you buys burn cream so we advertise burn cream to people like you." In this scenario the person who actually bought the item will get the ad too.

To me, advertising a product to people who already bought it is a sound strategy. I see no contradiction. I'd rather like to hear a cogent argument as to why that is wasted effort, as is usually implied.

In your example, maybe I left the bottle of burn cream at the office. So I will want to get another one on the way home. It's effective to get reminded in my Facebook about that item (and the brand) isn't it? Yes it's creepy but it works.

[reaperducer]

> facebook doesn't need to listen to what people are saying, it has a ton of other data streams

The same could be said for any of those data streams. But just because Big F has access to those other sources of data doesn't mean it excludes voice recognition.

I'd be very surprised if FB wasn't in this arena. That's how big tech companies work these days — they do what everyone else is doing. Like getting into self-driving cars.

Apple[1], Google, Amazon, MS, etc... all have voice recognition boxes. FB doesn't need to market a box for your living room, it already has one in everyone's pocket.

([1] Apple says it doesn't use any of your Siri words against you. If you don't trust FB's word, you might no trust Apple's, either. But at least on the Mac you can keep dictation off the internet, if you choose.)

[macNchz]

I agree that even with an enormous amount of data there are limits to what facebook can realistically infer...I particularly enjoyed the things that facebook’s ad platform determined to be my hobbies when I checked recently: https://m.imgur.com/nWCWn63

I’ve had an account since 2005, even if I use it much less than I used to, you’d think they’d have enough data to do better than that.

[jpalomaki]

Think smaller. They don't need to consider billion users at once, for modeling you can start with hundreds of thousands or few million. Trillion pieces of metadata can be reduced to tens of thousands of features (eg. instead of feeding in the raw status updates, assign each update to some class, put in some mood measurement etc). This reduction does not need to be fully automated, it can be human assisted (humans coming up with ideas on what kind of features to extract). Instead of considering all possible ads, you can start by taking out the long tail of ads with very little interaction data and focus on some specific categories.

Building this kind of predictive models is not black magic. This is what companies do to figure out who they should target on direct marketing campaigns. Applying some magical deep learning dust can quite likely improve the results (in Facebook scale), but it's not mandatory. Computationally the hardest part is building the models. Once they are done, applying them to millions and millions of customers is more straight forward.

When considering what is possible, you need to also consider the stakes. Facebook ad revenue is something like $10 billion per quarter[1]. To make more money, I can think of three ways: add more users, make users spend more time looking at their feed or make more money per ad slot. Better targeting means more money per ad slot since some customers are paying directly based on the clicks. Since improvements lead to better revenue, extra hardware costs are easy to justify.

[1] http://www.adweek.com/digital/facebook-raked-in-9-16-billion...

[halflings]

Patterns like these can definitely be inferred by machine learning, with well principled models.

P(product_class=bandage | job=factory_worker, pharmacy_visit_last_month=1) >> P(product_class=bandage)