Where is the Line Between Personalization and Creepiness?

A few days ago, I was checking my Instagram feed and I saw an ad for ‘Barkbox’. I own a dog and have a separate account for her where I post about her adventures. I have been doing that for the past four months, so Instagram was of course well aware of that since for a while now. And so Instagram showing me an ad for ‘Barkbox’ is the least surprising. But the timing was super interesting. Here’s some context. I am taking a class on HCI for which we are studying various tech products and how to improve them. During one of our evening classes, we had a peer review session of our redesign. I was talking to one of my friends from another team who was studying ‘Barkbox’ for this assignment. So we talked a lot about it. Then, right after I came back home, I saw an Instagram ad for ‘Barkbox’. Prior or during this discussion, I had never searched for ‘Barkbox’ and I had never seen its ad before this event. So the timing was quite interesting. This was the first time, I had experienced such perfect timing with Instagram ads (I don’t use Facebook but I am guessing it must be equally good).

Another example from several months ago was of this company called ‘Rotimatic’. We were visiting a friend in Seattle who had Rotimatic. We talked a lot about it since we were considering to buy one (this time, I had definitely searched for the product but much earlier). Three days later, we came back and I saw Rotimatic’s ad at my home, not while we were in Seattle but at my home. But the ‘Barkbox’ experience felt super creepy and like so many others, I thought Instagram was passive listening to conversations through mic. Not that it was transmitting data but just processing it locally and using keywords to enhance ad targeting. A quick search online brings up several articles around this.

Facebook has multiple times denied any such doing on their part. Their ad execs even took to Twitter to defend and officially state that it does no such thing. And to be honest, I believe them. I don’t think Facebook is actually listening through our microphones. But it does give that illusion because it’s so damn good in ad targeting, so good that it feels too good to be true. Gimlet media’s Reply-all podcast does a great job of explaining how Facebook can be so damn effective at ad targeting. You can give it a listen here.

This led me to think how we perceive personalization of content and services. Marketing has been personalizing ads and content since decades based on our data. Store reward cards is the simplest example. We all know about how Target’s coupon department figured out a teenage girl was pregnant before her father did. There are hundreds of such examples in the wild. Ad targeting that Facebook does is nothing new. But never before has any one single company harvested such huge amount of personal data of every user (yes, Google has vast amounts of data too). Because the amount of data is huge, you need machine learning models to make sense of this data. The more information about a user the model has, the better it gets in narrowing the list of ads most suitable for that user. Reply-all says that Facebook assigns every user about 50,000 categories. Imagine how insanely accurate profile you can build. I don’t think even I can lay out 50k variables about my own self.

I definitely don’t like when I see the pharma companies ads on TV channels. They are totally irrelevant to me. But my only reaction to those is of boredom and/or frustration. On the other hand, I felt creeped out like crazy when incidents such as ‘Barkbox’ and ‘Rotimatic’ (and a few others) happen even though those ads are extremely relevant to me. In this case my reaction is of disgust, scared, creepiness. Much stronger emotions than boredom or frustration. From a user perspective, if I had to pick one between those two emotional states, I would always, always go with boredom and frustration.

But how and where exactly do we draw a line between these two states where there’s that perfect equilibrium of relevancy and privacy? These ads are powered by deep learning models and neural networks and if you were to ask Facebook why you were served that specific ad, even they wouldn’t know the exact answer. It’s beyond human capacity to understand these models at that level.

Humans simply cannot comprehend the model the computer has built of itself. Yet it works.

Adam Geitgey

P.S. Checkout my friend’s blog on this same experience.

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.