Behavioral Targeting is Getting Scary
This fascinating article about how an “election management firm” — under contract with the GOP — mined social network data and used it to bombard impressionable voters with content designed to manipulate them is thought-provoking and troubling.
Go ahead, read it first, and then come back. Also, for balance, read this counterpoint from Bloomberg doubting the data’s effect on the election, and listen to this podcast from Michal Kosinski who is the Stanford researcher featured in the article. Kosinski is unsure what effect data mining and targeting had on the election, but has plenty of fascinating things to say.
It’s always eye-opening to me when I ask Trump voters why they voted the way they did. Some of these people are friends and others are taxi drivers or other people I’ve met randomly. Their responses usually start out with something negative like “I could never vote for Hillary”. Then, when you probe just one level deeper and ask why, they will say something like “she had that investigator murdered”. Then you ask why they believe that to be true, and it usually ends up with some form of “it’s all over Facebook” or “there’s a whole video explaining it”. Granted, my sample size is small and doesn’t represent all the reasons someone might vote Trump, but I always leave these conversations thinking to myself “how could someone actually believe that?” Turns out, they may believe it because it’s been hammered into their news feed, repeatedly, by a company who may know them even better than they know themselves. (The research in that link is amazing, by the way: by analyzing only 10 Likes on Facebook, computer models can predict your personality better than your work colleagues, and by analyzing about 300, better than your spouse.)
We’ve known that fake news and political propaganda have been a problem on social media for awhile, but until reading about how psychometric targeting actually works, I viewed the phenomenon more as opportunistic chaos than anything else. Now it feels a lot more like a large-scale data-driven campaign to manipulate voters.
On the one hand, you can look at these methods and say they are no different than the targeting we do when we sell products to people. But on the other hand, there is a difference between selling someone a toaster they don’t need, and actively poisoning someone’s ability to tell truth from fiction.
Technology like this, as well all the other forms of hyperpersonalized targeting, really make you question your participation in platforms that collect and expose your data. It’s no longer just a question of “will that Amazon ad keep following me around the web until I buy the shoes?”. Now you legitimately have to worry about your own government, and private parties wanting to be your next government, trying to target you based on your vulnerabilities. Your social data is the tracking cookie, and your vote is the conversion.
I vividly remember when I first saw Take This Lollipop. It was an amazingly powerful demonstration of the harm one person could theoretically do to one other person using basic Facebook data as fuel. I think with the 2016 election though, we just witnessed the social data bomb explode at world-scale. Perhaps even more troublingly, this form of weaponized data science is just getting started.
I think it’s too easy to let the tools and networks we build off the hook by claiming they are just reflections of what we, as humans, already want. If they encourage us to self-segregate in ways that not only create perspective-shrinking echo chambers but now expose us to political targeting by outside agents, it’s hard to see how that is something people actually want, or something that is good for society.
It is now more important than ever to ask ourselves not only what good the technologies we are building (and using) can cause in the world, but also what bad they may already be causing, either by accident or on purpose. The goals of any technology — however noble — are never as important as the effects.
(This post also available on Medium.)