Harlo has a blog.

Do not do what Donny Don't did

[24 Jun 2015] [tags: eating_crow medicine ]

Or, the one in which I apologize to Airbnb.

A few weeks ago, I presented a talk at the Eyeo conference called “I will find my people everywhere.” In it, I discussed the many bizarre, funny, and sometimes disturbing ways people find one another online. I covered a lot of ground: I talked about my realization that my own research eerily resembles XKEYSCORE; how advertisers miss the mark with retargeting but we could subvert it in funny ways; how the virality of language can be analyzed via the user dictionaries in each of our cell phones; and how we should beware digital redlining, because “black box algorithms” sometimes go awry and can cause subtle but unwanted effects.

It is the last bit that I wanted to talk about. In that chapter, I went through the really great research by Latanya Sweeney, and The Nation’s Astra Taylor and Jathan Sadowski, and followed it up with the most grave factual error. Let me explain, and hopefully, apologize.

I capped off this chapter of the talk with a video as “proof” of Airbnb using similar techniques. In 2013, I noticed that there was a discrepancy between the price I was shown and the price a unit actually costs at the moment my Facebook profile loads its information onto the page. I publicly (mis)charactarized this as a jump in price. (This discrepancy actually comes from a strange calculation of the cleaning fee that’s tacked onto each potential rental, and I was shown three different prices for the same potential sale.)

I suspected that the discrepancy was linked to Facebook, and thus, reputation management. And pushed further to say that I might have been either incentivized or penalized with a different price because my profile indicated something about me. While I did say that the discrepancy could go either way (and you’ll notice that, I was in fact incentivized by being shown a $2 discount!) I gravely mischaracterized this practice and absolutely lead an audience to infer something super pernicious in this practice.

Also, I relied on anecdotal “proof” rather than real, concrete proof. My eyes are not a legitimate debugging console, and therefore it is not responsible of me to present the just-in-time loading of tracking data into the page as proof of a cause-and-effect relationship in price adjustment. It could have very likely been the result of a specific campaign, or even A/B testing, which is absolutely that company’s right to do. Crafting experiments like those certainly pose a fun challenge for a team of data scientists and marketers who are doing a lot of fascinating and important work on how subtle changes in users’ perception can, when presented in a certain way, sway consumers to spend their money.

Finally, my framing of the issue was reckless and, quite frankly, horrible. It was definitely not my intention to insinuate that such a discrepancy was racially motivated (I made a specific clarification) but this “weak sauce” anecdote was placed in a chapter on digital redlining, and that carries a huge implication about how socioeconomic class, algorithmically inferred by personally-identifying data, come to bear on what we are shown online. This is a real thing that happens to us all; some of us benefit from it, and some of us are hurt by it, but by no stretch of the imagination did Airbnb participate in it.

It horrifies me that I left the anecdote sit in the air without bringing it to any real conclusion. (As I avowed on stage, I had absolutely no conclusion to give). I did not have any intention to infer that kind of injustice, but it was foolish of me to underestimate that the very optics of that chapter in my talk could easily lead viewers in that direction.

I’m deeply saddened that, instead of uplifting this type of research, I could have weakened its integrity. And also, I’ve hurt the integrity of those with legitimate grievances with the “sharing economy” and fight for it to stay more humane. I should not have contributed noise to what has always been a really clear signal. So, yes there are targets, but I had no right to use either this anecdote or the Eyeo stage for target practice.

And I deeply regret it. My mother always said, “you don’t apologize to make you feel better, you apologize so you will do better,” and these words have never felt more true.