Archive for November, 2015

Algorithms Without Empathy

I’m depressed.  I know this because there are days where it take all my energy just to get out of the house.  I know this by the way everything can feel so distant, like it’s happening to someone else.  And I know this because my phone is telling me so.  Not because I have a fancy mood tracking app. A couple weeks ago my phone autocompleted the entire phrase “I’m not doing so well,” which, for someone like me who struggles with talking about depression, translates to something between “I’m having a really bad day” and “I’m about to fall apart.” It was not a good feeling.

Part of having depression (for me, anyway) is having a little voice in the back of your head that hijacks your internal monologue in a way that distorts reality.  There’s a story that’s a bit of an Internet Classic which explains the inherent tradeoffs involved in living with physical disability by way of an analogy: you start the day with a certain number of spoons and everything you do (getting dressed, going to the doctor, work, having fun, etc.) has a “cost” in spoons.  In order to do everything you need to do (and maybe some things you just want to do) you have to plan your day around the number of spoons you have left.  I’ve found it to be an apt analogy for living with depression as well.  I strongly encourage you to read it, but if you haven’t, trust me when I say that dealing with depression is often a give and take between short and long term self care.  Case in point: I am absolutely convinced that talking about my depression is important for my long term mental health, but very often I make the decision not to talk about it.  Sometimes for better reasons than others, but almost always at least in part because I think it’s the best way to take care of myself — at least in the short term.  Watching my phone autocomplete that phrase was just the opening that voice needed to let me know how bad of a job I’m doing, not just at taking care of myself, but at pretty much everything else as well.  Being outed as a liar, no, a self-destructive liar, no, no, a lying failure by my phone is a peculiarity of two algorithms: the machine learning algorithms in the autocorrect software and the biochemical one running in my messed up head.

I’m sure the programmers and machine learning experts that made my autocorrect software never intended to ruin my day. Just as they never intended to enable the creation of those surreal artistic pieces that are autocorrect poems. As a species we’ve yet to find a tool we couldn’t abuse in some way its creators never expected.   (I guess that’s the flipside to the old saying “when the only tool you have is a hammer, everything looks like a nail.”)

I don’t know what those programmers could have done to prevent a situation like mine.  I don’t know if they even should have. I do know that it hurt to see those words repeated back to me. It kicked off another wave of depression that engulfed me for the rest of the day. It also made me realize how bad I am at talking about my depression, even to myself. I’m sure that’s not something they ever intended. I get notifications regularly congratulating me on saving another ten thousand keystrokes — not once have I gotten one for speaking honestly. But it’s becoming increasingly clear that good intentions aren’t enough.  Take, for instance, this story from the end of 2014.  The designers and engineers of Facebook’s new “year-in-review” feature blithely assumed that people only share happy things on Facebook, and that the only kind of year one can have is a good one.  Instead of respecting Eric Meyer’s choice not to make a year-in-review, Facebook pushed a picture of his dead daughter surrounded by partying cartoon figures exhorting him to share his wonderful year with others.  In an example slightly closer to home, my mother-in-law, who died in 2013, still appears in my wife’s feed no matter how what settings she selects on her mom’s account.  These algorithms aren’t just cute pieces of code that make our lives easier. They have real effects on real people’s lives.  They are pieces of ourselves distributed across platforms and devices and it’s time that we as engineers and designers take that responsibility seriously.

These algorithms are becoming increasingly complex and sophisticated in anticipating our desires — in emulating us. As we walk down this road with our algorithmic simulacra, it is urgently important that we consider empathy in their design. For better or worse we are slowly, painstakingly creating software in our own image.  In doing so, it would be a terrible shame to let a belief that these algorithms are somehow neutral to guide us towards removing our own humanity.

Thanks to Deana Rutherford and Peter Raleigh for reading drafts of this essay.



November 2015
« Oct   Dec »