Why Facebook's "It's too hard" excuse for Vietnam war photo takedown is bullshit

On Friday, Facebook started deleting posts containing "The Terror of War," Nick Ut's photo depicting a young Vietnamese girl fleeing a napalm attack on her village; Facebook approach this photo with a scorched earth (ahem) policy, even deleting it when it was posted by the Prime Minister of Norway.

Facebook's excuse for this is that "it’s difficult to create a distinction between allowing a photograph of a nude child in one instance and not others."

Dan Hon's magnificent rant in response goes like this: the engineering mindset that approaches difficult technical challenges with relish and gusto, but approaches difficult social challenges as though they were fundamentally insoluble and off-limits, is a giant, steaming cop-out. If you can build a billion-person, world-scale comms platform, can you credibly throw your hands up and give up on solving the problems it creates?

Now, some problems may indeed be different kinds of hard/insoluble (think of the inverse, in which policymakers say to technologists, "Don't tell me that crypto backdoors are 'too hard,' just do it, Poindexter"). But as Hon points out, there's plenty of advanced work on adjudicating exactly this kind of dispute, which may be costly, but it's certainly not impossible, and so another way to parse Facebook's objection is, "It's too expensive."

Because it *looks* like this: it *looks* like you're entirely happy solving hard engineering problems (designing and building new server architectures! "Saving" the planet by reducing your carbon footprint and designing more energy-efficient systems! Making it super easy for me to take a video of my newborn baby and effortlessly share it with my entire Dunbar's worth of family and friends! (AND I AM NOT BEING SARCASTIC ABOUT THAT ONE, I GENUINELY AM ACKNOWLEDGING THAT IT'S A DIFFICULT PROBLEM AND WORK WENT INTO IT AND SOLVED THE SHARING A BABY PHOTO PROBLEM).

But. These other problems. These soft, social, anthropological problems. These problems that do not fit in boxes. It looks like they are just too hard for you, as an industry, as a community, to deal with. So you ignore them and scratch the engineering itches (which again! Valid itches! But again, NOT ALL ITCHES!)

We - homo sapiens - and the countless civilizations and cultures that we have created over literally *thousands* of years - have been working on some of these problems for a long time. We have figured out some ways to deal with them. They aren't perfect (nothing is! Apart from, maybe, in our pure world of mathematics and set theory, right?) but they're what we've got and we've also figured out ways for those methods to be *malleable* and to change and to accommodate what is fundamentally a continuous system - an analogue, human one - that doesn't really quantize, that is variable and most definitely, completely, not binary.

It is, of course, always easier to do the easier work than it is to do the hard work. It's easier to do the attractive work, to solve the difficult problems that interest you rather than the difficult ones that don't. Or perhaps, even to be tempted to apply one toolset to a different domain. But, the *evidence* is that the difficult problem of bringing Internet access to people (and thus, let's all acknowledge here, being able to capture some cold hard cash value by being part of the access layer and an intermediary) is way more exciting and interesting and worthy of investment than the difficult problem of figuring out what to do when you can't tell the difference between an allowable nude photo and an unallowable nude photo.

s3e27: It's Difficult [Dan Hon/Tiny Letter]

Notable Replies

  1. Yes, facebook should suppress pictures of people screaming in pain. I don't want Abu Graib in my face, or all those sorry civilian casualties in those far-off countries in my news. I prefer images of raising the flag over Iwo Jima, and the "mission complete" stuff from 2004 desert storm, and other true depictions of the USA's military might. And would it kill facebook to airbrush happy faces onto all these funeral mourners and bereaved? All this other stuff is just gross and makes it hard for me to concentrate on my portfolio.

    And yeah as for nudity, it's of course freakin' disgusting all by itself. C'mon, facebook, do your job!

  2. Nelsie says:

    Engineers solve engineering problems because that's how their brains are wired, I guess. Which is fair enough. But when your solution is earning a lot of money, and other, equally-talented engineers' solutions aren't, that kind of implies that you've also got someone in to solve the money-management problems at some point. And if you can get someone whose brain is wired up to solve money problems, why can't you hire someone whose brain is wired-up to solve social-management problems?

  3. Church says:

    Simple. All Facebook has to do is develop an AI that understands context. Maybe they can get some help from the MPAA.

  4. I'm just glad to see a quote on BB pointing out that engineering problems are relatively easy in comparison to the much more complex anthropological ones.

  5. My thought is that yes, an algorithmic approach can be excused for sometimes making the wrong decision, but there always needs to be the recourse of someone noting a wrongly-removed photo. In this case, it would have been easy for a thusly-notified human to quickly assess that this photo is monumentally significant, culturally and historically, and to not just reinstate it but to both add that photo as one the algorithm never removes again, and to flag "similar" photos (same photographer, or same posting account, etc) as "humans should review future flaggings in this category til we feel the algorithm is heuristically steered back on course". From what I can tell, in this case Facebook stonewalled, even upon being notified of the photo's symbolism... and past that, I don't even understand why the photo wasn't already in a big repo of "do not remove these photos because they're incredibly well-known and valuable".

    (NB: I'm familiar with machine-learning/sorting/AI algorithms, have used and written several myself, and this type of thing seems eminently doable to me.)

Continue the discussion bbs.boingboing.net

75 more replies