Algorithms Don’t See

United States Supreme Court Justice Potter Stewart famously quipped “I know it when I see it” in describing his threshold test for obscenity in Jacobellis v. Ohio. The problem with this threshold, of course, is that everyone’s opinion of what is obscene and what is not is different.

For instance, to some, the work of Helmut Newton may be art. To others, it may be smut. But, thankfully, we all have the free will to decide such things for ourselves.

Of course, there are some things I think almost everyone can agree is not OK. Child pornography perhaps being the best example.

And, therein lies the problem with Tumblr’s direction in attempting to ban all porn from their platform. Instead of doing the very hard and very human work of figuring out how to best prevent those things we can almost all agree we know are not OK, they are taking the easy, broad, far reaching, and non-human approach of allowing an algorithm to decide.

Algorithms don’t see. Furthermore, they lack context and nuance. We’ve yet to reach an accurate AI based threshold test for “I know it when I see it.” Because of this, Tumblr’s plan to unleash such a tool to ban porn is not going so well and will likely be the company’s downfall.