About two weeks ago, my Twitter account was suspended. The message arrived while I was in the middle of messaging one of my cronies about Facilitated Communication. When I clicked “send”, Twitter told me, “You are not permitted to perform this action. Learn more.” When I clicked on “Learn more”, Twitter told me, “Your account is suspended. Learn more.” When I clicked again, Twitter told me that the suspension was for “making violent threats.” “You may not threaten violence against an individual or a group of people,” Twitter explained.
My account is public, visible to my students and colleagues. Even if I were inclined to do so, it would be a really bad idea for me to make threats—let alone violent ones. I wasn’t, and hadn’t—as many of the supporters who subsequently came out in my defense kept telling @Twitter.
So what had triggered this suspension?
It’s Twitter’s policy not to tell you who reported you or what the offending tweet was. For a couple of days I was certain that someone who doesn’t like my content was somehow responsible—perhaps, for example, a defender of Facilitated Communication. But, upon reflecting on my recent tweets, I found another possible culprit—a tweet I had posted some five days earlier. It went something like this:
Today’s text exchange with my #moderatelyAutistic, #HFA son:
Me: Did you finish your work?
Him: Almost, more messages = I smash your head into the wall.
Me: We need to work on your behavior.
Him: I wasn’t actually going to smash your head into the wall.
This was part of a series I’d been posting with the hashtags #moderatelyAutistic, #HFA. My goal was to showcase the co-existence of HFA (high functioning autism) with “moderate autism” (meaning more than mildly autistic)—a co-existence that some people think isn’t possible and doesn’t exist. I also wanted to contrast my son’s texts with the gentle bromides that people extract from other autistic children via Facilitated Communication (“I believe that communication is a basic right”, etc.). And, yes, I was also going for humor.
But could @Twitter take a joke?
I went back over my timeline and sought out the “smash” tweet. And, lo and behold, it was gone. Never before had I experienced this: to the best of my knowledge, no other tweet of mine has suddenly disappeared.
And so, the dust settled: Twitter must be relying on its artificial intelligence system to identify violent threats and suspend accounts. The AI had found the phrase “smash your head into the wall” but lacked the ability to parse the broader context. It couldn’t distinguish a threat made by me from a threat made to me and then retracted.
It’s actually no surprise that Twitter AI is yet another one of the machine-learning trained systems that we see everywhere from Siri (see earlier) to automatic essay grading (more on that later). These are systems that eschew the complexities of natural language processing and simply look for statistical patterns involving key words and phrases. No matter if these key words and phrases (say, “smash” and “head”) are embedded in a dialogue or a clause within a negated sentence.
But what was surprising, at least to me, was that Twitter would leave its decisions about account suspensions entirely up to its AI system, and that no human eyes would ever be involved.
Well, not quite. It turns out that Twitter’s suspension does involve human eyes, but not in the benign ways that you’d expect. I’ll have more to say about that in my next post.