Warning HashflagsPublished 7 years, 10 months past
Over the weekend, I published “Time and Emotion” on The Pastry Box, in which I pondered the way we’re creating the data that the data-miners of the future will use to (literally) thoughtlessly construct emotional minefields — if we don’t work to turn away from that outcome.
The way I introduced the topic was by noting the calendar coincidence of the Star Wars-themed tradition of “May the Fourth be with you” and the anniversary of the Kent State shootings in 1970, and how I observe the latter while most of the internet celebrates the former: by tweeting some song lyrics with a relevant hashtag,
#maythe4th. I did as I said I would…and Twitter blindly added a layer of commentary with a very simple little content filter. On
twitter.com and in the official Twitter app, a little Stormtrooper helmet was inserted after the hashtag
So let’s review: I tweeted in remembrance of a group of National Guardsmen firing into a crowd of college students, wounding nine and killing four. After the date hashtag, there appeared a Stormtrooper icon. To someone who came into it cold, that could easily read as a particularly tasteless joke-slash-attack, equating the Guardsmen with a Nazi paramilitary group by way of Star Wars reference. While some might agree with that characterization, it was not my intent. The meaning of what I wrote was altered by an unthinking algorithm. It imposed on me a rhetorical position that I do not hold.
In a like vein, Thijs Reijgersberg pointed out that May 4th is Remembrance of the Dead Day in the Netherlands, an occasion to honor those who died in conflict since the outbreak of World War II. He did so on Twitter, using the same hashtag I had, and again got a Stormtrooper helmet inserted into his tweet. A Stormtrooper as part of a tweet about the Dutch remembrance of their war dead from World War II on. That’s…troublesome.
Michael Wiik, following on our observations, took it all one step further by tweeting a number of historical events collected from Wikipedia. I know several of my British chums would heartily agree with the 1979 tweet’s added layer of commentary, but there are others who might well feel enraged and disgusted. That could include someone who tweets about the election in celebration, the way people sometimes do about their heroes.
But what about appending a Stormtrooper helment to an observance of the liberation of the Neuengamme concentration camp in 1945? For that matter, suppose someone tweets May-4th birthday congratulations to a Holocaust survivor, or the child of a Holocaust survivor? The descendant of a Holocaust victim?
You might think that this is all a bit much, because all you have to do is avoid using the hashtag, or Twitter altogether. Those are solutions, but they’re not very useful solutions. They require humans to alter their behavior to accommodate code, rather than expecting code to accommodate humans; and furthermore, they require that humans have foreknowledge. I didn’t know the hashtag would get an emoji before I did it. And, because it only shows up in some methods of accessing Twitter, there’s every chance I wouldn’t have known it was there, had I not used
twitter.com to post. Can you imagine if someone sent a tweet out, found themselves attacked for tweeting in poor taste, and couldn’t even see what was upsetting people?
And, as it happens, even
#may4th wasn’t safe from being hashflagged, as Twitter calls it, though that was different: it got a yellow droid’s top dome (I assume BB-8) rather than a Stormtrooper helmet. The droid doesn’t have nearly the same historical baggage (yet), but it still risks making a user look like they’re being mocking or silly in a situation where the opposite was intended. If they tagged a remembrance of the 2007 destruction of Greensburg, Kansas with
#may4th, for example.
For me, it was a deeply surreal way to make the one of the points I’d been talking about in my Pastry Box article. We’re designing processes that alter people’s intended meaning by altering content and thus adding unwanted context, code that throws pieces of data together without awareness of meaning and intent, code that will synthesize emotional environments effectively at random. Emergent patterns are happening entirely outside our control, and we’re not even thinking about the ways we thoughtlessly cede that control. We’re like toddlers throwing tinted drinking glasses on the floor to see the pretty sparkles, not thinking about how the resulting beauty might slice someone’s foot open.
We don’t need to stop writing code. We do need to start thinking.
But…. …realistically, pragmatically, the solution is rather simple. On twitter.com, they should provide a preview of how the tweet will look. Obviously, that has a whole host of other benefits as well.
So the problem is not the behavior, it’s the transparency. Make the results transparent, and the behavior stops being an invisible faux pas and intention can match results.
It’s ironic that the attempt to make algorithms seem more human is tripping us up. I thought about this the other day after asking Siri (on my iPhone) to set a timer for seven minutes. Siri responded with something to the effect of, “OK, but remember, a watched iPhone never boils.” The comment didn’t bother me at the time because I was merely trying to bake the perfect waffle. But had I been, for example, administering measured and critically-timed doses of epinephrine to someone suffering from anaphylaxis, the cheekiness would have been less well-received. I wonder if the development of some sort of “emotion coprocessor” could help? Sample several of the user’s biological signals, and if it is determined that she is stressed, anxious, or angry, deliver a “strictly business” response. If on the other hand the user seems to be happy and well-situated, deliver a more casual or even humorous response tuned to that mood. As it is, it seems the modus operandi of today’s technology crafters is to assume that the users of their products are in a normal, non-crisis mode, and hope that this applies to a majority of the population (“user base”). And it’s not working very well, is it?
Eric Meyer on Twitter’s Manipulation of #MayThe4th | b19y
[…] via Eric’s Archived Thoughts: Warning Hashflags. […]
This is “uncanny valley” stuff.
Technology used to be completely unsophisticated and unable to even approximate human behaviour.
Then we got a few things working and it was cute.
Now we passed the cute threshold and the increase in technology is resulting in lots of conspicuous anti-human behaviour.
Ron, I agree entirely! I had the same thought after reading Trace’s comment and tweeted the thought along with an article from last December covering that angle written by Nicholas Bowman. I think there’s a lot to this idea of an emotional uncanny valley, where we’re trying to emulate human behavior in code, but don’t have the tools/power/something to make it human enough. So the attempt comes off as really creepy or wrong, at times.
I am not seeing Twitter doing this now, I wonder if they got the message, or they just thought it would be cute for the day.
I agree with your premise. A little research would go a long way.
Jack: The list at http://hashfla.gs/star-wars/ shows they’re no longer flagging #MayThe4th, just things directly related to the new movie now. In fact, by the time I saw Eric’s post the other posts were no longer flagged, so I think that set was only active on that day.
The thing is, it’s a nice feature most of the time. During the last Soccer Extortion Cup they appended the flags of each country to each country’s three-letter abbreviation. It added a nice touch. For something like a date, though, just the presence of the hashtag isn’t enough, as all the examples pointed out.
Perhaps adding a keyword trigger to the system would help. #MayThe4th wouldn’t work by itself, but in conjunction with “Star Wars” or “wookie” it would trigger the emoji. A basic one like the flags would just match “*” and be done with it. It wouldn’t be perfect — you couldn’t match “shot first” for Kent State related reasons — but false negatives are better than false positives in most cases like this.
#WCNEO – WordCamp Northeast Ohio – Morsie Code
[…] to “stress-test” their sites include the Google April Fools Day Mic Drop Feature, the #MayTheFourth hashflag on Twitter, and the “Cheatin’ uh?” error message in WordPress […]