Die, Bot, Die!

When and How to Say Goodbye to a Bot Gone Wrong

Bad robot (Edward Liu via Flickr)

I love a good internet joke the same as anyone. I really do. It’s why, when Daniel Wheaton came to me and asked for my complete Twitter archive and didn’t really want to tell me why he wanted it, I just gave it to him. He was a student in a class I was teaching about little internet bots, and I knew this would be fun.

And, with a little Markov chain magic, it was fun. Daniel created a ebooks bot version of me. Sometimes random gibberish, sometimes Genuine Future of News Brand Quality Tweets. Not all that different from my real Twitter account, come to think of it.

But, in my archive are every mention of someone I’ve ever interacted with on Twitter. Sometimes, the interactions are kinda fun.

For the record, Tyler Fisher very graciously told my bot that it is welcome.

But, and here’s the nature of pseudo-randomness, it can get a little … weird.


Yeah. I really respect Margo Kaminski and Nabiha Syed, both first-rate privacy/tech/media lawyers, so random I-hope-she-gets-the-joke tweets are a risk. They might take them the wrong way—see the photo but not read the name. So the risk is not to the bot, but to me.

So, that got me thinking: When do you kill a bot? How? Why? Media are making little bots all the time–scrapers, Twitter bots, If This Then That triggers, Haiku poets, sportsball watchers. Most are getting set up as a simple repetitive task on a forgotten computer–I run a Raspberry Pi in my office as my little bot server and it works great. As these little bots proliferate, some will exceed their useful life.

When I offered a class in story bots, I was so focused on making them, that I never thought about killing them. What would a checklist for execution look like? Some things are kind of obvious. Some might not be immediately so. Some only became clear to me after watching @MattWaitebooks evolve over time. Here’s my list of reasons to kill a bot dead:

The Story Has Moved On

A large amount of work that journalists and developers in news do is story driven. At some point, most likely that story is going to end and we’re going to move on. A lot of bots can keep running until you turn them off. But do you when the story ends? If it’s a scraper, is the data worth keeping? Can you go back and get it all later if you turn it off? Is the audience still interested in updated information, or ongoing tweets? Honestly, of all the reasons I’ve come up with to kill a bot, this is the easiest one. If it’s public-facing and no one cares anymore, kill it. If it broke and you had to go fix it, that’s time spent on something no one cares about anymore. Shut it off. Move on.

The Audience Yawns

Sometimes we have a really good idea for a thing and the audience disagrees with us. They just never sign up like we thought. They don’t follow, they don’t engage, they don’t tell their friends about it. This is a little harder in that a bot doesn’t really cost anything, and if you have a small audience that’s really dedicated you might reconsider, but there should be some metrics to go with a project. X number of followers, Y number of subscribers, something like that. Miss the target, don’t see a future of success, bye bye bot.

The Bot Behaves Erratically

This is really for bots that use pseudo-random text generators or retweet bots with simple rules that can be exploited. If your bot randomly harasses people or says things that could be seen as threatening or weird or uncomfortable, kill it. Or, at the very least, beef up your filters. Everyone needs an editor, bots too.

The Bot Hurts the Brand

This goes along with the previous bot-killing reason, but I looked at the number of followers for @MattWaitEbooks and wondered why 100 people would follow it. Most of them are people who get the joke–current and former students, a couple of colleagues in the College of Journalism and Mass Communications at the University of Nebraska-Lincoln, news nerd friends around the country. A smattering of them are bots themselves, which is sorta funny. But one is a very senior administrator at my university. Another is a research lab on campus. I think one knows me well enough to get the joke, but does the other? Do they think the random gibberish is me? How can I tell? If you can’t tell, or you suspect the worst, kill it with fire before your brand managers or senior executives do it.

The Joke Isn’t Funny Anymore

One great thing about a lot of these bots is that they’re just kind of cheeky and fun. There’s a joke, and for a while anyway, it’s funny. Doing something for fun is not the worst way I’ve come up with as a reason to do something on the internet. Sometimes it’s magical. But, suffice it to say, if the joke isn’t funny anymore, time for your bot to die.

So what to do about my bot alter ego?

I’m sitting two feet from it’s demise right now. I could kill it surgically, by removing a single line from a cron tab. I could use the nuclear option and simply unplug the RasPi that’s been running continuously and without interruption for nearly 120 days as I write this. And just like that, @MattWaitEbooks dies.

And yet, I can’t bring myself to do it. There’s no story here, so it can’t move on. The audience, at least occasionally, finds it funny. Does it behave erratically? Yep. Is it hurting the brand? It just might be.

Given that, I thought I’d teach the bot a lesson. I went into Daniel’s code and added some regular expressions. Now, no more @ mentions. No more harassing people. No more awkward tweets aimed at people I respect. It’s just my words now. Or some sorta-random amalgamation of them.

Because every once and a while, the joke is still funny.

And disturbingly self aware.



  • Matt Waite

    Matt Waite is a professor of practice in the College of Journalism and Mass Communications at the University of Nebraska-Lincoln and founder of the Drone Journalism Lab. Since he joined the faculty in 2011, he and his students have used drones to report news in six countries on three continents. From 2007-2011, he was a programmer/journalist for the St. Petersburg Times where he developed the Pulitzer Prize-winning website PolitiFact.


Current page