Troll the bot, troll means love.
When I was in grad school for Digital Storytelling, oh so many years ago, there was a professor whose motto was red means love. It was his belief that marks on a paper was a good thing. It meant you could refine your idea. But this is a hard concept for a student to get used to. We all want to be right on our first go, even our second go.
However, that’s not how people get better at things. Especially not writing. So having a draft that was just littered with red pen marks was a sign that you had stuff to improve upon. It was a good thing.
This thought ran through my mind as I added testers to my Messenger weather bot. I would be able to get some nice feedback, see my blindspots, go back and build a cleaner interaction, a better bot.
However, the testing my friends put the bot through went beyond the expected interactions. It’s not that my testers just expected the bot to be smarter than it was. There were levels of trolling that I wasn’t prepared for. Oh, how naive I am…
Let’s start with the basic stuff.
First off, let’s talk about the repeating of a phrase. I think the ‘as the kids on the streets say these days’ is dumb but in a fun way. However, the conversation and whatever personality the bot had grows stale quickly. I don’t imagine the average user to do multiple inquiries, but then again there’s this feeling of testing bots as much as possible from the user side of things. People want to know how much a bot can do.
Also, my conversation design differs from Hi Poncho. While that bot relies on a mix of text and CTAs (call to actions), I’m relying just on text. This makes it a little harder to steer the conversation, in fact, it makes it impossible. I’m not able to guess what the user will be doing or steer them into a path. Their options aren’t limited to simply yes or no.
This can be seen as bad design especially when thinking of users as customers. However, I don’t find CTAs fun. I’m of the mindset that a bot should be about the conversation via texting. That’s what is fun about it and what will endure people to bots, not more clicking and filling out forms which is how Hi Poncho feels to me.
So feel that I should be jumping into open conversation right off the bat, even if it’s more painful as a designer.
RECOMMENDED FURTHER READING:
A call to better bot building from by Alex Bunardzic.
Back to ripping my conversation design apart, the second issue with the above is “When will it rain?” being asked by the user. I think every one of my seven testers did something like this. Either they expected the bot to know the location right off the bat or expected the bot to remember the location.
I don’t believe that I can get the location via Facebook messenger, and I didn’t even think about storing the location for future reference though I know it’s possible thanks to a current bug in my conversation design…
At the end of the day, this is the kind of information I was expecting. Things that seemed like a good idea in theory but weird in practice. Or just overlooked stuff, no matter how stupid I feel about overlooking it.
What I didn’t expect was stuff like this:
This was just weird, and I’m not 100% sure how to build for this. The easiest and most human thing, I think, would be to simply ignore it.
Don’t feed the trolls.
Probably because I’m not an angry person most of the time I didn’t plan on a lot of that. Not that the above user was really angry. The tester is just someone that really likes the word fuck and throws it in a sentence aggressively. But we all know that just reading can cause weird interpretations which is something that I believe we’ll need to think about when creating bots.
Bots just don’t know how to feel that feel, you know? So it comes down to reading it cold and guessing for the bot during the design process. Or maybe there’s another way?
I’ll grab “What the fuck!?” and add it to my lists of intents for help.
Funny I thought of “this sucks” but not “what the fuck”. Hindsight.
Now these interactions seem like they shouldn’t be ignored. They aren’t abstract or weird. There is substance behind the words. It’s just a matter of getting it out and processing it for the bot.
Maybe I’m being a little optimistic on this, but I think there should some thought put into what kind of trolling should be addressed. A blanket don’t feed the trolls policy doesn’t make sense. People are going to troll whether intentionally or not. We need to think of them as opportunities to convert them into better bot users. Not just for my bot or your bot, but we need better bot users going forward.
There will always be trolly outliers, but creating better bot users will ultimately allow us to create better bots.
As I finish this article I’m excited to clean up my current conversation design, and let go through another round of testing. I’m excited to see them troll this so I can gleam some idea as to what this dum weather bot might be capable of and what my future bots should be addressing.
I just have to remember troll is love.
