Well, What Do You Say to an Anxious Bot?

This week, over 100 people talked to my MEL essay bot. This may seem like a minuscule amount — at least compared to the hundreds of thousands of viewers accrued by Facebook Live videos of exploding fruit — but the results surprised me. Each of those bot users (or readers) made the decision to pick up their phone, text the number, and then exchange dozens of text messages with a robotic voice. These readers weren’t just viewing the piece, clicking into it, skimming, and then closing the tab. They were intentionally interacting with the bot’s very own argument—which is that bots really aren’t that useful.

Here are some brief observations from those responses.

• Unprompted, many of the users mentioned “loneliness” as a motivation for interacting for so long with the bot. Not enough people active on Slack, or just boredom in the workday? It might be a larger ennui. Of bots, the novelist Elliott Holt writes, “these dolls are also a response to the loneliness that seems systemic in global digital culture.”
• When asked what kind of personality they wanted for their bots, almost all users first chose “useful.” Then they chose “commanding.” “Cute” came last. Personally, I’d rather have cute.

• There were a lot of fond memories of ELIZA, the therapy bot, and SmarterChild, the AIM bot. Maybe text bots are just hitting the 20-year nostalgia cycle.
• Many of the users said that they had never used a bot in a functional way before, but a handful had coded bots of their own. Those respondents were most likely to critique the bot hype. “I don’t think bots will matter 1/100th as much as apps,” one wrote. “All the hyped startups in the ‘bot’ space are really useless or actually are people,” another wrote.

https://twitter.com/robinsloan/status/729689059166031873

• Most users agreed with New York Times writer Jenna Wortham’s argument that bots are mostly a convenient way to disguise human labor.
• In fact, some users felt sympathy for the poor bot. “Working retail enough, I certainly feel like a bot to some customers,” one wrote, in what was probably my favorite response.

• At a certain point, the bot asks you if you want it to aggregate bot news from around the Internet for you. Literally only one person did. The bot interaction might often be more interesting to users than the function, at least in the novel early stage of text bots. The potential exists for more bots based only creating positive interactions rather than servile functions.
• What was most interesting to me about the results was not the answers to the questions I got but rather the fact that the bot prompted people to think about the arguments, formulate a response, and write it out. That’s a big result from just a little text, in a much more controlled fashion than the plague of public Internet comments.

I think the Essay Bot was a success. It modeled a way for bots to create a more aggressive dialogue with their users and used a unique medium to get the argument across. It’s an “essay as persona,” as the novelist Robin Sloan tweeted.

But was it really a bot? Since the Essay Bot was just a flowchart of responses with no real intelligence of its own, I’m tempted to call it a “guided conversation” or a script more than a bot. If I were more technically adept, it could be smarter and more varied, but I’m not sure the experience would be any more compelling. As many users pointed out, what made the Essay Bot compelling was knowing there was a human behind it, monitoring their interactions.

https://twitter.com/SerenaEGolden/status/729805341941071872

The whole project doesn’t make me totally optimistic for bot-writing, however. Current bot services cost money to upkeep; the Essay Bot ran up a tab of over $100, which would be difficult to scale for those without their own technology. Like any article, many readers also abandoned the bot partway through. Only a fraction made it to the very end. If I rewrote it, I would cut it off at around 500 words rather than 1000. “This would be so much faster as a normal essay,” one user complained. True!

If a bot had a plot, then perhaps readers would be motivated to finish it, similar to the text-based iOS adventure game Lifeline. In the end, the issue with literary bots might be the same that plagued the great cul-de-sac of hypertext literature: the projects are more fun to write than to read.

Kyle Chayka is a journalist in New York whose writing has appeared in The Verge, Racked, Curbed, Bloomberg and more.

To talk with MEL’s Essay Bot, text ‘Essay’ to 203–872–5806

Talking to a Bot About Bots