Building a basic chatbot with WNYC’s John Keefe

Plus, a brief look at some of the potential applications for small news publishers

Joe Amditis
Dec 7, 2016 · 4 min read

John Keefe, Senior Editor for Data News at WNYC, took to the stage today on day two of the #FutureToday Summit at the 92nd Street Y to give a crash course in basic chatbots. Keefe’s build-a-bot workshop was a quick, useful introduction to the world of chatbots, which are quickly becoming one of the most sought-after pieces of tech in the journalism industry.

John Keefe kicks off his Build-A-Bot workshop at the Future Today Summit at the 92nd Street Y in NYC.

The bots Keefe taught us to build are simple, call-and-response bots that draw on two columns in a Google Sheets spreadsheet in order to receive and respond to user queries.

The rest of the back-end infrastructure that informs and guides the behavior of the KeefeBots is still somewhat of a mystery to me, but Keefe did give us a brief introduction to the machine-learning apparatus that operates quietly behind the scenes at — I just have a little more work to do before I fully understand it.

The level of skill required to program a KeefeBot is incredibly low. If you can use a spreadsheet, you can program a basic KeefeBot. This was my first encounter with bot technology. I know there are other tools out there like Motion.AI that will allow you to build a bot of your own, probably with a bit more complexity, but I found the simple, cell-by-cell entry style very easy to pick up and play around with.

There are clear limitations to this type of chatbot. This is especially true if you’re looking to create a bot for the purpose of general user interactions or something of a similarly broad nature.

For one, it would take a very long time to go in, anticipate, and add every possible question or query that your users might want to ask your bot. Luckily, the KeefeBots have at least some basic artificial intelligence at work in the background. You can ask simple questions about the date, such as “What was the date 13 days ago?” and the bot will tell you. Go ahead, give it a shot.

It’s not very comprehensive or adaptive, but the machine-learning at work behind the scenes via means that it is possible to teach an old KeefeBot new tricks.

Another limitation of the KeefeBot and similar bots is the fact that they are mostly unable to provide conditional responses, or responses that are based on previous responses or conditions. So, for example, I can ask a KeefeBot, “How can I contact the news desk?” Assuming the creator of the KeefeBot has that query in the “question” column of their spreadsheet, the KeefeBot will provide the answer, perhaps an email address or phone number.

If I were to respond with the follow-up question, “When are they available?” the KeefeBot would not be able to parse the language of the question and accurately determine who “they” refers to in that question. It would not be able to go back to the previous question, identify the subject as “the news desk,” and then respond to my follow-up question as it relates to the subject of my first question.

Again, there’s an entire back-end at work here, of which I have almost zero knowledge, and there are many other, more complex and adaptive options when it comes to building a bot. But the low level of technical expertise required to build something like a KeefeBot creates an extremely low barrier to entry, of which publishers of all sizes should look to take full advantage.

Potential applications for small publishers

First, these kinds of bots seem like they would be incredibly useful (and neat) at various journalism and media events like town halls, conferences, or trainings. I can imagine the creator filling the bot with useful event information and answers to common event-related questions like “Where’s the bathroom?” and “What’s the Wi-Fi password?” or “What room is the [session name] in?”

You might also use these simple bots to handle some of the more basic customer relations management functions and social media interactions (many simple bots already boast integrations with Facebook Messenger and other chat apps).

Simple informational queries from new or unfamiliar users that would normally require precious time and attention from what little staff members many smaller publications have on hand might eventually be replaced by chatbots.

The key thing here seems to be upkeep. Bots like the KeefeBot can only be as useful and responsive as the people who create them. As long as you pay close attention to the questions your users ask and continue to update the question and response fields in the bot spreadsheet, you should be able to create a novel and valuable experience for users who encounter your new bots.

NJ Mobile News Lab

Experiments in Mobile News Innovation from the Center for…

NJ Mobile News Lab

Experiments in Mobile News Innovation from the Center for Cooperative Media

Joe Amditis

Written by

Joe is the associate director of the Center for Cooperative Media at Montclair State University and the host of the WTF Just Happened Today? podcast.

NJ Mobile News Lab

Experiments in Mobile News Innovation from the Center for Cooperative Media