The Washington PostDemocracy Dies in Darkness

How bots will change the Web, according to a bot we built to answer that question

Chatbots are the new darlings of Silicon Valley — the officially anointed Next Big Thing. They’re on Slack, taking your lunch order. They’re on email, scheduling your meetings. On Tuesday, Facebook CEO Mark Zuckerberg announced that chatbots would soon join Facebook en masse, suggesting that, one day, users may get everything from groceries to Google within (and via) chat.

But let’s say you haven’t been following the bot revolution very closely. You probably have a lot of basic questions. Like: How do computers participate in conversations? And where exactly did this whole bot thing come from? Why is this happening all of the sudden?

To answer those simple questions, we made an equally simple bot. It’s built like a “choose your own adventure” game — follow its prompts, and it will take you on a 101-level tour through the wild world of bots. (Don’t even try to get it to chat about Trump or the Holocaust.)

Intersect Bot, as we’re calling it, has two purposes. The first is obviously to drop some knowledge about chatbots, like when the first one was invented.

The second is to approximate — in all Intersect Bot’s glitchy, stripped-down glory — the basic mechanics of bot conversation. To wit, a chat with a bot is just a series of if/then relationships. The bot is just a program that tries to estimate which “thens” go best with each “if.”

To build our bot, we relied on the Cleverscript tools developed by Existor, a commercial chatbot company based in the U.K. Cleverscript powers hundreds of website “help” bots and other applications, as well as Cleverbot, a chatbot that has come perilously close to passing the Turing Test. The company has developed some pretty sophisticated technology, on par with the latest wave of bots — but they also make it possible for beginners to build more rudimentary bots, which made it perfect for our purposes.

We began by sketching those if/then, or input/output, pairs: what sorts of things did we want the bot to know about? How did we want it to respond to specific user prompts?

The next step was coding those input/output pairs in a way that a computer could read. In Cleverscript’s case, that way is a spreadsheet. The spreadsheet maps inputs to outputs on a 1-to-1 basis, but it also allows some wiggle room: you can also, for instance, set multiple inputs to trigger one output, or set an accuracy threshold that will make sure the output triggers even if the input contains a typo.

After that, the spreadsheet gets uploaded to Cleverscript and published out on its servers, where it’s pushed to us via API. Voila! You are now chatting with a spreadsheet in almost-real time.

How similar is this to a bot like Tay? Wellllll … Tay also operates according to that basic if/then or input/output structure, and the choice to connect a specific input to a specific output is also made according to an unseen model.

But Tay (and other cutting-edge chatbots) have layers and layers of technology on top of that: machine learning and natural language processing and massive, real-time databases, which means both that they recognize a larger array of inputs and that they produce new outputs without a human manually programming them in.

So think of it like this metaphor: Intersect Bot is a typewriter and Tay is a shiny new Macbook. Most of the new chat bots joining Facebook, Slack and other platforms will fall somewhere between the two.

But don’t take my word for it — go ask Intersect Bot! It is desperate to show off its single spreadsheet of hand-coded knowledge.

Tina Matheisen and Tim Wong contributed to this post.

Liked that? Try these!