That first iteration of the bot, where we asked for an email address, was intended to minimize missed connections. With Driftbot, on the other hand, our goal was to prevent conversations from coming to an end. So if a user goes off-script and Poncho gets stumped, it doesn’t necessarily ruin the experience when that conversation comes to an end. There’s a defined set of things users can expect the chatbot to be able to tell them. That’s because Poncho’s purpose is to programmatically respond to specific questions. Ultimately, where a chatbot falls along the spectrum should depend on its particular purpose.įor example, it makes sense that the weather forecasting chatbot Poncho for Facebook Messenger is all the way over on the no-human/chatbot-only end of the spectrum. No position on the chatbot spectrum is inherently better than another.
How we decided what type of chatbot to build Move over even further and you’ll find Driftbot’s predecessor: a super-minimal chatbot that would ask people a single question before connecting them with a human (“Hey, just in case we get disconnected, what’s your email address?”).įinally, if you strip away that super-minimal layer of chatbot interaction you’re left with no chatbot interaction at all - You’re chatting human-to-human. Then it connects you to that human as quickly as possible. It asks you questions not so it can attempt to resolve your issue on its own, but so it can figure out who the best human is for you to talk to. While the chatbots on the right side of the spectrum were focused on replacing or minimizing human-to-human interaction, the chatbots on the left side are geared toward enhancing it.ĭriftbot, for example, acts sort of like an intelligent switchboard for live chat.
#Drift chatbot has scrollbars upon opening software#
But eventually, if you’re persistent (or you repeatedly mash the “0” key with your thumb), you can reach a human.Ĭross the halfway point and you get to where our chatbot software fits into the picture. It’s like when you call a customer support line and get that pre-recorded voice that asks, “What is the nature of your call, this, this, or this?” and then you respond and get another set of questions, and another, and another. Move over a bit and you have chatbots that try to extract as much information from you as possible on their own, but can ultimately fall back on a human if it’s absolutely necessary. It’s a strictly human-to-chatbot conversation. And all along that spectrum you’ll find chatbots that can engage with you in a variety of ways and to varying degrees.ĭrift product manager Matt Bilotti explained it to me like this:Īll the way on the right you have totally programmatic chatbots, where you give them specific prompts, or ask them specific questions, and they give you specific responses. But in reality, today’s chatbots fall along a spectrum.Īt one end of the spectrum are chatbots that are designed to replace human-to-human communication.Īt the other end of the spectrum are chatbots that are designed to facilitate it. to beat the Turing test - to become indistinguishable from humans during conversation). We often think about chatbots as a single category that has universally shared goals (e.g. Instead, it describes what the chatbot landscape looks like right now. No, that isn’t the opening line of a dystopian science fiction novel (yet).
0 kommentar(er)
