Sometime around Christmas, when we look back at the year 2016 in tech, one of the biggest trends will be chatbots.
Suddenly, all of the Silicon Valley engineers who live in detached houses and drive to work realized that most of the planet’s population doesn’t have as much private space and would rather text, not talk to, their A.I. assistants. (Seriously, if you think saying “OK Google” sounds cool in a crowded subway car, you’re probably the one holding the progress back and still wearing Google Glass.)
As much as texting while hauling a giant rig is an equally giant no-no, my team and I have carefully analyzed the way our beloved truck-driving customers use our app Trucker Path Pro. The app is designed to assist truckers with their everyday searches, like service stops or cheap fuel. We found that offering them a chatbot makes sense for when they’re parked, resting, and planning their trip.
The bot helps them find essential facilities like parking lots and weigh stations quickly and conveniently. Here are a few observations I made while playing with other bots and developing our own.
State of the bots
Some businesses rush to take advantage of the hype and end up offering simplistic, underdeveloped tech. Their customers may not be used to this type of man-machine communication, yet they get a fun-for-a-minute experience with no immediately clear benefits. Forgettable interactions don’t help to engage and retain customers.
Others aim for a very specific one-way type of communications, creating things that are not quite chatbots but more like a glorified email subscription repackaged for 2016. Don’t get me wrong, some of them have pretty high retention rates, but their engagement power remains to be seen. Some may feel overwhelming to users because they spit out information constantly.
Finally, I’ve seen a few chatbots that seem incredibly intelligent for A.I. — until the media looked into them and discovered there was, ahem, a human backend somewhere in Bangalore. This is all the television show Silicon Valley (season three, to be exact) in reverse, but the guys are probably just hoping to fake it until the whole industry makes it.
Notifiers are the most primitive type. Their conversational logic is a small part of their overall code base. As a consequence, they are very common and easy to build, and there are thousands of them out there. They typically just broadcast messages and don’t react to user input. Several bot constructors, like Facebook’s wit.ai, are mostly used to build this kind of bot.
What actually works
The next step are bots that are able to consume human input and respond with a meaningful message (let’s call them Reactors). They don’t know much about the context of the specific conversation and have no memory of any prior conversation. More complicated systems are what I call Responders. These guys have the memory of a 5-year-old and are able to keep track of the location and the context. Talking to one, you notice the flaws of the current state of technology because you sometimes get random messages out of context.
The most advanced category of bots to date is what I call the Conversationalist (and which some people just refer to as A.I.). Not only are they able to use the location, user profile, context, and previous conversations’ data to deliver the message, but they also learn and improve based on patterns using technologies like machine learning and neural networks.
To offer our customers a true conversationalist is our current bot-related goal. To demonstrate any tech’s true potential to the masses, you may sometimes need one large key player (what Pokémon Go is for AR) but also a thousand smaller ones. Chatbots look like they may be the latter, and it’s our responsibility to provide ever-increasing value to our customers and warm them up to the concept.