Chatbots are in their awkward pre-adolescent stage; what this means for developers

Chatbots are in that awkward pre-adolescent period. We don’t really know what they can do yet, we’re unsure where there fit, and we don’t even know if they have any real usefulness. They have serious potential, and might one day rule the world, but first we have to figure out if they can even speak using a full sentence or engage in an intelligent conversation.

Since this was a big week in automotive technology, especially for Uber and Ford, the nascent quality of chatbots reminded me of the early days in that industry — right around 2003 or 2004 — when cars were first showing signs of autonomous behavior. I was working for a now-defunct magazine called Business 2.0 at the time and wrote about being a passenger in Stanley, the self-driving car. I remember being blown away by the idea of robotic driving even if, in practice, the engineer only let the car drive in a closed-off parking lot. He didn’t want the Volkswagen Touareg to travel too far off the digital leash.

That’s essentially what’s been happening with chatbots. When you use a bot to order a pizza at Pizza Hut, you only have a few options. You can’t ask a pizzabot to sauté your green olives in butter. That might not happen until 2021. What you can do is order a pepperoni pizza, click a few buttons to pick a size, and confirm your address. The pizzabot can’t go “off script” and tell you about the caloric differences between Swiss and Cheddar cheese. It can’t tell you about the history of pizza, and it can’t give you dating advice when you take your girlfriend out this Friday. It can’t tell you that the name “Pizza Hut” was the only phrase that would fit on the sign at the first restaurant. (If they’d had a bigger sign, we’d probably be calling it Pizza Emporium today.)

With cars, it’s a similar challenge, although autonomous driving is much further along. Self-driving cars can’t really deal with bridges yet, they can identify bicyclists on the road but can’t always predict their path, and they don’t know there’s a new four-way stop that the city just installed to solve some traffic congestion problems. That’s one reason Ford recently announced it would not offer fully autonomous cars until 2021. It’s the reason I’m so skeptical that Uber will really be able to offer a fleet of driverless cars anytime soon, even though they will start testing the idea with drivers who have one hand on the steering wheel this month in Pittsburgh. It’s why Tesla doesn’t call their Autopilot mode “autonomous” yet.

All robotic technology relies on more than just natural language process. It is dependent on a vast number of inputs about the world. When you chat with a pizzabot, it’s not that intelligent yet. It can remember that you like the large pepperoni and that you live in San Francisco, but not much else. It’s partly that the bot hasn’t been spoon-fed this information and partly that the A.I. algorithms can’t figure it out independently. There are no pizzabots that can look at my own taste preferences based on ordering history, look at where I live, my family heritage, and the fact that I’m on a diet, and determine that I should stick with a  basic cheese pizza. There are only pizzabots that remember I like green olives.

This reminds me of the pre-adolescent stage in life, that strange period of time when you barely know anything, but you’re given adult-level responsibilities. You don’t know what you don’t know yet. You can be a babysitter, work at the lawn and garden store across town, and even bike to school without supervision. You have the keys to the kingdom, but you barely know how to find any doors. You almost look like an adult, and you can act like one at times, but your future is an open book. That’s essentially what is so appealing about chatbots. They have remarkable (untapped) potential.

That’s also likely why chatbots are so popular and so widely available — in record numbers. They have captured the attention of an admiring public. One chatbot that adds a filter to a selfie attracted 300,000 users within seven days of launching. Chatbots might replace apps for many different reasons, but for now, they are all elbows and knees, bumping into things in their attempts to do something useful and practical. They don’t seem to know what they don’t know yet.

Developers, this is your challenge to start training chatbots to break out of this early awkward stage. Give them more utility, more intelligence, and functionality beyond basic app replacement. Make them so indispensable that we can’t imagine what life was like before we handed over our pizza order to an intelligent assistant. Give us a chance to get comfortable having a chatbot that can monitor our connected homes, turning on the lights and the sprinklers with a simple command. When chatbots become so intelligent and feature-rich that we use them all day at work and in our personal lives, we’ll know this is more than a trend. Chatbots will become an integral, necessary part of our digital existence.