What AI companies can learn from airline autopilot
Mar 28, 2017
Tags: ai , Casting.AI , machine learning
Image Credit: Shutterstock.com / Melpomene
Only a few weeks ago, I was living in my blissfully ignorant AI -will-eat-the-world bubble. Then I realized things were not always so blissful. I had the good fortune of being accepted to Boomtown , an accelerator in beautiful Boulder, Colorado. I was sure everyone there would be as enamored with our AI vision as we were. Well, turns out, not so much. Showing how our system shaves hours off their daily routine would elicit the kind of relief generally reserved only for people who realize that the sheet of paper in their windshield wiper is just a cheap advertisement and not a ticket. Everyone loved the features, but as we started describing how our clever AI was making it all possible, almost on cue their excitement dissipated. It appears walking people through the process on the factory floor of the proverbial sausage factory was ruining their appetite for the finished product. Keep the AI behind the scenes
At first, I was prepared to chalk these findings up to the tech-averse nature of our industry, where I have had real clients lament how they miss making phone calls over sending emails. However, another AI startup in our cohort was receiving similar feedback, and it’s using AI to automate hiring tech talent. If any industry should be comfortable with tech, it is the people building tech. I was ready to write the whole world off as a bunch of Luddites, but Ian Barnor of RecruitSumo managed to draw more insight from this undesirable feedback. He compared it to airline autopilot. When he was young he dreamed of being an airline pilot, but a friend who is an airline pilot recently told him that he was lucky his childhood dream went unrealized. He told Ian that he was just in the cockpit to make people feel comfortable and that AI did most the flying now. I fly frequently enough to be comfortable ignoring the safety tutorial. I am fully on board with all things AI. I also knew that most of the flying is being performed by automated systems and that statistically speaking the pilot is more likely to cause an accident than solve one. Even still, I would feel nervous boarding a plane where the cockpit had been converted to fold-down business-class seats. The inner workings of AI are at best confusing and at worst deeply unnerving to end users. Most passengers don’t even realize that most of their flight is being executed by autopilot, and there’s a good reason for that. Explaining how the sensors work in concert to measure the yaw and altitude is enough to make the safety video look like actual entertainment. It also invites people to wonder what would happen if these things went wrong. Which isn’t a concern when it comes to human pilots, although it should be. At the end of the day, people are hardwired to trust other people more than technology, even when it isn’t warranted. The confident nod the pilot gives as you board the plane does more to build your trust than hundred of safety reports detailing the intricacies of the extremely sophisticated autopilot systems that have saved literally thousands of lives. Promote the human touch
However unfounded the emotional response to AI may seem, it should certainly be a factor in how you launch any AI startup. Your pitch deck to investors could be laden with language that is scarcely heard outside of university computer science departments, but it is a mistake to have the same language on your public-facing website. This is clearly a lesson that many of the tech giants have already learned, since most people don’t even realize that many of their favorite features are their favorites exactly because they are based on AI. Millions of Netflix viewers depend on “what to watch next” recommendations for finding their next binge-watching pleasure. If you start explaining to people just how it makes such good recommendations, the most interesting thing to watch is their discomfort growing. Netflix is storing your every move, and it remembers everything. It remembers how you quit watching the insightful documentary eight minutes in, instead choosing to binge-watch that kids’ cartoon. It knows that you watched that super dumb movie … twice. The inner workings of AI are at best confusing and at worst deeply unnerving to end users. Despite this fact, most startups are going to great lengths to put their AI components front and center of their customer messaging campaigns. Ironically, most small-scale AI features actually involve a lot of human touch points behind the scenes to make it work. Knowing this might help ameliorate some of the concerns of end users, who are not comfortable putting all their faith into faceless AI agents. However, many well-intentioned founders are instead doing everything in their power to bury the humans involved in their AI magic. Pride about flawless tech may be the downfall of many of the early entrants into the AI revolution. Perhaps eventually we might reach a tipping point where the general public can trust AI systems as much as they trust people, but we certainly are not there now. In the meantime, we should all take a play out of the airlines’ playbook. During the mobile revolution, many startups would “mechanically turk” their process when launching an app, which meant having a real person do something behind the scenes to make the tech look more appealing to users. With the AI revolution, we may need to flip these old best practices on their head, shifting that real person to the forefront of the user interaction so that the AI tech on the backend can look more appealing. Remember you can have the most well-trained, technically sophisticated, and highly effective AI system in the world, and it will still not be half as reassuring to users as a confident nod from an official-looking human. Aiden Livingston is the founder of Casting.AI , the first chatbot talent agent.