When people tumble into the Uncanny Valley
The other day I got a call from my insurance company. As my wife’s pregnancy progresses (obligatory knock on wood) we find ourselves doing things we never thought we’ll do, like taking out life and disability insurances. You know, adult people things. I navigated through the automated telephone system and then a very nice lady picked up the phone and started interviewing me. As she fired her questions at me a nagging thought kept bothering me:
Am I talking to a human or a bot?
See, her speech was almost mechanic. Her tone was steady, not going up or down at all, the rhythm very static. For a good part of our conversation, I couldn’t say whether she was a real person or an automated system. Not for sure, at least. And it drove me crazy.
There’s a fun game that was making the rounds a few months ago called Detective. It’s a chat based game where you need to figure out if you are chatting with a bot or a person pretending to be a bot. It was fun. I wasn’t very good at it.
Talking to the Insurance Lady felt kind of like the same. I asked her to repeat one of her questions, and did so in a way that I hoped would be hard for a bot to understand. She repeated the sentence without missing a beat, and did so in the same exact tone, intonation, and speed as she said it the first time. I wasn’t getting anywhere.
Not so long ago, the mere idea that the person you are communicating with might not be a person, but a collection of algorithms built to respond in a way that resembles a human, was pure science fiction.
These days, though, it’s getting harder and harder to tell apart the bots from the people. Bots, or rather the people who program them, have gotten better and better at simulating the way people talk. So good, in fact, that one can argue they have passed the bottom of the valley and are steadily climbing up the slope out of the valley and into acceptance.
There are a lot of things helping these bots ascend the slope: better algorithms, more computing power, advances in NLP, etc. And there’s one more factor: the robotization of us, humans, especially at work. People who work in jobs like the Insurance Lady’s have to do the same thing over and over. They’re given scripts to follow, instructions on pronunciations and tone of voice. They read the same sentences time after time off their computer screens and say those sentences again and again. No wonder it becomes mechanical. No wonder they sound like robots.
Just a couple of days ago I was standing in line for the cashier at Staples. There were 3 cashiers and a person who was their supervisor. The line was about 5 people long in average, and moving frustratingly slow. Every 30 seconds or so, one of the cashiers would say loudly “Sorry for the long wait.” The phrase was always the same, in the same intonation, and if they didn’t the supervisor would say it himself, and/or shoot them a reprimanding look. It made the wait that much worse.
But at least since I saw them in person, I knew they were people, and not bots.
In the last couple of months I had a few interactions in voice and writing where I had a hard time telling if I were communicating with a person or not.
There was the “assistant” to the person who was trying to set up a meeting. I had my suspicions from the start, but an email asking me why I didn’t reply earlier which arrived at my inbox at 4:45am convinced me it wasn’t a real person I was communicating with. Weirdly enough, I still found myself being polite and writing “thank you” and other pleasantries even when I was convinced I was talking to a bot.
There was the sales person(?) who assisted me in a chat on a site where I was looking for a hardware component. All of the lines he wrote were prewritten for sure, and his help consisted of searching the site for the product I was looking for and sending me a link to the search results.
“Assistant-as-an-App” is a growing niche and the “next big thing” according to some. Some of these apps use only algorithms to communicate with you, some use real people, and some use a combo of the two.
The lines are getting blurry, and any certainty you might think you have about what or who is on the other side of the virtual line is probably exaggerated. Or maybe it’s just me who feels uncomfortable in that situation?
As for Insurance Lady, as we continued the interview I continued to be to unsure whether I’m talking to an actual person. Time went by, the interview was getting closer to the end, and still I had no answer. And then it happened. I don’t remember what she asked and what answer I gave, but I made a breakthrough: I made her laugh.
That set my mind at ease. I was talking to an actual person. Unless, of course, bots have gotten that good. I’m pretty sure they haven’t. Yet.