My unwitting conversation with an AI
Kismet, a robot with rudimentary social skills (Photo credit: Wikipedia)
Today I got a phone call. Here's what happened, as best I can remember.
I pick up.
"Hello," I say.
"Hi Mike," comes a voice on the other end. It's a woman with a very pleasant phone voice. If you rated the last hundred strangers I've talked to on the phone, she'd come out way on top.
"You're as hard to get hold of as it is to get my kids to clean up their room," she continues.
I chuckle appropriately.
"I'm calling on behalf of the breast cancer association," she goes on and launches into a spiel. Something tips me off. Maybe she's a bit too smooth.
So I interrupt. "Wait a minute. Are you a real person? Or are you some kind of a robot."
I expect she's going to keep going. But she stops, in mid-spiel.
"Well," she says smoothly, "to ensure quality control, I am using prerecorded segments of speech. But there is a real live person on the call."
My brain freezes. Somewhere in my unconscious, I realize that I just asked a robot if it was a robot, and the robot understood my question and played the pre-recorded speech that explained and excused itself for being a robot. Or maybe a real live person was listening and switched from one pre-recorded segment to another. I don't know, but I'm inclined to believe that it was AI all the way.
"Sorry," I said, mind-numbed. "We don't respond to telephone solicitations of any kind."
"That's OK," the voice continued smoothly. "Just tell three of your friends something something." I hang up in a mild panic state.
Thirty seconds later, I wish I hadn't. I would have liked to explore whatever was on the other end of the phone. Would it have answered other questions? How would it have managed the transition to the "real person" if it had to? If I got to a real person, would it be the one with the awesome phone voice, monitoring twenty lines, and jumping in only when necessary? Or would it be someone who called herself Jane but had a think Indian accent, like so many Janes do.
This is the shape of things to come. Here's a computer system handling a one-on-one sales conversation. The repertoire for this conversation is pretty limited, and I outed it pretty quickly, but I'll bet that it takes a lot of people to the end of a conversation without them suspecting it's a computer, not a person.
And there are lots of jobs that don't require a lot more skill than that. Customer support, for example. It's all flow charted. A computer can follow a flow chart as well as a low-wage operator. Better, actually.
Every interaction provides information for improvement. Alternate speech segments can be A/B tested. The algorithm can match the voice of the "operator" to the profile of the client, based on information that they've scraped off the net. I got someone who sounded like a well-educated white woman. If I was black would I get a well educated black woman? If I was a woman, would I get a man? Or is a woman better? Which one did I respond to? Give more like that. Where does the conversation break down? Fix that segment. What question did someone ask for which there wasn't a pre-built sound bite? Build that bite.
The analysis can be done automatically. The system spits out the metrics. Coming up with a solution takes people, right now. But each solution scales across all campaigns and all domains. The AI gets better and better. The jobs that people do become less.
When the system can't deal with a situation, it temporizes while, in real time, a human being gets all the contextual information needed to fill a gap. The person fills the gap, speaking in their own voice. The result is parsed and synthesized in the "voice of the ideal operator for this customer." The repertoire grows larger.
And it's early days.
It's going to better.
On the other hand, if you're hoping that your interpersonal skills will keep you from being replaced by a computer, it's going to get a lot worse.