In early May, Google’s “Duplex” dominated the news (in addition to preparations for a certain UK wedding). Demos of the robo-caller, sounding disarmingly alive while scheduling haircut appointments and dinner reservations, had us humans talking.
Its authenticity was bolstered by a smattering of ums and uhs. All that was missing were a few “likes”, and an upward inflection at the end of sentences — as in, “I want to um, like, make a reservation for Saturday night?” The initial creep out factor for many was high.
Duplex raises a number of ethical questions, because it’s so lifelike. Among them: Should a bot be obligated to identify itself as such, when there’s a human on the other end of the line – especially if the call is being recorded?
A pre-recorded telemarketing message doesn’t necessarily start out with, “This is a recording…”, but then again it’s pretty easy to spot after a few seconds. If the first few sentences don’t tip us off, we realize we’ve been duped once we try asking a question.
Then, we usually hang up.
Duplex is interactive. It can pivot, for example, if the flesh-and-blood hostess replies that there are no tables available at 6:30. It will then inquire about a table for two at 7 p.m.
I’m not a product visionary. I don’t see a two-inch wire and think, “If I bend this a few times, I could use it to clip loose papers together and call it a ‘paper clip’.” But it’s easy to see the utility of a robo-caller that gets mundane tasks done, while sounding authentically human.
So, what if the task isn’t mundane, at least to some of the humans involved?
As I watched the Duplex demos, I flashed back to a job interview from a few years ago. It was a first-round screener — an exercise that doesn’t usually require much prep, but is a way for company and candidate to size each other up. Mostly it’s a forum to assess whether the candidate is fundamentally under/over qualified.
It’s also an opportunity for a candidate to get a first read on cultural fit. Occasionally, companies – especially the “hot” ones – forget this.
The aforementioned interview involved a recorded phone screener. I had never heard of such a thing. I went through a 20-minute online “training” to learn how the tool worked, and how the interview would go.
I was issued an applicant ID, and could call at any time within a date range. A series of pre-recorded questions would be put to me, and when I was ready to answer each one I was to press a button and begin speaking at the beep. When I stopped speaking after each answer, the tool would take my silence as a sign I was finished, and move on to the next question.
It wasn’t like a voicemail message that could be re-recorded if I wasn’t happy. I had one shot.
I am not sure why I went through with the interview. I have no clue who, if anyone, even listened to my recorded responses. Because I spent that interview essentially talking to myself, I never got a sense of how it had gone. What’s more, I learned absolutely nothing about the role. There was no way to ask questions.
Here’s what I did learn about the company and its culture:
The company was more interested in checking boxes – “Did she confirm she has 5+ years of experience doing XYZ?” – than in interacting to assess personality and cultural fit.
The corporate culture was not for me. Whether I was, as I suspected, overqualified for the role, the company had demonstrated that employees were just numbers. Widgets.
The company misjudged the balance of power between us. I was interviewing them too, and they bombed.
I can easily imagine Duplex serving as a more sophisticated version of the unfortunate interview tool I experienced. It could certainly help move widgets along the assembly line faster, and at lower cost. But, does it stand to reason that a bot will be as effective at screening out the “bad” fits?
And what if the strongest candidates remove themselves from consideration, because they believe they deserve a flesh-and-blood partner for the interview dance?
I hope companies that ultimately gain access to Duplex technology use it appropriately. Booking a table for Saturday night dinner is a mundane task. The two-sided evaluation of a human being’s fit for a role is not.
This is not to say that a non-bot interviewer cures all ills. I also recall a phone screener where the interviewer – who sounded like a summer intern – read verbatim from a script.
But that’s a story for another post…