Life with my robot secretary

By Mark Wilson

This article originally appeared on Fast Company and is reprinted with permission.

I'm not important enough to have what the rich and powerful call "people." Most of us aren't. The average personal assistant is paid about $31,000 a year, while the average wage for the US job market is roughly $5,000 less than that. In other words, for a majority of Americans, it would be a sizable step up just to become a personal assistant.

This is why the prospect of having a Clara was so attractive. Clara, a blob of algorithmic code assembled by the Y Combinator-funded startup Clara Labs, is an AI assistant that lives inside your email. Unlike shouting at Siri or Cortana, to get Clara involved, you simply CC it on an email. Clara's specialty? Handling the gruntwork surrounding that one particular conversation we're all forced to have over and over: scheduling.

Clara's also adept at another, slightly unexpected thing: making you feel like a total boss.

Credit: Clara Labs

The Perfect Assistant

I was addicted immediately.

"Clara, can you set us up for coffee this week?" I typed into an email thread. I was unsure. And I felt silly, to be honest, like I was bringing an imaginary friend into an important business meeting. At any moment, I feared, she was going to somehow embarrass me.

"Happy to help!" she replied within minutes, listing a few meeting times that might work. They agreed on one without my typing another word. She sent out a Google Calendar invite, complete with the address of my favorite local coffee shop to take meetings, she thanked them for me, and the deed was done.

And here's the thing: This was about as annoying as Clara ever got. Because a lot of the time, she'd juggle times without me even CCd in the email thread. I'd just see calls appear on my calendar, complete with dial-in instructions or restaurants listed.

It's hard to express Clara's value in quantifiable terms, because the time I've saved was probably only 30 seconds here or there. But Clara has saved me from an omnipresent distraction, like any personal assistant would. And unlike me, Clara didn't once mix up the time conversion of PT to CT to ET, she didn't confuse anyone's names, and she was, more often, more polite than I tend to be.

The Interface You Already Know

After the initial setup, where I listed my preferred location for coffee meetings and lunch meetings on Clara's site, all other settings were handled via email. If I wanted Clara to give me a 15 minute buffer between meetings, I could just write her and tell her. There were no dropdown menus or long lists of settings toggles to navigate; just requests, delivered in natural speech and a familiar interface that requires no learning.

To teach me about some of her deeper features, Clara would just email me. "Tell me about your scheduling preferences!" she wrote once. "You can use me to schedule internal meetings too," she let me know another time. It's the same sort of scheduled email blast that other online services have used for a decade now to let you know more about a product. But in this case, I could feed right back into the mechanism at work. When Clara told me about scheduling, I could respond immediately, "never book a call after 4pm." And it was done.

Prepping for a vacation, I wrote Clara, "I'm on vacation all next week."

"Thanks for letting me know! I'll certainly make note of it on my end. Have an exciting yet restful vacation!" she responded flawlessly. It was faster than setting up my Gmail autoresponder, but far more pleasant, too. Even though I was totally aware that she was fake, I appreciated the sentiment.

Over time, I began speaking more tersely to Clara, trusting her enough to take the training wheels of semantics away, and rely on her mental processing to save me the trouble of properly articulated communication. The other day I sent a couch email with the subject "meeting tomorrow" and the body "can we move it up to the am." Clara spotted the one afternoon call I had scheduled, figured out the contact for their assistant (also an AI in this case!), wrote them, negotiated a new time, and moved it.

Another time, I asked Clara to schedule dinner with a colleague at a specific restaurant when I was "in town" for Fast Company's upcoming Innovation Festival . (I offered no dates or address.) It ended up this colleague used Clara too, so she looked at both our schedules, figured out when the festival was, and just picked a date—offering to change it if we didn't like it. (If only Clara was as full featured as Facebook M, she could have booked us a table, too.)

The Turing Test

Over the course of the month, Clara did the unthinkable for a robot: she always passed for human. I'd CC Clara to set up an appointment, only to ask my contact later if they'd realized she was an AI. Not one person did from her speech. (Some, no doubt skeptical at how I could have a personal assistant, spotted that her domain was from Clara Labs and figured it out. If I'd spend more on a custom-named Clara from a custom domain, though, they wouldn't have seen this.)

As one PR rep put it when I revealed the nature of Clara:

Another person had a very similar reaction:

And when I asked if she suspected that I—little old me—had an assistant?

If there was one complaint about Clara, it was that she would respond to emails too fast, with schedule options too in the ready. But frankly, that wasn't my problem. An overeager employee will do nothing to make me look bad. And besides, Clara got shit done.

The Real Awesome, Creepy Stuff

There were a few genuinely surprising moments, where I had a gut-punch feeling I was living in a sci-fi movie like Her or Ex Machina. For a lunch meeting, Clara had silently scheduled me to a nearby southern restaurant that I liked. I asked the person to share the conversation they'd had with Clara to arrange it. It ended up that he'd asked where I was, she answered "Mark is close to the Andersonville area," and my friend just picked his favorite spot in the area for lunch.

In another instance, Clara Labs founder Maran Nelson CCd her Clara to schedule a meeting, and I CCd my Clara. So then two AIs were digging through our schedules and negotiating a meeting time, without either of us seeing the conversation. I couldn't help but look forward a decade or two, when we all have bots—wrapped in emails and text messages, indistinguishable from real people—negotiating meetings, or prices, or deadlines, or raises, or court appearances, or credit limits, or tax loopholes, or any number of things on our behalf. Will custom, tricked-out AI be the new advantage of the 1%, or will the technology be so scalable and universal that it can benefit the other 99%?

Due to the nature of reporting, much of my day consists of emailing people—often people more important than me, who make more money than me—to figure out when in their schedule they can fit into my schedule.

More often than not, I'm actually dealing with their assistants. It's, to be frank, a long grind on the ego that sets up the entire conversation as if I'm the lesser. And it's a psychology that, no doubt, the people and companies I talk to benefit from. They're the play-it-coy, maybe I can pencil you in, prom queen. I'm constantly the overeager suitor.

And so while it may be petty—heck, it may be downright childish to admit—my favorite part of Clara was feeling like I had people of my own.

The Secret Ingredient Is People

If all this sounds too incredible for technology to handle in 2015, well, your suspicion isn't completely unfounded. Because currently, an unspecified amount of human workers are still writing copy, double checking every email that Clara drafts, and categorizing conversations over and over to teach Clara about etiquette and the nuance of conversation.

In other words, Clara sounds human because she IS human.

"You have to get really really perfect data of that conversation, and you need it really really really well labeled to understand it," Nelson explains. "How do you do this in a very human way? The answer we believe is to introduce humans to this."

When I asked Nelson how much of what I read from Clara was AI, and how much was human—just a ballpark estimate—she wouldn't answer. One industry insider I spoke to has heard that Clara Labs is flat-out brute forcing Clara's intelligence by human hand. Whether or not this is the case, Nelson's thesis is that humans are training the machine, and so no human input is wasted, and that human input is training more and more fringe conversations around and beyond scheduling—like, what does Clara say to someone whose child is sick? What would Clara say if you wanted to offer someone a job?

But in the long term, Nelson doesn't seem to feel the human touch negates the scale or power of the Clara AI system.

"We have to get to a place where this thing doesn't need to be 100% automated or our company dies," she says. "That's a shitty place to be in. You want to provide an excellent service to a lot of people, even if it takes some time in some cases to give the right answer, intelligently."

In other words, the most clever bit about Clara may be that she's not an AI. Rather, she's an AI that's capable of shorthanding work for a massive human force. But for us, the clients and conversation starters, Clara can be dozens or hundreds or thousands of people—and that's fine—because Clara wraps it all into one vessel whom we can actually talk to. After a month working with Clara by my side, my greatest realization was that, quite simply, I don't care if she's AI or not.

And maybe that's the real genius of Clara. If nothing else, she's a designed construct that gives everyday people the semantic keys to the power of micro-labor.

She's a platform to make all of us feel like a boss.

[Photo: Her, Annapurna Pictures 2013]