With love from my robot: virtual assistants may secretly be emailing you

It started as a normal email exchange with a tech CEO. He was up for a coffee, and passed me to his assistant to find a date. But then it turned a bit strange.

Her emails were too good: all written in the same carefully casual, slightly humourless style. All formatted the same. All sent at socially convincing times. And all at believable intervals from my own messages. But they were off just a little.

Hi Nellie,

No worries! Unfortunately, Swift is unavailable tomorrow morning. Can you talk at one of the following times?

Tuesday (Nov 10) at 3pm EST

Tuesday (Nov 10) at 4:30pm EST

Let me know!

Best,

Clara

I stared at her notes for a few minutes before it hit me: she was a bot.

Leaving aside the issues around giving the admin bot a female name, as all these services seem to do, this feels like one of those moments the future promised us. So now it’s here, I thought to myself, staring at her emails. It has arrived. She is among us. And she’s excellent.

“You asked me how I conceived of ‘her’, not ‘it’. When you talk about ‘her,’ we’re 90% there already,” said Dennis Mortensen, whose company, X.ai, has pioneered email bot personal assistants. “You already conceive of her as a human being even though you know she’s a machine. Now we have something to work with.”

Virtual email assistants, which started as clunky tools for hacker hobbyists trying to streamline their lives with artificial intelligence, have grown enormously in both their abilities and adoption in the last few months. My email contacts (venture capitalists, startup founders, bloggers) are a skewed and terrible sample (if someone has friends they could introduce me to, that would be great), but the tech world shapes workplace trends across industries and across the world. So, from the industry that brought you open plan offices – Silicon Valley – email bot assistants are coming next.

“It’s an inflection point right now. There are just a lot of tools to look at data like Hadoop that just weren’t available before,” said Amy Stapleton, a former Nasa executive who runs Virtual Agent Chat, a blog about assistant technologies. “I don’t know where it’s going. The whole thing is a little spooky, though, isn’t it?”

There are varying degrees of digital assistants. Facebook’s M, launched in August 2015, is an algorithm-human hybrid, in which the humans are hands-on “trainers” helping the bot learn to be a full-scale personal assistant.

Other services like Crystal or Google’s email autocomplete service, launched in October, try to just be a layer of intelligence over human activities, tweaking email phrasing and wording with data but without creating a whole bot personality. Clara Labs, which created Ms Stark who’d emailed me earlier, has algorithms that ping humans who can step in with empathy responses if, for example, someone’s canceled a meeting because of sickness.

And X.ai is almost entirely bot, a persona they’ve called “Amy Ingram”.

“If we were going to humanize Amy, we had to do it completely,” Mortensen said. For Mortensen, this means Amy has to be a cohesive personality. His staff doesn’t talk about building features, they talk about teaching her things. So he says he hired a Harvard drama major to embody her.

“Everything Amy might say has gone through her,” Mortensen told me.

“In the Amy person, you would want certain human qualities. Let’s call it empathy. If you’re canceling a meeting, she adds a little bit of love,” he said. “Not random love. Not Citibank, ‘we appreciate your business.’ Love.”

He and his team of about 60 add a delay to Amy’s responses to keep up the act, as though she’s typing.

The strangest surprise has been how users respond. Amy gets chocolates. She gets flowers. Someone sent her Red Bull. Mortensen says that in 11% of of the meetings they schedule, someone sends Amy a note of gratitude or emotion, even though they know she’s a robot.

“People who we told this was a machine, they still write her these things, like they just want to give her a virtual pat on the back,” he said.

So far, Amy’s set up half a million meetings.

“Every meeting we do, the wiser Amy will be,” he said. “We want to get to the inflection point where she will able to go learn and work on her own, independently.”

Not everyone in the email automation space likes where it’s going. Drew D’Agostino’s Crystal scans social media profiles and past writing to suggest better ways to write to people, and can prompt a specific tone or slang someone might want in an email.

“If you’re talking to someone with a personality type that’s more direct, Crystal can pick up when you’re dancing around a topic,” D’Agostino said, or she’ll help you to use more effusive language like “excellent” instead of “nice”.

D’Agostino’s skeptical of the all-bot assistants.

“I know I should like them because I’m in this too,” he said. “But I still, when somebody sends me to a bot instead of planning their own meeting, I’m like, ‘Do I really not matter enough to you to do a little calendar thing?’”

The bot also plays the purpose as scapegoat, D’Agostino said: “It’s like, ‘Oh my bot says I’m too busy, sorry, not my fault.’”