Jonathan Foster from Microsoft's Cortana group on the Microsoft Campus in Redmond, WA. Matt Lutton / Boreal Collective for the Financial Times
The voice behind the robot: Jonathan Foster, leader of Microsoft's Cortana group © Matt Lutton/FT

Halloween may be nine months away, but Microsoft executives are already dreaming up responses to the traditional “trick-or treat” taunt.

A team in the Seattle office is bouncing suggestions around. “Give me a scary story!” is tried out, only to hang in the air. “Here, I got you some invisible treats!” suggests one. “The best I’ve got is Halloween jokes,” ventures another.

We are behind the scenes with the editorial team working on Cortana, Microsoft’s personal digital assistant, whose job is to breathe life and personality into a robot. They are riffing on potential responses to playful questions from an imaginary owner.

After throwing lines about, the team hits on the idea of granting “virtual treats”. They try out: “I’m giving you virtual treats right now — but they’re, you know, virtual. They might taste like air.”

Concerns are raised. Might that response sound borderline pornographic? The decision is deferred — there is plenty of time, after all.

The eight around the table are part of a broader global team of 28 working on Cortana’s script. All have backgrounds in the arts, and include technical writers, a playwright and a children’s novelist.

Microsoft's Cortana group, led by Jonathan Foster (second from right), on the Microsoft Campus in Redmond, WA. Behind the team is a piece of art featuring Microsoft's founder Bill Gates. Matt Lutton / Boreal Collective for the Financial Times
Microsoft's Cortana group discuss ideas for the assistant's voice and personality © Matt Lutton/FT

These writers, Microsoft hopes, will enhance Cortana’s personality and give it an edge on its peers as the market for personal digital assistants heats up. Cortana, which is available on mobile devices and personal computers in 13 countries, is facing a challenge. The competitive market includes established services such as Apple’s Siri and Google’s Assistant as well as Alexa, Amazon’s rising star.

As experts and economists worry that robots will replace humans in the workplace, the gathering in Seattle is intriguing. At Microsoft, robots are creating jobs — for the time being at least — as writers are hired to assist the development of artificial intelligence. Google has reportedly hired writers from The Onion, the satirical website, and Pixar, the film company. Jonathan Foster, principal content publishing manager for Cortana, sees future opportunities in this line of work for linguists and

Science fiction has set a high standard for virtual assistants. In Her, the 2013 film, the protagonist, a depressed soon-to-be divorcee falls in love with the voice on his phone, played by Scarlett Johansson. Assistants usually have female voices, variously explained by researchers as being more pleasing than male voices. Others suggest they might in future replace the predominantly female personal assistant role.

Cortana is no exception — in the US, her voice is peppy and professional, her gender amplified by the fact that her name is a reference to a buxom character clothed in a transparent sheath in the video game Halo.

Cortana plucks most of her responses to common questions from the internet — for example, to find out where the nearest sushi restaurant is or the chronological list of all the US presidents. Other responses are, to use Microsoft’s jargon “non-utilitarian”. In other words, fun.

People have emotional responses when they are engaging with digital assistants, according to Mr Foster. “If you accept it, you either ignore it or embrace it.”

Microsoft is embracing it. By giving a digital assistant a personality, the interaction between human and assistant is smoothed, he says. But does a utilitarian product need to have opinions?

Why, for example, does Cortana have a favourite movie? “Because people are asking that,” says Mr Foster. For a while, her favourite film was ET (she skews to science fiction) but today it swings between Star Wars and Star Trek films. Her favourite TV show is Star Trek: the Next Generation.

Users often ask their machines cheeky questions. “We want to delight them. We know they are in play land [then]. People like it when they are delighted.”

While most of Cortana’s human responses are generated in the US, the words are adapted to local markets, so they feel, say, German or Mexican. Mr Foster, who has written for television, as well as a feature film Not That Funny, starring Tony Hale, says “cheekiness” was found to be more important in the UK than the US, where humour is “watered down”.

Humour has to be appropriate. If Cortana comes up with the wrong answer to a question then the user will be frustrated. “We don’t want to be flippant or funny,” he says.

When Microsoft considers applicants for Cortana scriptwriting jobs, it scrutinises samples of candidates’ writing for an ear for conversational dialogue. The job is to figure out how to make Cortana interact as if she is having an interesting human conversation. Yes or no answers are avoided as much as possible.

Steve Barker, another Cortana scriptwriter and an essayist and author, says the job is creative.

“This work has made me a better writer,” he says. He is forced to cut
everything down in his personal writing, making him think ruthlessly about how each word must count.

Ron Owens, who is also a playwright, says the job has helped him sharpen dialogue. Both like the camaraderie of discussing words as a team, sparking ideas off one another.

They all must ensure that users are in on the joke, says Mr Foster. Cortana cannot be hostile to her owner. The team must always remember that, while Cortana has personality, the aim is not for her to be human.

Conversation can lead the script team at Microsoft down dark avenues. For example, they took a long time to discuss how to deal with questions such as whether she likes Nazis. In the end, they decided her answer should be a straightforward “no”.

Last year, Tay, Microsoft’s chatbot, was let loose on Twitter to engage in conversation and reply to questions. Tay had to be pulled after users encouraged her to make racist slurs and endorse Hitler.

The company said at the time: “The AI chatbot Tay is a machine-learning project, designed for human engagement. As it learns, some of its responses are inappropriate and indicative of the types of interactions some people are having with it.”

And rather than encourage sexually provocative, or downright abusive conversation, Cortana will shut the conversation down. “I’ve learnt a lot about the naughty realm,” says Mr Foster.

There is also the “uncanny valley” problem, in which digital voices that are too human-sounding are found to be disturbing.

Microsoft has considered this. If you ask Cortana what her favourite food is, she will not reply “waffles”.

Rather, in the US, she responds: “I dream of one day getting to taste waffles.”

Cortana must always be otherworldly.

Get alerts on Microsoft Corp when a new story is published

Copyright The Financial Times Limited 2019. All rights reserved.
Reuse this content (opens in new window)

Follow the topics in this article