Could your next personal assistant be built using artificial intelligence?

x.ai believes that its personal assistant won't only be able to schedule meetings, but also interpret tone, sarcasm and lies

Meet Amy or Andrew (whichever you prefer) - your new personal assistant. But this isn't just any personal assistant it's one based on artificial intelligence (AI) and is far more sophisticated than your usual virtual assistant.

Founded in 2014, US-based x.ai's business is based on a simple idea; it wanted to make it easier for people to schedule meetings. But Dennis Mortensen, CEO of the company, explains that this goal presented an "extremely hard" challenge. So hard, in fact, that he suggests that it ought not even be possible.

But surely, given that it's now possible to have remote-controlled lawnmowers and driverless cars, it can't be that hard to have a virtual personal assistant that can plug into your Outlook or Gmail application and set up briefings if you tell it to?

Mortensen suggests that this probably would be easy, but x.ai has loftier ambitions than that. What x.ai aims to achieve is a human-like personal assistant, one that you would communicate with in the same way you would a real-life personal assistant.

"We're trying to recreate Her [referencing the 2013 romantic science fiction comedy-drama film]. Think about anything you can do if you hired a real person, we want to mimic that in a machine instance, that means there is no app, no UI, no settings.... You'll write an email and ask to set something up with Dennis when you're in New York on Friday and Amy [the AI assistant] needs to understand what you want and what's achievable.

"She has to be able to have human-to-human like dialogue with the other party to figure out if that Friday works, when on the day would work, and then she'll send out an invite. That's what we're trying to pull off and I think we're in good shape," he says.

Indeed, the firm hasn't gone unnoticed. In January, TechCrunch reported that x.ai raised nearly $10m in new funding at a $40m valuation. This isn't so surprising when you take into account the fact that Mortensen was CEO of Visual Revenue, and COO of Indextools, companies that were acquired by Outbrain and Yahoo respectively. And yet, still somewhat surprising when you consider than the firm's product is still in beta.

But as Mortensen explains, the likes of Apple's Siri, Microsoft's Cortana, Google Now and IBM's Watson are doing what he calls the simple tasks, while x.ai is striving to do more.

"You can ask them a lot of easy things, what's the time in Singapore, set my alarm for tomorrow morning at 6:20am- but they aren't really verticalised AIs. Where they do tasks - x.AI does jobs," he says.

By this, he means that Amy or Andrew will take care of the tedious matter of scheduling a meeting. Not just by sending out a calendar invite, but also by making sure the time fits around your schedule, that a reminder goes out before the meeting if you'd like it to, and nudge people to accept the calendar invite.

Mortensen says that according to x.ai's data, two humans usually take an average of two days to arrange a time that works for them.

But whilst this all sounds fine, it doesn't really justify the $40m valuation. That sum starts to make more sense, however, when you look a little deeper. x.ai's product requires the user to give Amy or Andrew access to their calendar, of course, but it also allows the user to sit down with their personal assistant and tell him or her that they're not a morning person, or that they like to keep Fridays free to close off the week.

Amy or Andrew will keep note of your preferences and book in meetings accordingly.

Mortensen says that a key area in which machines are superior to humans, and one that x.ai is looking to exploit, is the analysis of historical data.

"She will go in and look at your calendar, so if you hired a human you wouldn't go back to January 1st 2010 and look at who you met with, why you met with them, how long you met with them, and do the same for the last five years. But Amy will do that for you," he says.

But how does Amy understand such phrases as "I'm not a morning person"?

"The only way we can be allowed space in the market is if we solve [this challenge] - not something where you have to come out with specific syntax where you have to speak to Amy in a certain way or she won't get it because then we would be handicapped. We are trying to create something so that the user doesn't have to think about how they phrase things," he says.

This is why x.ai is training the product for US English to crack the American market before it has to alter its understanding for other regions, and also why the firm is taking its time before going live with the full version of the product.

Mortensen says that the company spent about a year and a half trying to define the "universe in which we exist".

"I wish a meeting was just a date, a time and a location, but sadly it's not - so we spent almost half a year trying to define in conceptual terms what that universe looks like and changed it many times over. The conceptual model needed to be mimicked in our database, too," he says.

This is why the company uses NoSQL database provider MongoDB, he explains.

"MongoDB allows us to be extremely agile, both in applying new models and changing existing models."

x.ai seems far more sophisticated than the usual "virtual assistant", but is it really artificial intelligence that is behind it?

"I think that's a very fair question because I think it's quite easy to slap on the AI moniker to any new venture from three guys and a garage," Mortensen says.

But he insists that the fact the firm has spent two years or so without going to market shows that it really is doing the required research to come up with machine intelligence that can handle human-to-human like dialogue.

[Turn to page 2]

Could your next personal assistant be built using artificial intelligence?

x.ai believes that its personal assistant won't only be able to schedule meetings, but also interpret tone, sarcasm and lies

Going for the enterprise

When Computing asks Mortensen what the target market is for the app, he responds with a question of his own: "What is Dropbox?"

For him, x.ai will have the same effect that Dropbox has had, or is having, on the enterprise. It will first and foremost be a consumer app, that will grow in use within the enterprise, and finally catch the CIO's attention.

"So we're probably a consumer app but we're aiming to make all of our money from the enterprise," he says.

"For us it's going to be the same [as Dropbox], after one colleague uses it, it becomes 10 and keeps spreading, the CIO will say ‘what, you're exposing our emails?', we should have a corporate agreement for this'."

How human can AI be?

Mortensen says the company has invested heavily in ensuring that its personal assistant is as human as possible.

"It's not about being a game of fooling people that this might be a human; we think we will be able to do better if she is humanised," he says.

He adds that data scientists at the start-up are currently working on how to predict gratitude, because many of its beta users are writing back to Amy saying "thank you for setting up this briefing" - even though they know Amy is not real.

Mortensen believes there are countless possibilities as to why people are responding, which is why he wants to get to the bottom of it, in order to ensure that x.ai can react accordingly.

Other elements that x.ai is taking into account are things like tone, sarcasm, humour, lies and compromises.

Capital letters or specific words such as "must" could indicate a change in tone, while Amy or Andrew need to be able to know if you email him or her at 3am about a reschedule on the same day, whether "tomorrow" is actually tomorrow.

If all of someone's meetings take place in one office or coffee shop, the personal assistant needs to understand why certain meetings aren't the default.

"[Amy needs to know] when I work the meeting or when do I surrender the meeting because it's more important than the default," Mortensen says.

And while x.ai is limited to email, Mortensen can see it working with the likes of Siri and Cortana in the future.

"We're not competing with Siri here, Siri is very likely going to be an enabler in the future. Someone could ask Siri to ask Amy to set up a meeting with Dennis, for example," he says.

x.ai could also conceivably be integrated with other apps in the social sphere such as Tinder, Facebook and LinkedIn - again allowing people to schedule meetings with contacts without having to do it themselves.

But despite the potential of x.ai, Mortensen and his team clearly have a challenge to incorporate all of these features into their product, and make sure it works. And if it does work, the question will be whether it is human enough to do the job.

Mortensen uses an example of people having to trust a driverless car. Likewise, users would have to trust x.ai to schedule meetings - and not get anything wrong. Otherwise, those same users may have increased their time scheduling meetings, because Amy or Andrew aren't up to the job.