According to 2017 Pew research, 75 per cent of Americans think that robots and computers will eventually perform most of the jobs currently done by people. With artificial intelligence improving every day, it's little suprise.
We look back at our interview with x.ai founder Dennis Mortensen to see what he had to say about creating artificial intelligence that can schedule meetings and whether we should all be fearful for our jobs...
Where did it all begin? How did the idea get started?
Back in 2013 we'd just sold our last company and I had a little bit of extra time on my hands and I ended up counting up all the meetings I did the year prior. Come the end of that, I found that I did 1,019 meetings and I set them all myself. No help, no nothing. Just me, sitting at home alone, crying with a bowl of cereal, trying to figure out who to meet when and where.
What's even sadder is that I had 672 reschedules that came along with that. If you look at that from just two feet away what you'll see is a massive amount of pain. And I think anybody who goes through that amount of pain would try to solve that in some way shape or form. Most people would solve that by getting up, hiring a human personal assistant, taking the hit on the $50,000 and that's it. Or they don't have the cost setting to justify that, try to put in place some software, seeing how it doesn't work, ends up on email pingpong. So that seems like a very good setting and starting point but I wish it could be more romantic. But it's really just all pain.
How did you go about creating something that works so well? I can't imagine that your first version of this was such a smooth transaction for users.
That's never the case. We had a few truths that we believed in where we've designed the product with these as the backdrop from day one. The first one is that we want this to be invisible software - we don't want this to be yet another app, a plugin, an extension, a web service, a Doodle, a something where you have to do additional work. We wanted something where we see software disappear and you hand over the job to that agent and that agent runs away, works on it and comes back once solved. That was the first battle.
We also wanted software that did the job in full. Don't assist you in doing it, don't do half a job, don't give you a few times and then you can write the email yourself, don't not follow up. Don't just help you be a little bit more efficient. We want to take this off your hands so you don't have to think about it at all. As soon as you've CC'd in Amy [Andrew’s female counterpart] and asked her to set up a meeting with Dennis, you shouldn't think about it. You should immediately archive that email and await an invite to be injected into your calendar and all is good.
That, of course, provides some stress on the engineering and data side of it because that suggests that if there can be no interface, no buttons to click, no dropdown, no calendar to look at, then it all had to happen in natural language and that you'd be able to communicate with it like any other human.
So I think the more it simplifies on the user end, the more you increase the complexity on the engineering side. But we are just so eager for this to turn into a solution that completely democratises the idea of a personal assistant. That it's not a luxury for the few, but it's something that everybody can get access to it. Just as I'm sure you’d never ask at an interview whether you would get access to email or not, you would just assume that of course you would get an email, how else would you be able to do your job? I think this needs to be something equally ubiqutous where, 'of course, I'm going to get an assistant to help me manage my calendar, why would they be paying me what they pay me to sit and do email pingpong, that doesn't seem right'. That's what we're trying to move towards.
Do you see artificial intelligence replacing PAs in the future? If the software is able to do this kind of scheduling job, would it be able to take over the whole role?
Do you have a PA yourself?
Definitely not, I am not important enough.
So, there is part of my answer. Almost all of us are 'not important enough' or not at the right place in life to have a personal assistant. That means that the vast majority, if not almost all, of the meetings being set up are being set up by people themselves - the Natalies, the Dennises, the Stefanies. We're not out to take any jobs. We're out to give something that was only for the few to the remainder of the organisation.
I'll give you a good example of my past company: We were 450 people and we had two assistants. That means everybody was on their own, but a few executives - and good for them! - but that's beside the point. Those PAs do a lot outside of just managing their calendar so we don't even think about that. This is not about taking their jobs, it's about giving this to you and I.
And what have your users said about it?
I think what takes the prize is that we see that the feedback is not 'it' or 'the software' or something else, it's 'him' or 'her' and people speak to it, even though they know it's a machine, as if it's a human being really. That I find really fascinating. We've seen it go all the way to the extreme where we receive chocolates, flowers, whisky to the office, we see Amy being asked out on dates (which I think is super sad!).
It certainly seems like with Amy you’ve achieved what anyone who is working on AI software is aiming for in creating something that seems so human.
If you decide to persuade 60 of your friends to go and work on an intelligent agent for two years, one of the very first decisions that you have to make is whether you humanise the agent or not. You can't do something in between, you either have to be Google Now, or Siri, Amy, Cortana. We decided to humanise the agent and we have invested heavily into that because we think that part of the success of the agent is how human she feels. Even if I tell you that it's a machine, I want you to walk away thinking what a nice person - even though you've been told. With that in mind, you see that she's got a full name, she's got a history, she signs off as you would, she's certainly formal but knows how to reply back to people who ask her out on a date or to meet her in the lobby or what have you.
I think there's two distinct choices that we made early on. One is that we're not trying to reinvent the typical dialogue that you would have with a human personal assistant, we're trying to emulate that experience. It's not that people dislike that experience. So we've gone through tens of thousands of meetings to work out what that dialogue looks like so we can emulate that.
The other is that we don't want Amy to be a bag of templates, we need her to be a whole being so that if you talk to her yesterday, when you speak to her today, you have a feeling of that being the same person and when you speak to her three weeks from today you remember how she communicates and how she wants to have things arranged. That is not an easy feat.
We ended up, not hiring another data scientist or PhD to help on that but actually a drama major, someone whose education and experience had been one of putting people on stage and creating characters, making sure that if you're on Broadway, you don't want to hear a pool of sentences, you want to have some sort of relationship with the characters that they've chosen to portray. And we've done the same here and it's super exciting to see that evolve. That's an entirely new job we created, instead of replacing personal assistants with this technology, we're actually creating entirely new kinds of jobs like the AI interaction designer.