Dennis Mortensen, expert in analytics, optimisation and Big Data, is soon to be a household name. Just give it some time.
The CEO of x.ai and a visionary in the field of Artificial Intelligence (AI), Dennis has, in the past, come under fire for potentially putting a lot of humans out of work. “We’re not out to get the job of the personal assistant,” he says. “This new setting is really a… kind of paradigm shift in how software is being delivered.”
Meet Amy and Andrew Ingram
A scroll through x.ai’s homepage will reveal some telling feedback about Ms Ingram. One from an individual named Patrick stands out: “I’m not exaggerating when I say she has saved me close to 10 hours a week.” Then, from the Financial Times: “When I first met Amy, I thought she was real…”
Amy Ingram is, in fact, a bot, the brainchild of Mortensen, who in 2012 had manually – and somewhat painfully – scheduled over 1,000 meetings in 12 months. The website cheerfully states: “Magically schedule meetings. That’s us. That’s all we think about.”
Amy and Andrew are personal assistant bots with one specific role: to schedule meetings using plain English.
There’s something about Amy
Mortensen remembers a time in which software was used in a very particular way.
“There’s a job which I want done and for me to be able to do that job, I will install some application. I’ll use that to do that job a little bit faster and a little bit more accurate… I will take credit for that job. I did it, I might have been empowered by software, but I bloody did the job,” says Mortensen.
“I think this new paradigm is one for where I won’t do the job… That is a rather dramatic shift in how we think about software.”
Mortensen and his cofounders at x.ai believe that Amy is reflective of a change in the way we as consumers expect products and services to be delivered to us – and it’s been happening for a while. Put simply, AI is machine learning, and tech giants like Apple and Amazon are already using their own bots for predictive services: think product and ad recommendations.
The difference, he says, is in their bots’ accuracy.
“Let’s play out this idea that four of us right now want to go build a self-driving car. If that’s our goal, we probably have some demand of accuracy where we need close to 100%… Certainly where there’s a pedestrian in front of the car and if there’s some models where you simply can’t achieve that, then that’s the end, that’s the ceiling of the performance you can extract from that service.”
There are plenty of other services where accuracy does not need to be close to perfect. Take Facebook’s algorithm that matches names to faces in uploaded photos.
“If that’s not running at near 100%, that’s probably not the end of the world,” says Mortensen.
A truly autonomous intelligent agent
“Now the truly autonomous agent, that is where the real value creation, at least in my opinion, will arrive again over the next half decade,” says Mortensen.
By autonomous agent, he simply means being able to speak with the bot in conversational language and have it understand what it is you need. “I think that is extremely important and that is obviously what makes this super hard.”
But there’s more to Amy and Andrew than just setting up the meeting in your calendar. They’re bots, but they were designed as characters.”
Think of a standard email you’d send to set up a meeting with a colleague. It would probably sound something along the lines of: ‘Good to talk to you yesterday. Do you have time to meet up later today, tomorrow, or perhaps early next week? I am free most days after 1 pm.’
“That is really a standard email, but look at it. It’s really a riddle of sorts. What the hell do you want me to do here? What we’re trying to do here is create this new setting where all you do is CC’ed your agent – in our case, it’s called either Amy or Andrew – and reply back, ‘I’d be happy to meet up next week. Please work with Amy CC’ed to get this on the calendar. Looking forward to it.’”
There’s a lot of implicit logic that comes with simple commands and, at the end of the day, Mortensen believes this is the direction we need AI to go for true innovation within the sphere of machine learning.
“If you really think about what we do here, even for something as simple as setting up meetings, is really a text-extraction challenge and a text-generation challenge.”
But there’s more to Amy and Andrew than just setting up the meeting in your calendar. They’re bots, but they were designed as characters. And they’ve done that by getting rid of canned response templates and teaching them to respond with natural-sounding dialogue.
“You can’t constrict avenues they walk down. You must assume that people aren’t following some vector from beginning to end, and it’s always going to be where we’re going to have that dialogue. People just inject all sorts of things in the middle of the conversation that you need to be able to participate in.
“It’s kind of hard,” he says, “but super f*cking sexy.”
Originally posted on BeMyApp Media. This article was based on Rand Hindi’s talk for With The Best’s web conference on AI.