Meet Amy, the virtual assistant that you don't have to assist
In April, NUI Central hosted Dennis Mortensen, the CEO of x.ai and creator of New York’s most famous AI, Amy. Opening with a lengthy discussion of how we’ve gone from software to the cloud to apps, Mortensen argued that Amy is the next evolution: invisibility.
“Gartner called it...invisible software...that’s the space that we’re in,” said Mortensen. “A space for where Amy in concept doesn’t exist as a plugin, or as an extension, or as a web service, or as a website, or as an app. Amy only exists in dialogue...with this one capability: that she can set up meetings.”
Amy, the name of x.ai’s first artificial intelligence, was released in 2014 as a personal assistant that could schedule meetings when copied on email correspondence. The company raised nearly $10M earlier this year from a consortium of venture capital firms, and is looking to roll out their offerings to a larger audience.
While at first glance it may seem that Amy’s functionality is limited, Mortensen argued that it is but one of many necessary steps toward a complete AI. It won’t happen overnight, said Mortensen. In fact, despite a straightforward UX, there’s already a lot going on under the hood with meaningful ramifications for NUI practitioners.
Siri can translate requests into immediate answers. She’s a shortcut but she cannot intelligently solve problems within unique contexts. The first and foremost challenge that Amy has had to overcome is becoming aware of all the variables that impact the universe of scheduling meetings.
Previous attempts at solving this problem only tackled a single, well-defined task, Mortensen said. And it wasn’t worth the human effort to use the technology for minimal benefits. But Amy is more ambitious, seeking to replace a personal assistant entirely when it comes to scheduling meetings. Essentially, Amy is not too ambitious to become a complete AI, but ambitious enough that when she works effectively, she is well worth the effort. That’s why it’s of utmost importance to the x.ai team that she comprehenend and respond to all the potential variables and still be able to schedule a meeting on her own.
To be successful, Amy must be able to compete with a human personal assistant. She must understand preferences, whether the user doesn’t want to have meetings on Sundays or wants their meetings spaced out throughout the week. She must have total recall for past events to establish context. She must come across as human so that the person on the other end of the email can type casually without having to consider unauthentic syntax.
“There’s a whole set of concepts she must understand,” said Mortensen. “What becomes equally hard is that there is so much ambiguity that arises when machines interact with humans.”
For example, we’ll often send an email late at night, after midnight, asking to move a meeting ‘tomorrow’ when we really mean ‘today.’ This is just one of numerous use cases Amy must account for.
Mortensen explained how Amy heavily relies on the user’s calendar as a set of important data points for decisions made in the past. She’ll suggest convenient locations based on established preferences and coordinate meetings around those locations. The goal is for her to actually become more competent than human assists who ‘rarely go back’ to meeting records to learn for the future.
With Amy still available only to beta users, it’s tough to give her a test run. But we’re on the waiting list and excited for what x.ai has in store over the coming months.