When I signed up, I synced my calendar with Amy and provided her with three alternate emails I could use. It then collects my meeting options via an open request – “You can also email me if there’s anything else I should know to get started with; For example, you attend most of the meetings in your office or hate meetings before 10am. However, I have a feeling that I didn’t cover all the important details when answering. Then Amy just waited in the ether network for the next time I schedule a meeting. When an issue arises, I just copy Amy to the chain and the AI will take it from there – though Amy forwards the mail to me to make sure I’m comfortable. That is a thoughtful lease.
When I spoke with Amy to arrange a meeting with Mortensen, I found that I was really worried about what I wrote. Should I do a person sentence? “Hello Amy, Monday at 10 a.m. is best for me.” Or since Amy is really a machine, should I go straight into the data? “Yes: 10am Monday.” But that feels rude! As conversations with AI become more and more common, I wonder if we will start homogenizing our speech out of fear that computers will misunderstand.
“That’s one of the things I’m scared of,” says Mortensen. “We’re trying very hard not to introduce the syntax.” (Amy sometimes responds with smileys, which seem to trigger Uncanny Valley for me.)
Behind the scenes, the machine is cutting emails into simple data – timestamps, names, etc. – and sifting through the textual content of the email to find out, among other things, who was involved in the creation. calendar and social nuances are possible. X.AI even breaks down email signatures into seven possible parts. “There’s a lot of sweet stuff,” Mortensen said of the signature.
Amy assigns a confidence level to each email it sends – meaning a percentage of the chance Amy understands the questions and content correctly and is responding with a reasonable follow-up. If the AI drops below a certain threshold, an alert will be sent to the X.AI team for help. But not all syntax problems. In a hypothetical attempt to schedule a meeting with my brother, he tricked Amy by asking about safe places to lock his bike. Amy gave me his question.
“It doesn’t seem like it’s a message that I can provide an answer to so I recommend contacting Brett directly,” Amy said. So now I’m back in the meeting scheduling cycle. X.AI also wants to automate these customer-specific small details. But what if the bike parking is always full before 9:30 am or the meeting takes place in a secluded part of town, meaning locking the doors at night isn’t a good idea? In order to fulfill its full potential, X.AI intends to teach Amy the complexities beyond yes and no.
But for reasons of debate, suppose Amy learns to properly schedule meetings on “next Monday”. And it can figure out when an executive is going to take a taxi through town, or know how to scan Google Street View for a bike lock location. Assistant takes on many other duties for their boss. Maybe Amy starts to book flights, hotels and rental cars, hand out the powerpoint slides of this week’s board presentation, and order everyone’s favorite sandwiches from the catering provider. , all without asking.
It’s easy to carry. Maybe Amy got a promotion. Amy is now a master of all options and preferences from either side. Perhaps, I suggest to Mortensen, it’s good enough to negotiate business deals. The contract is just an optimal set of data points, right?
Dennis pressed on the brakes. “I try not to think about it,” he said of the AI promises. “We were disappointed a lot of times.” Currently X.AI is focusing on scheduling meetings. That’s it. “Once we get good at that, then I’ll move on to the next idea.”