Amy Ingram, an AI-powered 'assistant,' simplifies the task of planning meetings. Here's how 'she' works.
"Hey Amy," I typed into Gmail, "can you move my lunch meeting with Kristin to 1 instead of 12 next week?"
In minutes, Kristin and I got an updated Google calendar invite, along with a friendly message from Amy. But Amy, it happens, is no normal secretary--she's an artificial intelligence (AI) program.
The term AI often conjures sci-fi scenarios of a world run by robots or the close-to-reality concept of driverless cars navigating their way through city traffic. But AI is often embedded in the most simple tasks. Now, the company x.ai has created a free AI-helper called Amy Ingram--"AI," get it?--that founder Dennis R. Mortensen told TechRepublic aims to "democratize access to an assistant."
Here's how Amy works.
- Sign up for access to Amy. You'll get put on a wait-list, but be patient!
- Once you've got access, you can link up your Google calendar and set preferences, like "give 20 minute breaks between calls," preferences for where to meet, whether Skype/phone are best, etcetera.
- Then you simply put Amy@x.ai in the Cc line of your email when you're coordinating a meeting.
- In the body of the email, put something along the lines of "I'm Cc'ing Amy, my AI virtual assistant, to help set up our meeting."
- Amy will take over from there. She'll reach out to your contact and offer potential meeting times. She'll go back and forth for however long is necessary to set something up. But all of it is invisible to you--Amy will work directly with your contact until the meeting is finished, and will send a calendar invite when it's ready.
- You can ask Amy for an update, like "How's it going with planning the lunch meeting with Kristin?" She will respond and give you a status update.
- If you change your mind and want to cancel an appointment, just tell Amy and she'll fix it.
- If something is a low (or high) priority, Amy can schedule it later (or sooner). Tell Amy to push a meeting into the future (back channel or explicitly). "Get this on my calendar this week!" or "This one can wait 'til June!"
- If you have what Mortensen calls a "Hollywood moment"--you need to "clear my schedule immediately!"--Amy can do that too, working with everyone you have meetings with on those days and pushing the meetings back.
Amy is still in beta mode, but will be available to all later this year. (She also has a "brother," Andrew, in case you prefer a male assistant.) In the last year, x.ai has been continuing to work on the mechanics of the system, and has also introduced my.x.ai, which is an easier way to specify preferences.
How did Amy become a reality?
It's based on a personal experience. Mortensen "went from CEO to scheduling assistant overnight." He wondered: could he schedule meetings on someone else's behalf with little to no context?
He found that yes, he could.
Mortensen learned that he didn't need "yearlong, intimate details" about the people he was working for. He simply scheduled meetings. He also discovered that the scheduling began taking on a similar format. "There were boundaries," he said. "Not only could I do this without context, but this clustering suggested it was a tractable problem."
When he saw the joy people experienced after getting an assistant, he was convinced. He hired a full-time assistant to "expand on the test."
Amy is an example of "narrow AI"--a type of artificial intelligence that is programmed to solve one very concrete problem, like playing the game of chess, for example. Amy is great at scheduling, but can't make a dinner reservation.
Mortensen sees Amy as being integrated into different programs, like Siri, for example, so that you can ask Siri to get something on your calendar, and "she" could reach out to Amy for help.
The mechanics of building an AI autonomous agent haven't been easy. While coordinating a date, time, and place may seem to be a simple task, it's actually got a lot of complicated variables. "How do we define time, space?" asked Mortensen. "It's not as simple as saying 'Wednesday at 3:00 pm.' People say 'afternoon,' 'later,' 'once we've done the demo,' 'upon my return,'" he said.
So how does x.ai learn what time "means"? Through massive data labeling. Unlike the lower-skilled labeling jobs, Mortensen said "our data needs to be labeled by semi- or highly-skilled workers. We have a 16-page guideline to memorize just for time," he said, "and a similar one for people and for location. If the labels aren't good, the training that you do will suffer."
The business model for x.ai is similar to Slack or Dropbox or Evernote, said Mortensen. Eventually, they may ask for companies to purchase subscriptions. And while Amy is meant to help people with no access to an assistant, another potential effect, he believes, is its ability to transform the role of the assistant, freeing him or her up to do more interesting tasks.
The 3 big takeaways for TechRepublic readers
- An AI assistant has the ability to take over communication to plan meetings, freeing up a lot of time and eliminating unnecessary email correspondence.
- Amy is an example of narrow AI, which is AI that is very good at completing a specific task.
- The AI assistant is an example of learning from a large amount of labeled data--part of a growing field of work to assist in machine learning.
- 7 trends for artificial intelligence in 2016: 'Like 2015 on steroids' (TechRepublic)
- The original robot "butler": What Zuckerberg can learn from Carnegie Mellon's HERB (TechRepublic)
- Why robots still need us: David A. Mindell debunks theory of complete autonomy (TechRepublic)
- How AI and automation could hollow out the US job market (TechRepublic)
- Q&A: A powerful look at the future of AI, from its epicenter at Carnegie Mellon (TechRepublic)
- Smart machines are about to run the world: Here's how to prepare (TechRepublic)