AMY emailed me three times overnight, roughly once every 3 hours between 2 and 8 am. She鈥檚 setting me up on a call with Dennis Mortensen, head of X.ai, the startup company that built her. While her email address looks just like anyone else鈥檚, her unusual persistence betrays her true identity. Amy is an artificially intelligent personal assistant designed to automatically set up meetings for its owner.
Amy鈥檚 architects are part of a growing cadre of researchers and startups that want to hand over to machines many of the scheduling tasks that humans have traditionally had to look after.
Amy is made to be easy to use: anyone who signs up to the service simply copies in Amy on an email and it takes over the scheduling, recognising phrases like 鈥渟ometime next week鈥� or 鈥渘ext Wednesday鈥�. Amy looks at the user鈥檚 calendar and figures out when they are free for meetings. It then sends requests to the other person, or perhaps their AI assistant, asking questions to find out when they are available. When it finds a mutually suitable time, it sends out a calendar invite, hopefully saving two people from using up their time playing email ping-pong.
Advertisement
Ankit Agarwal, who runs a startup called Micello that makes indoor maps, says he now uses Amy instead of a human assistant for all scheduling tasks. 鈥淎my has set up hundreds of meetings for me,鈥� he says. 鈥淚t鈥檚 one of those magical services where you CC Amy and it gets done.鈥�
The system is designed to work with any communication system 鈥� SMS or voice, not just email, Mortensen says. And while it only works for scheduling right now, in future it could expand to control things like smart home appliances and robots.
Sriraam Natarajan of Indiana University in Bloomington says AI assistance could be incredibly useful in hospitals. An intelligent triage system, for example, could handle the influx of patients on its own, using digital health records and symptoms databases to offer more complete analyses than doctors and nurses often have time for.
鈥淭hink about entering a hospital,鈥� Natarajan says. 鈥淭he physician asks a few questions 鈥� did I have an injury in the past five years? My answer will probably be no because I don鈥檛 remember. How cool would it be if a system steps in and notes that I do have a certain preexisting condition?鈥�
鈥淕iven the growth of electronic health records, and that hospitals are linking records together, I think it鈥檚 possible in the near future,鈥� says Natarajan.
Other assistant services are looking to help out in other areas of life. For example, an app called Humin was launched outside the US on 9 December. It aims to help you remember people you meet and predict who you might like to contact by tying their details into the context of your relationship. This lets you search for 鈥減eople I met last week鈥�, say, just like you might ask a friend about someone you met at their party.
聯The app can search for 鈥榩eople I met last week鈥�, like asking a friend about someone at their party聰
Google Now, Siri and Microsoft鈥檚 Cortana assistants all contain a machine learning component that allows for similar predictive services, like scanning your email and notifying you of upcoming flights you鈥檝e booked.
For AI assistants to perform these tasks, they have had to take an important step in understanding language and how humans communicate, including common phrasings like 鈥渉ow about we set up a chat鈥�, as well as distinguishing time and place names.
As Amy鈥檚 overnight missives to my inbox attest, there is still much about human behaviour that machines don鈥檛 grasp. But, as Mortensen points out, such pestering ensures that a meeting isn鈥檛 missed. And if an AI assistant can competently take over these small, mundane jobs, he says, it will free up our brain power for more productive purposes.
And while we are very far from building computer intelligence on anything approaching a human level, Mortensen thinks that developing many smaller applications like Amy into a mesh of services that all talk to one another may one day provide a sort of super-helper, with each specialised bit of software tackling a narrow facet of daily life better than a human.
鈥淚f you assemble those companies together a decade from now, we might have something that looks like the AI we were promised,鈥� Mortensen says.
This article appeared in print under the headline 鈥淢eet your AI assistant鈥�
