Apple’s virtual personal assistant, Siri, may never sleep, but it also lacks the smarts of a real personal assistant. A human assistant, in contrast, can work on much more complex tasks but comes at a much higher price.
A new alternative, called Premier, aims to surpass them both. Outwardly, it’s a capable, round-the-clock personal helper; behind the scenes, it consists of a crowd of workers corralled by smart software to respond intelligently to any kind of request.
Premier was developed by MobileWorks, a crowdsourcing company, which is currently testing the new assistant with some existing customers. Testers are given a private e-mail address and encouraged to treat it as they would a conventional personal assistant. “You talk to it in natural language and it talks back to you the same way,” says Anand Kulkarni, cofounder of MobileWorks.
Tasks Premier has handled so far include compiling reports, replying to e-mails, and even grading assignments for teachers. “We are really starting to see some unexpectedly intelligent behavior coming out of Premier,” Kulkarni says.
MobileWorks plans to launch a trial version of the service in 2013. Kulkarni says he usually kick-starts a person’s relationship with Premier by asking it (that is, the crowd of people behind it) to introduce itself with an e-mail listing ways it might be useful to that person. That list is compiled by doing online research about the new user.
When an e-mail is sent to a Premier address, it is immediately read over by a crowd worker, who breaks it down into a list of tasks, which are then fed into an automated routing system. The system directs those tasks to the workers best qualified to take them on, drawing on information about individuals’ specialties and past experience.
A simple task, such as pricing flight options for an upcoming trip, could be tackled by one person. More complex tasks, such as compiling a report on the state of the tobacco industry, as one person testing Premier asked, are broken up and handed to several different members of MobileWorks’ roster of crowdworkers. Producing such a report might mean recruiting several people to scour the Web for the most relevant information sources, several others to summarize the results succinctly, more workers to compile those into a single report, and a final group to proofread the document.
Premier’s most crucial ability, says Kulkarni, is that it can “remember” the history of each person’s interaction with the service, even though different groups of workers might handle each request. “You may well be interacting with somebody who is completely different, but you don’t know it,” he says.
That’s possible thanks to a system that requires each worker handling a Premier task to update a Wiki-style record of what has been done for the user. Those records then serve as a quick reference allowing any worker helping a Premier user to pick right up where the others left off. Periodically, some crowd workers are assisgned the task of tidying up those records so they make it easy for others to get up to speed. That feature of Premier was inspired by a Rochester University research project called Chorus, which created a crowdsourced virtual assistant that could carry on conversations via instant messaging (see “Artificial Intelligence, Powered by Many Humans”).
MobileWorks finds its workers in developing countries around the world, including India and Jamaica.
The company is considering targeting versions of Premier to users in particular professions. Some users of the prototype, for example, are teachers who have used it to help with simple grading.
Kulkarni says he’s not quite sure how Premier will be offered as a product, but he hopes it will broaden the appeal of crowdsourcing. Typically, considerable work is needed to figure out how to break up a task and assign it to many people.
“Here we’ve shifted a lot of that burden to the crowd itself,” says Kulkarni. “This is something that might bring crowds down to a consumer level, because it’s a simple e-mail address, the crowd manages itself, and we don’t have nearly as much of a barrier to entry.”
When designing an embedded system choosing which tools to use often comes down to building a custom solution or buying off-the-shelf tools.