Putting Robots To Work To Make New York City’s 311 A Better-Oiled Machine http://www.fastcoexist.com/3031202/putting-robots-to-work-to-make-new-york-citys-311-a-better-oiled-machine
Click here to preview the new Fast Company

Want to try out the new FastCompany.com?

If you’d like to return to the previous design, click the yellow button on the lower left corner.



Putting Robots To Work To Make New York City’s 311 A Better-Oiled Machine

In the future, 311 will let computers answer the easy questions to give humans more time for the difficult problems.

Since opening in 2003, New York City’s pioneering 311 center for non-emergency questions and complaints has become a massive operation, handling an average of around 60,000 questions a day via phone, text message, website, and mobile app. That’s added up to more than 180 million queries to date, processed by the hundreds of real, live humans that staff a call center in Manhattan 24 hours a day, seven days a week.

Yet it all seems rather antiquated at a time when a quick query to Siri or Google can almost instantly provide answers in other realms of life.

"It looks like some kind of NASA control center," says Markus Mobius, a principal researcher with Microsoft Research, who is now working with New York City to automate parts of the system. "It’s a very useful system. It’s high demand. But it is very, very labor intensive."

For Mobius, after visiting the 311 center, the goal was to develop easy-to-program software that can answer the easy queries, like whether schools are open or closed today, so humans can focus on the more difficult ones. It’s not as simple a problem as it sounds. While search engines like Bing and Google can scan many results and provide direct answers to some kinds of questions, such as "When was Barack Obama born?" they have a harder time doing that with more local searches that can only reference a smaller universe of documents.

It’s also a question of getting it right every time: "If you tell people alternate side parking is not in effect, and it is in effect, people get very mad about that," Mobius says.

Along with three collaborators at Microsoft Research, including a 16-year-old intern who helped program the system, Mobius developed and presented a prototype that he says is now being integrated into New York City’s 311 mobile app. The system includes a "router," which uses natural language processing and machine learning to understand the question being asked and extract the relevant information, like a date, an address, or a license plate number, as well as a series of simple "‘bots," which automatically supply the answers.

In one example, if someone asks the app if schools are closed today, the system could be trained to recognize the question and all variations of it and understand the word "today" should be translated to the date, and send the date to the "Are schools closed? bot that will automatically supply the answer.

The innovation in the system, says Mobius, is that it separates the activities that a computer is good at—the machine learning to process the meaning of the question—with the part that humans are good at, which is programming the bot to supply the correct answer. The researchers made the system so easy to program on the fly that anyone with minimal training could set it up to answer new questions on the fly, such as providing information about an impending hurricane or storm.

"With a relatively small number of bots we think we can automate between 20% and 40% of the questions submitted to 311," says Mobius.

Though there is no official launch date, Mobius says the city wants to make the system available through its app, which doesn’t currently allow people to ask interactive questions for fear the operators could be flooded with too many. He says it could be useful for the phone service too, but that would require an extra layer of uncertainty in using voice recognition.

"They are really concerned about peak demand. On days when there is peak volume, it’s always the same questions, like, ‘When is the snow plow coming?’ On the date where they are most desperate, that’s exactly the date when the kinds of questions are predictable," he says.

[Image: New York City via Shutterstock]

Add New Comment