Computer scientists are looking to improve on the performance of artificially intelligent personal assistants by devising a way to use the power of a human crowd to chat you instead. The system, known as Chorus, was designed by researchers at the University of Rochester to allow a number of users to act as a single agent that converses with a single end user in real time.
Chorus was made to try and deal with a couple of problems - the limited knowledge base of a single human user, and the often stilted conversational ability of AI that can leave you feeling like you would be better off talking to your dog.
UPGRADE TO NEW ATLAS PLUS
More than 1,200 New Atlas Plus subscribers directly support our journalism, and get access to our premium ad-free site and email newsletter. Join them for just US$19 a year.UPGRADE
The use of a multitude of human users means that everyone can suggest answers, providing a large pool of possible responses, with the crowd voting to reach a consensus about the best way to proceed.
Given that humans are typing up the responses though, as well keeping a running record of the conversation to date for continuity purposes, the answers should seem a little more natural.
In a study using the software, the researchers found that the crowd was able to successfully answer user queries 84.6 percent of the time.
Chorus is also designed to be drop-in drop-out friendly. The individual participants don’t matter as much as the crowd itself, and are free to stay for as long or as little as they like, or even participate across multiple conversations.
“Chorus is capable of maintaining a consistent, on-topic conversation with end users across multiple sessions, despite constituent individuals perpetually joining and leaving the crowd," writes the study team. “This is enabled by using a curated shared dialogue history.”
The question, though, is with all the curated histories, suggested responses, and user voting going on, will the conversation be as natural and back-and-forth as regular human chat? Will it be as fast as a response from an artificial intelligence? We tried to get in touch with the research team to find out, but they were not immediately available for comment.
Interestingly, Technology Review have also reported that the system could work in a similar way to Amazon’s Mechanical Turk, with crowd users being paid in micro-transactions of a few cents to participate in the crowd.
There’s no word on whether Chorus will be commercialized at this point, or exactly how such payments would work. Nonetheless the idea presents an interesting, human-driven alternative to the traditional concierge, and a little more tailored and personable than a regular old search engine responses.View gallery - 2 images