Microsoft: Cortana Programming Reaches Human Parity
According to a report from Microsoft, company researchers working on the Cortana programming for the Xbox’s speech program have achieved human parity in programming her, a historic milestone in the case of artificial intelligence.
According to the researchers, the program’s word error rate fell from 6.3 to 5.9, around the same error rate as living, professional transcriptionists.
The drop in word error rate is the lowest ever recorded against the Switchboard speech recognition task. Within Microsoft itself, the decrease in WER signifies them reaching a milestone that they set for themselves less than a year ago, and means that the Cortana programming will be able to recognize a conversation as easily as a regular person would.
Speech programs like Cortana first came into prominence with Apple’s Siri program, and soon spread out to many other smartphones. While there’s still a lot of work to be done before they could pick up all of the nuances of human speech, Cortana, Siri, and the various other programs like them have reached a new benchmark with this new achievement.
Five years ago, for instance, Harry Shum (the executive vice president of Microsoft’s Artificial Intelligence and Research group) believed that something like this could never happen, but the Microsoft team has proven him wrong.
While it’s likely that the Cortana programming efforts will continue until they’re as perfect as possibe, Microsoft getting Cortana’s word error rate to drop to the same level as a normal human is still a landmark accomplishment.
Cortana has been a part of the Microsoft Windows phone for a good while now, but it only came to the Xbox One back during the summer, first appearing on the console’s Preview option before finally being placed on the Xbox One’s dashboard as an app to use (that’s likely better than the Kinect voice-command system.