The “parity with the human being” is reached. By inflating the chest of well-deserved pride, the leader of Cortana Xuedong Huang project announced the milestone reached by the voice recognition system available for platforms Windows 10, and Xbox One: the error rate dropped by one percentage point, from 6.9 to 5.9%.
That which on paper seems a negligible improvement is actually symptomatic of fidelity now acquired by Artificial Intelligence Microsoft: 6% in fact represented the threshold that has always divided the car man or more specifically by the typing professionals. Until yesterday, in fact, only the latter could give less than 6% of errors, the act of transcribing a speech or conversation.
The milestone from Redmond is the culmination of a research forty years, started outside of Microsoft: the first studies on speech recognition date back to the 70s, initiated by DARPA the US government agency for advanced research projects for the defense.
As is always the case when we talk about I.A. ( see the case of taxis in Singapore autonomous driving ), it is not all gold that glitters. In fact, despite having reached a level of word recognition identical to that of humans, c ‘is still a lot of work on another branch of the voice transcription: the understanding of the meaning of elaborate phrases, or that feeling that allows to a good typist to instantly correct the errors expressive of who is speaking, as well as to adapt the text in those practical situations where the background hum prevents capture every single syllable is pronounced.
In short, the “human equality” will also been achieved in the laboratory; but in real conditions the human being is likely to remain indispensable for a long time ‘.