Send an email from Microsoft Outlook promising to send a document to a colleague by the close of business, and Microsoft will put a message in your system that pops up on whatever screen you’re using to remind you of the promise you made.
The feature, which will only work if you give Microsoft specific permission to analyse your emails (and indeed which only works under very limited circumstances right now, though Microsoft promises to make it more widely available), uses machine learning to help Microsoft’s artificial intelligence assistant, Cortana, figure out when someone has made a commitment in natural language in an email, however vague that language was.
You could promise to complete a task by the end of the month, and towards the end of the month, Cortana will remind you what you promised.
Piece by piece, artificial intelligence is becoming a standard part of the ‘‘software stack’’. Where once an operating system such as Windows provided the ability to save files to a disk and maybe print them over a network, now AI is being embedded into the OS, to be called upon when needed, just like any other service in the stack.
Steve Clayton, the general manager in charge of AI at Microsoft, says after five decades of research into AI, it’s only getting infused into everyday computing now that three stars have aligned: we have ‘‘almost infinite’’ computing power in the cloud; there are vast amounts of data that can be used by machine learning; and there have been major breakthroughs in the algorithms through which all that computing power learns from all that data. ‘‘We’ve started to put that sort of [AI] capability into day-today products,’’ he says.
On top of adding AI to Outlook, Microsoft has built AI services into its business-oriented social media site, LinkedIn, that help job applicants turn their profile into a job resume. Skype, Microsoft’s video- and voice-chat platform, also uses AI to do real-time speech translation in a dozen languages.
‘‘We’ve made some real advances ... particularly around speech and vision where we’re now at the point where vision services can pretty much recognise the world with the same level of accuracy as a human can. And the speech services can do even better than that, can take a transcription of a speech and actually do a better job than humans can in transcribing it,’’ Clayton says.
(Putting that claim to the test, we got an AI-based transcription service, Trint, to transcribe our interview with Clayton, and compared that with the results from a human-based transcription service. Trint was faster and cheaper, but the human transcription was vastly better. Perhaps not all AIs are created equal.)
The next step, Clayton says, is to get customers to start building AI into their own applications. Microsoft has just announced a version of its software development platform, Visual Studio, in which AI services such as computer vision are available for developers to call up, just like they would other services like networking and disk I/O.
While AI services aren’t yet so cheap to incorporate that developers would think ‘‘why not?’’ implement them, ‘‘we’re now at the stage where many industries are looking at it and saying ‘well how can I apply this?’ It’s out of the ivory towers of the AI researchers and it now is becoming a commodity service that anybody can use,’’ he says.
网友评论