22
Jan

Machine learning and Windows 10 and Windows Phone 8.1

2-Color-Design-Hi-Res-100px-widthMicrosoft’s Windows 10: The Next Chapter (http://news NULL.microsoft NULL.com/windows10story/) event, which was held on Wednesday, January 21, 2015, included a number of new product and features announcements. A lot of editorial copy has been produced on the Holographic computer, the “Hololens”. But other announcements, about machine learning capabilities, while, perhaps not as dramatic, also deserve some comments.

Windows 10 Leverages Machine Learning

During the 2 plus hour webcast, Joe Belfiore demonstrates the new role Microsoft’s Personal Assistant, aka “Cortana”, will play in Windows 10. He makes the point of mentioning Cortana’s capacity to learn, over time.

This machine learning capability is also demonstrated much later in the presentation, within the introduction of the Hololens.

Belfiore’s claims are overstated, at least when they are judged against my use of Cortana on Windows Phones. We own two of these, both are Windows 8.1 Lumia 925 smartphones. Cortana has operated as a feature on these phones for at least the last 90 days, and perhaps even longer. The biggest missing piece for us has been the lack of any improvement in Cortana’s understanding of either of our two users. One of our users has a pronounced accent, which has proven to be the basis of erroneous responses from Cortana, which she has gotten to simple questions.

Our other user, me, recently asked Cortana what it knows about him. Cortana’s reply included mention of a “Notebook”. The “Notebook” is presented in the “Windows 10: Next Chapter” event as a valuable new feature. In the webcast, one of the presenters (probably either Terry Myerson or Joe Belfiore) makes mention of the “Notebook” as an important control people will be able to use to determine just what personal information is available for processing and to limit the cognizance of the system of personal information.

Cortana’s reply to a simple question about what this personal assistant application “knows” about a specific person: “Well, I have my Notebook, so I know what you know you’ve let me know. Y’know?” seems to be more of a disclaimer than anything else. We took a look at the “Notebook”. We framed specific questions about information included in it, but could not get an answer from the system on any questions about any of the entries in the “Notebook”. Recommendation: turn down the hype on machine learning as this component of the system does not seem to have developed much at all since we last took a look at it.

In fairness, we need to also note we have an Android 4.4 KitKat smartphone manufactured by LG and an Apple iPad Air 2. Both of these devices also include personal assistants, “Siri” and “OK Google”, which are capable of understanding verbal commands and formulating audible responses. Neither of these devices are actually any more useful than Microsoft’s Cortana when the requirement amounts to an extended audio discussion with one’s computer device.

Machine learning in all of these applications has a long way to go before it is tangibly useful for personal computing.

Ira Michael Blonder

© IMB Enterprises, Inc. & Ira Michael Blonder, 2015 All Rights Reserved