Machine learning powered features will be the main point of differentiation of mobile phones and Google’s prowess in data mining will set Android apart.
A few years ago, hardware distinguished mobile devices. Setting Apple’s Retina display aside, most Android devices and iPhones are functionally equivalent and equally powerful. Then, content became the battleground. App stores raced to cultivate critical masses of applications, movies, books and music. Today, Android, iOS and Kindle Fire have reached the same plateau in these categories. In the next few years, mobile operating systems will use machine learning expertise and its application as points of differentiation and I believe Android will win disproportionate market share because of Google’s superiority in machine learning.
Voice / Siri
Voice recognition and voice commands, like Siri, represent the most salient example of machine learning on a mobile phone. Voice input machine learning problems ingest human speech and output text or trigger actions such as scheduling a meeting or setting an alarm. Successfully implementing this feature over many languages is no small data mining feat. Google developed this feature before Apple acquired Siri and because of its Google Voice transcription service plus the number of voice search queries, it likely has a much larger data set on which to train its models.
Google Now / Notifications
Google’s Now feature, which intelligently predicts information you might need during your day, impressed me most at Google IO. A Jelly Bean powered phone automatically informs you when you should leave to arrive on time for an appointment (including public transportation transit time) and learns that you may want to visit the gym in an hour based on your habits. Digital personal assistants have been a dream for a while. The first versions are finally here. While Apple’s Siri can’t be far behind, Google’s hundreds of millions of GMail users and Google Apps for Enterprise users form a much larger data set than Apple’s or Amazon’s.
Search is the original machine learning problem. Google runs the tables on search for the moment. Clearly, Apple’s move into Maps diminishes Google’s local query share and it invites speculation of Apple delving deeper into search. After all, search query data could inform other features within the OS. But Google’s search queries and data richness are unmatched. Because of the breadth of its search index, Google can build relevance and relationships across web data no other OS vendor can match, save Microsoft.
Not to be overlooked, content recommendations in the app and content stores will be important over the long-term. Finding the right content for the right person at the right time set Netflix apart. Despite Apple’s iTunes query and purchase advantage, Apple relies mostly on editorial curation for content recommendation, an approach that fails to personalize shopping. Amazon has an edge with all its existing purchase data and behavioral intelligence from commerce. Android lags in this area, but given enough handset distribution and even a modestly successful store, Google’s data asset will bloom.
The Foundation of ML
Machine learning’s life blood is data. The more data an algorithm has, the more accurate it becomes. Capturing, aggregating and leveraging this data is the mobile OS battlefield for the next few years. The scale of Google’s machine learning teams eclipse its competition. To supplement their ranks, Apple, Microsoft and Amazon will likely acquire many teams building consumer applications using machine learning. The most valuable ones will have large proprietary and datasets eminently useful for machine learning based features.
The race is on to build the best machine learning enabled mobile personal assistant at all costs. Ultimately, I believe Google’s access to significantly larger data sets, many of them proprietary, will enable superior Android features conquering larger market share.