Allo is a child that Google just sent to school to become your personal assistant. The first release looks like a messenger app because it is a common user interface (UI) that almost all smartphone users know, and it is conversational. Like a child learns to speak through conversation with adults, Allo will learn to be a personal assistant through quadrillions of messaging conversations if it succeeds as the next big platform.
Google Now and Apple’s Siri are rudimentary compared to the personal assistant that Google Allo could become. Google Now’s and Siri’s voice to text is pretty accurate, but the user is limited to a fixed set of commands, navigate, play music, search, etc.
+ Also on Network World: 25 surprising facts about Google +
Google and Facebook have already said they want to create a personal assistant better than Google Now and Siri—one that is like a human personal assistant. A personal assistant knows your history, details, preferences and habits. It also knows what you mean, not as a general interpretation of spoken and written phrases, but based on how a phrase is used in common language and the way a user says it. A field of machine learning called word sense disambiguation (WSD) teaches systems the meaning of words in the context of the conversation, such as this example of the two meanings of the word “bank”:
The boy leapt from the bank into the cold water.
The van pulled up outside the bank, and three masked men got out.
Three levels of machine learning underpin Allo. First, of course, is the meaning of language with all its nuances, idioms and exceptions using WSD to understand meaning within the context and situational queues. Second, with over a billion Google accounts, email data and search data, machine learning can predict much about a user’s intent. The last, and hardest, part is personalizing Allo in real time with the user’s interests by applying neural networks to deduce inference—in other words, to derive logical conclusions from the phrase and context.
Inference is a well-researched and understood academic field that produces reliable conclusions from test datasets. Implementing inference engines at the billion-user scale with live data is more difficult. Optimizing the inference engines to work quickly and responsively on a mobile device is even harder.
We knew Google was working on this
Allo the personal assistant is no surprise. Google’s open approach to innovation through participation in open forums and open-source projects, published papers and the release of application programming interfaces (APIs) publicly signaled its development goals.
Allo is built with three of Google’s technical domains of expertise: search, natural language processing and machine learning. Google’s knowledge graph includes over a billion people, places and things and more than 21 billion semantic relationships between them, which can be used to respond to spoken questions.
Much of the technology used on the backend of Allo is available for developers to use today to build intelligent digital assistant capabilities into apps. It’s possible for developers to add intelligent assistant features now and add more intelligent assistant features when Google releases more robust versions of the APIs listed below:
Google and Facebook have the best chances of winning the race to be the personal assistant platform because both companies have the two most important resources. Both companies have invested in building top AI and machine learning teams, and both have enormous stores of live data, the raw material needed to train a personal assistant to serve at the billion-user scale.