- New digital personal assistant part of broader AI strategy
- Assistant to appear in Allo messaging app, then Home device
Google is turning to artificial intelligence to make sure people keep using its search engine, even if they’re not spending as much time on the Web and personal computers.
The Alphabet Inc. division unveiled a new mobile messaging application Wednesday called Allo containing a digital personal assistant, based on AI technology that powers other Google services like Inbox.
At its I/O developer conference near its Silicon Valley headquarters, the company also showed off a voice-based search device called Google Home that uses the same assistant technology to answer questions when people are in their houses, a potentially potent rival to Amazon.com Inc.’s popular Echo gadget.
Google Chief Executive Officer Sundar Pichai said the goal is to develop an “on-going two-way dialog with Google” and build billions of people their own “individual Google.” The CEO sees the Google digital assistant as an “ambient experience that extends across devices.”
Google became one of the world’s most valuable companies by making a search engine that sucks in billions of queries people type into web browsers on PCs and phones. Google sells ads based those indications of intent and desire. But that search advertising money machine is at risk as computing evolves and gives people new ways to find what they want -- and new avenues for competing companies to satisfy those wants.
The new products unveiled Wednesday -- and future ones using the same Google AI technology -- give the company a chance to keep its search engine relevant in an era of new, connected devices.
With the future of search -- and intent-based advertising -- up for grabs again, AI has become a big strategic area of investment for many technology companies. The bet is that whoever makes the most engaging and useful digital personal assistant, also known as the conversational interface, will control the layer between a person and their digital life, and collect the most revenue and profit from being that privileged broker.
Google has been working on artificial intelligence -- technology that lets computers teach themselves about the world -- for more than 15 years. The company hopes this expertise can help it build conversational computing products that beat the competition.
"It’s absolutely strategic," said Scott Huffman, a vice president of engineering for search. "If you think about this simple idea of having a conversation, that is the interface that all the people around you have."
Google hopes that if it makes it easier for people to access its services, they’ll use them more. That’s a pattern it has seen with other technology initiatives. "Every time we’ve had an improvement in voice recognition we see a corresponding jump in usage," Huffman said.
Conversational computing is a crowded field. Amazon’s Echo, which has been a bestseller for the company, lets people use their voice rather than type to search for things and order them from Amazon’s online store and play music through Amazon-owned services, cutting out Google. Microsoft Corp.’s Cortana AI assistant embedded in Windows 10 lets people ask questions that the company answers via its Bing search engine. Facebook Inc.’s digital assistant, M, uses some AI to let people get answers to questions and perform actions such as ordering flowers, and its recently unveiled chatbot platform gives companies a way to chat directly with consumers, no Googling required.
Over time, Google said it will develop other products and services using the same digital assistant, which will stay with people across devices, and remember their habits, Huffman said. That will let Google’s search and other services follow people from smartphones and smart watches into their car and homes. Google plans to add more AI capabilities to its assistant, some of which won’t necessarily appear in its main search engine, Huffman said.
The company’s Allo messaging app will come out this summer. It will suggest responses to messages by reading and understanding people’s text conversations. A contact named "@google" can be summoned by users to provide AI-powered services, like finding restaurants and booking tables, or searching for movies.
The AI in Allo is based on technology already deployed in Google’s Inbox program, which reads through e-mails and suggests appropriate replies. It also understands images sent in text messages, using its AI to look at what is in the picture and suggest its own comments.
The more people use Allo the better the AI will get. The system works by converting words and images into sequences of numbers explicable to Google’s machine intelligence, letting it develop intuition so it can guess the word "dog" is semantically similar to puppy, or that the appropriate response to a picture of someone skydiving is "brave."
In a recent demo, Erik Kay, a Google director of engineering for communications products, took a photo of colleague Amit Fulay, who posed with an exaggerated grin. Kay texted the photo to Amit using Allo, and on Amit’s phone the AI studied the photo, noticed the grin and came up with the response "sunny smile :-)". These kinds of automated responses are meant to give people a more satisfying conversation, but has the additional benefit of generating more data for Google to use to further develop its AI.
After Allo, the company plans to release the Google Home device that people can speak to. The gadget will play music, communicate with other Google devices, and answer questions using Google’s AI assistant and its search engine, along with managing other Google products like Calendar and Gmail. People will summon it by using the same call out -- "OK, Google" -- that is used in other existing Google mobile apps. "Hey, Google" will also work.
"We don’t necessarily have to be first every time, that’s not actually our goal, but we want to be the most scalable solution," said Rishi Chandra, a vice president of product management for Google, noting the product is "not an assistant that does three things but it can really do anything."
Over time, Google said its AI will gain new capabilities. The company plans to give it a more flexible memory, so when people have a conversation where a friend references their home address, it will learn to associate that location with them. A technology called “Expander” will help the technology work in multiple languages, with insights gleaned from AI deployed in one language feeding the intelligence of the software in another tongue.
But it will take time for the AI to be able to anticipate and respond to every request, said Huffman. "It’s nice job security for us because it’s going to take it a while to let it do anything people want to do."